<script data-pm-proxy="intercept"></script><?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:itunes="http://www.itunes.com/dtds/podcast-1.0.dtd" xmlns:googleplay="http://www.google.com/schemas/play-podcasts/1.0"><channel><title><![CDATA[LakeD-AI Unbundled]]></title><description><![CDATA[Exploring patterns at the crossroads of AI technology, investments, and human connection—one insight at a time.]]></description><link>https://lakedai.substack.com</link><image><url>https://substackcdn.com/image/fetch/$s_!sJyA!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa0f8ece2-b176-49d4-bec5-7431afde3248_500x500.png</url><title>LakeD-AI Unbundled</title><link>https://lakedai.substack.com</link></image><generator>Substack</generator><lastBuildDate>Sat, 05 Sep 2026 03:40:17 GMT</lastBuildDate><atom:link href="/__u/lakedai.substack.com/feed" rel="self" type="application/rss+xml"/><copyright><![CDATA[Lake Dai]]></copyright><language><![CDATA[en]]></language><webMaster><![CDATA[lakedai@substack.com]]></webMaster><itunes:owner><itunes:email><![CDATA[lakedai@substack.com]]></itunes:email><itunes:name><![CDATA[Lake Dai]]></itunes:name></itunes:owner><itunes:author><![CDATA[Lake Dai]]></itunes:author><googleplay:owner><![CDATA[lakedai@substack.com]]></googleplay:owner><googleplay:email><![CDATA[lakedai@substack.com]]></googleplay:email><googleplay:author><![CDATA[Lake Dai]]></googleplay:author><itunes:block><![CDATA[Yes]]></itunes:block><item><title><![CDATA[AI Infrastructure Is Moving Up the Stack]]></title><description><![CDATA[What Q2 2026 revealed about the software control plane for agents]]></description><link>https://lakedai.substack.com/p/ai-infrastructure-is-moving-up-the</link><guid isPermaLink="false">https://lakedai.substack.com/p/ai-infrastructure-is-moving-up-the</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Wed, 22 Jul 2026 04:23:07 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!a2Ie!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>For the past two years, most AI infrastructure discussions have focused on chips, data centers, and power. Those constraints still matter, but Q2 2026 made another shift clear: the next bottleneck is software.</p><p>Better models do not automatically create reliable products. Companies still need systems that decide which model to use, give AI agents the right data and permissions, recover from failures, control costs, and measure whether work was completed correctly.</p><p>This software layer is becoming the control plane for enterprise AI. It includes agent runtimes, distributed computing, inference optimization, memory, observability, evaluation, security, and governance. As models become easier to switch, these surrounding systems can become more durable and valuable.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!a2Ie!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!a2Ie!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png 424w, /__u/substackcdn.com/image/fetch/$s_!a2Ie!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png 848w, /__u/substackcdn.com/image/fetch/$s_!a2Ie!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png 1272w, /__u/substackcdn.com/image/fetch/$s_!a2Ie!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!a2Ie!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png" width="1456" height="819" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/d91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:819,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:1555616,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/208008209?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!a2Ie!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png 424w, /__u/substackcdn.com/image/fetch/$s_!a2Ie!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png 848w, /__u/substackcdn.com/image/fetch/$s_!a2Ie!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png 1272w, /__u/substackcdn.com/image/fetch/$s_!a2Ie!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd91ef94a-b31d-4bc5-95ee-1d2000a534d0_1672x941.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><h2>5 signals from Q2</h2><h3>1. Building an agent is getting easier; operating one is still hard</h3><p>An agent that answers a question is relatively simple. An agent that works for hours, uses company systems, and makes decisions is a distributed application. It needs memory, secure tool access, retries, approvals, and a record of every action.</p><p>This explains why agent platforms are expanding beyond frameworks into managed runtimes. The most valuable products will not simply help developers assemble an agent. They will keep fleets of agents reliable, secure, and efficient in production.</p><p>The opportunity is especially strong for neutral platforms that work across models and clouds. Large vendors will bundle basic agent capabilities, but many enterprises will not want one provider to control every model, data source, and workflow.</p><h3>2. MCP is becoming a common connection standard&#8212;and creating a new security layer</h3><p>MCP gives agents a standard way to connect with tools and data. Its progress toward enterprise authorization in Q2 was important because connectivity without control is dangerous.</p><p>When an employee asks an agent to update a customer record, approve an invoice, or review confidential documents, the agent should inherit that employee&#8217;s permissions. Companies also need to know which tools are trusted, what information was accessed, and why an action was taken.</p><p>As MCP adoption grows, demand should increase for agent gateways, identity management, tool verification, policy enforcement, and audit infrastructure. This resembles the rise of API management, but the risk is greater because agents decide when and how to act.</p><h3>3. AI observability must measure behavior, not just uptime</h3><p>Traditional monitoring asks whether software is available and fast. AI monitoring must also ask whether the system made the right decision.</p><p>An agent can return a technically successful response while choosing the wrong tool, using stale information, revealing sensitive data, entering a loop, or spending far more than expected. Teams therefore need to inspect the full sequence of decisions and actions&#8212;not only the final answer.</p><p>LangChain&#8217;s SmithDB and MLflow&#8217;s Q2 releases show this category maturing. The longer-term opportunity is larger than debugging. Production activity, human feedback, evaluation results, and user outcomes can become the data that continuously improves an agent. The platform that owns this record can become a critical system of record for AI quality.</p><h3>4. Inference economics are increasingly controlled by software</h3><p>The best model is often unnecessary for routine work. A production system may use a small model to classify a request, a specialist model to extract data, and a frontier model only for difficult reasoning.</p><p>Routing, caching, batching, quantization, and scheduling determine how efficiently these models run. In May, the vLLM team reported 230 output tokens per second for one DeepSeek V3.2 deployment&#8212;more than four times most providers in the cited comparison.* In June, a vLLM-Omni integration reduced one 30-billion-parameter model checkpoint from 66 GB to 25 GB and improved guided-generation speed by 1.55&#8211;1.67&#215; in the tested configuration.* <a href="https://vllm.ai/blog/2026-05-11-vllm-tops-artificial-analysis">Performance</a> &#183; <a href="https://vllm.ai/blog/2026-06-02-vllm-omni-autoround">Quantization</a></p><p>These results are workload-specific, but they illustrate an important point: software can materially change AI cost and performance without waiting for the next chip generation.</p><h3>5. Data, memory, permissions, and governance are converging</h3><p>AI becomes useful when it understands proprietary business context. That includes documents, transactions, customer history, policies, and real-time operational data.</p><p>But context must come with permissions. An agent needs to know not only what information exists, but who may access it and what actions it permits. Snowflake and Databricks are therefore bringing data governance, agent memory, model access, and tool execution closer together.</p><p>Incumbent data platforms have a distribution advantage because they already hold enterprise data and permissions. Independent companies can still win by providing cross-platform control, stronger security, better performance, or a neutral layer across fragmented systems.</p><h2>Capital is following these bottlenecks</h2><p>An illustrative group of four Q2 financing announcements totaled <strong>$232 million</strong>:</p><ul><li><p><strong>$32 million</strong> for Parasail&#8217;s distributed inference and training platform. <a href="https://www.prnewswire.com/news-releases/parasail-raises-32m-series-a-to-build-the-supercloud-that-puts-developers-in-control-of-their-ai-302742856.html">Source</a></p></li><li><p><strong>$20 million</strong> for Tensormesh&#8217;s inference-caching platform. <a href="https://www.morningstar.com/news/business-wire/20260527958597/tensormesh-raises-20m-from-investors-including-amd-ventures-coreweave-nventures-launches-tensormesh-inference-to-fix-ais-most-expensive-problem">Source</a></p></li><li><p><strong>$100 million</strong> for Runpod&#8217;s AI development and deployment platform. <a href="https://www.prnewswire.com/news-releases/runpod-raises-100m-led-by-summit-partners-to-accelerate-the-ai-developer-cloud-302808689.html">Source</a></p></li><li><p><strong>$80 million</strong> for Sail Research&#8217;s long-running agent infrastructure. <a href="https://www.prnewswire.com/news-releases/sail-research-raises-80-million-to-build-max-efficiency-infrastructure-for-ai-agents-302810497.html">Source</a></p></li></ul><p>This is not a complete market total, but it validates where demand is forming: inference efficiency, distributed execution, developer infrastructure, and reliable agent operation.</p><h2>What this means for investors</h2><p>We see five areas where durable value can emerge:</p><ol><li><p><strong>Agent runtime and orchestration:</strong> infrastructure that keeps long-running agents reliable and recoverable.</p></li><li><p><strong>Evaluation and observability:</strong> systems that measure task success, safety, cost, and continuous improvement.</p></li><li><p><strong>Inference optimization:</strong> routing, caching, scheduling, and other software that reduces cost or latency.</p></li><li><p><strong>Security and governance:</strong> identity, permissions, policy enforcement, and audit for agents that can act.</p></li><li><p><strong>Neutral infrastructure:</strong> platforms that work across models, clouds, data systems, and agent frameworks.</p></li></ol><p>Bundling will raise the bar. Startups offering only a thin feature may struggle as cloud and data platforms add similar functionality. Stronger companies will own a difficult technical layer, a critical workflow, proprietary operational data, or a cross-platform system of record.</p><h2>Hardware is the tailwind, not the full story</h2><p>Hardware progress remains important. NVIDIA reported that Vera Rubin delivers 10&#215; higher agent throughput at scale than Grace Blackwell and introduced software designed to optimize token output per megawatt.* <a href="https://nvidianews.nvidia.com/news/vera-rubin-full-production-agentic-ai-factory">Source</a></p><p>But larger compute budgets make software efficiency more valuable. A modest improvement in utilization, routing, memory management, or reliability can compound across enormous infrastructure spending. Hardware provides the capacity; software determines how productively it is used.</p><h2>What is next</h2><p>The market is moving from building individual agents to operating fleets of them. We expect agent authorization and gateways to become standard, evaluation to enter normal software-release processes, and model routing to become a core economic control.</p><p>The most important AI infrastructure companies may be the software systems that decide where intelligence runs, what it can access, whether it performed correctly, and how it improves. That is where I believe a substantial share of the next generation of AI infrastructure value will be created.</p><div><hr></div><p><em>Disclosure: Anyscale is a Sancus Ventures portfolio company. Lake is Founder and Managing Partner of Sancus. Company-reported performance claims are marked with an asterisk and should be interpreted within the workloads and configurations tested.</em></p>]]></content:encoded></item><item><title><![CDATA[AI M&A: Shifted From “Buying Models” to “Buying Control Points”]]></title><description><![CDATA[The big pattern: AI buyers are purchasing bottlenecks]]></description><link>https://lakedai.substack.com/p/ai-m-and-a-shifted-from-buying-models</link><guid isPermaLink="false">https://lakedai.substack.com/p/ai-m-and-a-shifted-from-buying-models</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Sat, 25 Apr 2026 04:07:51 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!-h4D!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!-h4D!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!-h4D!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png 424w, /__u/substackcdn.com/image/fetch/$s_!-h4D!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png 848w, /__u/substackcdn.com/image/fetch/$s_!-h4D!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png 1272w, /__u/substackcdn.com/image/fetch/$s_!-h4D!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!-h4D!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png" width="1456" height="971" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:971,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:1804245,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/195412336?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!-h4D!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png 424w, /__u/substackcdn.com/image/fetch/$s_!-h4D!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png 848w, /__u/substackcdn.com/image/fetch/$s_!-h4D!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png 1272w, /__u/substackcdn.com/image/fetch/$s_!-h4D!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92b605b8-fc7c-4975-abf3-c9027a99d1e3_1536x1024.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>The AI acquisition market has moved beyond &#8220;buy a model company.&#8221; <strong>The biggest acquirers are</strong> <strong>buying bottlenecks: secure cloud environments, governed enterprise data, identity, agent safety, coding workflows, AI data-center power, cooling, networking, chip interconnect, and robotics.</strong></p><p>That shift matters because it tells us where strategic value is forming. In 2023 and 2024, much of the venture market asked: &#8220;Who has the best model?&#8221; In 2025 and early 2026, the M&amp;A market is asking: &#8220;Who controls the layer that every model, agent, or AI application must pass through?&#8221;</p><p>AI demand is creating pressure on data centers, power, cooling, networking, and financing. Reuters reported that global data-center dealmaking hit a record through November 2025, with more than 100 transactions and nearly $61B in value; another Reuters report cited UBS data showing AI data-center and project-financing deals surging to $125B year-to-date from $15B in the comparable 2024 period.</p><p>Let&#8217;s take a quick look of some of the acquisitions in the last 6 months:</p><ol><li><p><strong>SpaceX &#8594; xAI</strong></p><ul><li><p><strong>Price:</strong> xAI valued at $250B; combined SpaceX/xAI valuation $1.25T</p></li><li><p><strong>Why acquired:</strong> Consolidate AI, compute, space, satellite distribution, and capital formation ahead of IPO path</p></li><li><p><strong>Strategic indication:</strong> Massive but unusual related-party consolidation<br></p></li></ul></li><li><p><strong>BlackRock GIP / EQT consortium &#8594; AES</strong></p><ul><li><p><strong>Price:</strong> $33.4B (including debt)</p></li><li><p><strong>Why acquired:</strong> Secure power generation for AI data center demand</p></li><li><p><strong>Strategic indication:</strong> Energy is becoming part of the AI stack<br></p></li></ul></li><li><p><strong>Google &#8594; Wiz</strong></p><ul><li><p><strong>Price:</strong> $32B</p></li><li><p><strong>Why acquired:</strong> Strengthen cloud + AI security across multicloud environments</p></li><li><p><strong>Strategic indication:</strong> Security is a core control layer in AI adoption<br></p></li></ul></li><li><p><strong>IBM &#8594; Confluent</strong></p><ul><li><p><strong>Price:</strong> ~$11B enterprise value</p></li><li><p><strong>Why acquired:</strong> Real-time data streaming and governance for enterprise AI</p></li><li><p><strong>Strategic indication:</strong> AI requires live, governed data pipelines<br></p></li></ul></li><li><p><strong>Permira / Warburg-led group &#8594; Clearwater Analytics</strong></p><ul><li><p><strong>Price:</strong> $8.4B (including debt)</p></li><li><p><strong>Why acquired:</strong> Investment/accounting platform enhanced with AI capabilities</p></li><li><p><strong>Strategic indication:</strong> PE is buying vertical workflow + data platforms<br></p></li></ul></li><li><p><strong>Salesforce &#8594; Informatica</strong></p><ul><li><p><strong>Price:</strong> ~$8B equity value</p></li><li><p><strong>Why acquired:</strong> Data governance, integration, and metadata for agentic AI</p></li><li><p><strong>Strategic indication:</strong> Trusted data layer is critical for AI agents<br></p></li></ul></li><li><p><strong>ServiceNow &#8594; Armis</strong></p><ul><li><p><strong>Price:</strong> $7.75B</p></li><li><p><strong>Why acquired:</strong> Cyber exposure management across IT, OT, IoT, and cloud</p></li><li><p><strong>Strategic indication:</strong> AI expands attack surface &#8594; security consolidation<br></p></li></ul></li><li><p><strong>Ecolab &#8594; CoolIT Systems</strong></p><ul><li><p><strong>Price:</strong> $4.75B (cash)</p></li><li><p><strong>Why acquired:</strong> Liquid cooling for AI data centers</p></li><li><p><strong>Strategic indication:</strong> Thermal management is a key AI bottleneck<br></p></li></ul></li><li><p><strong>SoftBank &#8594; DigitalBridge</strong></p><ul><li><p><strong>Price:</strong> ~$4B enterprise value</p></li><li><p><strong>Why acquired:</strong> Data centers and connectivity infrastructure</p></li><li><p><strong>Strategic indication:</strong> AI infra ownership and financing are strategic<br></p></li></ul></li><li><p><strong>Marvell &#8594; Celestial AI</strong></p><ul><li><p><strong>Price:</strong> $3.25B (cash + stock)</p></li><li><p><strong>Why acquired:</strong> Photonics for AI chip interconnect</p></li><li><p><strong>Strategic indication:</strong> Bandwidth and memory efficiency are critical<br></p></li></ul></li><li><p><strong>Meta &#8594; Manus</strong></p><ul><li><p><strong>Price:</strong> Undisclosed (~$2B&#8211;$3B reported)</p></li><li><p><strong>Why acquired:</strong> General-purpose AI agents</p></li><li><p><strong>Strategic indication:</strong> Platforms want agent capability, not just chat<br></p></li></ul></li><li><p><strong>Apple &#8594; Q.ai</strong></p><ul><li><p><strong>Price:</strong> Undisclosed (~$1.6B reported)</p></li><li><p><strong>Why acquired:</strong> Audio AI and edge intelligence</p></li><li><p><strong>Strategic indication:</strong> Device-native AI is becoming strategic<br></p></li></ul></li><li><p><strong>Mobileye &#8594; Mentee Robotics</strong></p><ul><li><p><strong>Price:</strong> ~$900M</p></li><li><p><strong>Why acquired:</strong> Humanoid robotics / embodied AI</p></li><li><p><strong>Strategic indication:</strong> Physical AI is entering strategic phase<br></p></li></ul></li><li><p><strong>Rumble &#8594; Northern Data</strong></p><ul><li><p><strong>Price:</strong> ~$767M (all-stock)</p></li><li><p><strong>Why acquired:</strong> AI cloud + GPU infrastructure</p></li><li><p><strong>Strategic indication:</strong> Compute access remains strategic<br></p></li></ul></li><li><p><strong>CrowdStrike &#8594; SGNL</strong></p><ul><li><p><strong>Price:</strong> $740M</p></li><li><p><strong>Why acquired:</strong> Identity security across humans, machines, AI</p></li><li><p><strong>Strategic indication:</strong> Identity = core AI security layer<br></p></li></ul></li><li><p><strong>Marvell &#8594; XConn</strong></p><ul><li><p><strong>Price:</strong> ~$540M</p></li><li><p><strong>Why acquired:</strong> Networking for AI data centers</p></li><li><p><strong>Strategic indication:</strong> Connectivity bottlenecks are consolidating<br></p></li></ul></li><li><p><strong>OpenAI &#8594; Neptune</strong></p><ul><li><p><strong>Price:</strong> Undisclosed (&lt; $400M reported)</p></li><li><p><strong>Why acquired:</strong> Model training observability</p></li><li><p><strong>Strategic indication:</strong> Internal AI tooling is strategic infrastructure</p></li></ul></li></ol><p>Two other deals matter strategically. <strong>OpenAI agreed to acquire Promptfoo</strong> for AI security testing and red-teaming, and <strong>OpenAI agreed to acquire Astral</strong> to deepen Codex and Python developer workflows; both were undisclosed, but they show how important agent safety and coding tools have become. <strong>Cohere&#8217;s acquisition of Aleph Alpha</strong> was also undisclosed, but it is a clear signal that sovereign and regulated-market AI are becoming strategic themes in Europe.</p><p><strong>SpaceX&#8217;s option to acquire Cursor for $60B</strong> is not yet a completed acquisition. But it is a powerful signal that AI coding tools have become strategically important enough to justify mega-deal structures.</p><div><hr></div><h2>Strategic indications from the last six months</h2><h3>1. Security is now one of the most valuable AI categories</h3><p>Wiz, Armis, SGNL, Promptfoo, and Seraphic all point in the same direction: AI creates new security surfaces. Enterprises now have to secure cloud workloads, code, runtime environments, non-human identities, AI agents, prompt injection risk, tool misuse, data leakage, and automated workflows.</p><p>This is why the security deals are not small. Google paid $32B for Wiz; ServiceNow paid $7.75B for Armis; CrowdStrike paid $740M for SGNL and also bought Seraphic for about $420M.</p><p>From a venture investment perspective, AI security should remain one of the strongest seed categories, but &#8220;security for AI&#8221; needs to be specific. Good wedges include agent identity, runtime authorization, AI red-teaming, evals, secure tool use, data-leak prevention, browser/workflow security, and compliance evidence.</p><h3>2. Data governance has become an AI platform feature</h3><p>IBM-Confluent and Salesforce-Informatica show that enterprise AI depends on governed, real-time, high-quality data. The acquirer logic is straightforward: if agents are going to act inside enterprise workflows, the platform needs to know which data is current, permitted, private, trustworthy, and auditable.</p><p>From a venture investment perspective, seed-stage data infrastructure is still investable, but it must be closer to the AI execution layer. A generic ETL or data catalog company is harder to underwrite. A company that gives agents trusted context, permissions, lineage, and evaluation loops is more strategically relevant.</p><h3>3. Compute is not enough; the bottlenecks are power, cooling, networking, and memory</h3><p>The AI infrastructure theme is broadening. DigitalBridge is about data centers and connectivity; CoolIT is about liquid cooling; Celestial AI and XConn are about photonics, switching, and connectivity; AES is about power. These are not all &#8220;AI startups,&#8221; but they are becoming AI M&amp;A because AI deployment requires physical infrastructure.</p><p>Big Tech&#8217;s AI infrastructure spending is also enormous. Reuters cited Bridgewater analysis estimating that Alphabet, Amazon, Meta, and Microsoft would collectively invest about <strong>$650B</strong> in AI-related infrastructure in 2026, up from <strong>$410B</strong> in 2025.</p><p>From a venture investment perspective, early-stage AI infrastructure can be attractive, but capital intensity is dangerous. Pre-seed and seed investors should prefer software-like infrastructure, infrastructure orchestration, utilization, scheduling, inference efficiency, power optimization, cooling intelligence, chip/runtime tooling, and design automation over businesses that require massive balance sheets.</p><h3>4. Developer tools and coding agents are becoming strategic assets</h3><p>OpenAI&#8217;s Astral deal and the SpaceX/Cursor option both indicate that developer workflow is one of the first AI markets with real usage, revenue, and strategic urgency. OpenAI said Astral&#8217;s tools would be brought into Codex, and Reuters reported that Codex had over 2 million weekly active users at the time of the Astral announcement.</p><p>From a venture investment perspective, coding-agent startups still have room, but they need a defensible wedge: codebase context, enterprise permissions, testing, migration, security, CI/CD integration, workflow memory, or vertical developer environments. A general-purpose &#8220;better coding assistant&#8221; will be hard to defend against OpenAI, Anthropic, Cursor, GitHub, JetBrains, Atlassian, and the hyperscalers.</p><h3>5. Physical AI is moving from theme to transaction</h3><p>Mobileye buying Mentee for roughly $900M, Tesla&#8217;s agreement to acquire an AI hardware company for up to $2B, Apple buying Q.ai for audio/edge AI, and SoftBank&#8217;s near-window ABB Robotics deal all point toward AI moving into devices, robotics, sensors, and embodied systems.</p><p>From a venture investment perspective, pre-seed robotics is investable, but founders need a narrow initial market. A &#8220;general humanoid labor&#8221; may not win against a constrained workflow: warehouse picking, inspection, eldercare assistance, manufacturing QA, lab automation, agriculture, logistics yards, or defense-adjacent autonomy.</p><h3>6. Sovereign AI and regulated-market AI are becoming acquisition themes</h3><p>Cohere buying Aleph Alpha is a signal that governments and regulated enterprises want AI they can trust, deploy locally, and govern under regional policy constraints. Reuters reported that the deal is aimed at government and business customers in highly regulated European markets, with Schwarz Group investing $600M in Cohere&#8217;s upcoming round.</p><p>From a venture investment perspective, sovereign AI is not just about building a local foundation model. More practical seed opportunities may sit in deployment, compliance, audit logs, data residency, model routing, private inference, procurement tooling, and regulated vertical workflows.</p><div><hr></div><h2>New Question for Investors</h2><p><strong>If AI adoption grows 10x, what breaks &#8212; and does this startup own the fix?</strong></p><p>For AI early-stage investments, you may want to ask:</p><ol><li><p><strong>Is this a feature, a product, or a control point?</strong><br>The best companies become unavoidable infrastructure inside a workflow.<br></p></li><li><p><strong>What proprietary asset compounds?</strong><br>This could be data, workflow memory, integrations, permissions, evaluations, customer-specific context, deployment footprint, or domain expertise.<br></p></li><li><p><strong>Who are the likely acquirers, and why would they need to own it? Does the startup create strategic urgency?</strong><br>Strong companies should have multiple credible strategic buyers. The best M&amp;A targets force incumbents to ask: &#8220;Can we afford not to own this?&#8221;<br></p></li><li><p><strong>Is there a budget owner today? Can the startup prove ROI quickly?</strong><br>&#8220;AI innovation&#8221; budgets can disappear. Security, compliance, infrastructure cost reduction, developer productivity, and revenue automation are more durable. In this market, seed companies need measurable cost savings, risk reduction, revenue lift, or time compression.<br></p></li><li><p><strong>Will the company survive platform bundling?</strong><br>If Microsoft, Google, Salesforce, ServiceNow, OpenAI, Anthropic, or CrowdStrike bundles the feature tomorrow, what remains defensible?</p></li></ol><div><hr></div><p>AI M&amp;A is no longer about models&#8212;it&#8217;s about control. The biggest checks are going to the layers that make AI work: security, data, identity, infrastructure, and robotics.<br>For early-stage VC, the message is simple: own a bottleneck, or get commoditized.<br>The winners won&#8217;t be &#8220;AI apps.&#8221; They&#8217;ll be the control layers everyone else depends on.</p>]]></content:encoded></item><item><title><![CDATA[Dual-Native Design for the AI Future]]></title><description><![CDATA[Intuitive for People, Structured for AI]]></description><link>https://lakedai.substack.com/p/dual-native-design-for-the-ai-future</link><guid isPermaLink="false">https://lakedai.substack.com/p/dual-native-design-for-the-ai-future</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Wed, 04 Mar 2026 23:47:34 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/21674063-8c40-4105-82eb-c75cdc6e887d_1312x816.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p></p><p>For decades, software infrastructure was built around one primary assumption: the user was a human. A person clicked buttons, read menus, filled forms, waited through flows, and made judgment calls at the edge. That assumption is starting to break. In 2024, automated traffic surpassed human traffic for the first time in a decade, accounting for 51% of all web traffic, according to Imperva. Not all of that traffic is useful AI agency, of course, but it is a clear signal that non-human actors are now a first-class part of the internet. At the same time, frontier products from OpenAI and Anthropic can now navigate websites, fill forms, use spreadsheets, and operate software on a user&#8217;s behalf.</p><p>The implication is bigger than &#8220;software should add AI features.&#8221; It means <strong>software now needs to serve two different classes of users at once: humans and agents. </strong>Designing for humans is still about usability, trust, and clarity. Designing for agents is about signals, structure, permissions, and control loops. The challenge is no longer just generating intelligence. It is turning intelligence into something <strong>legible enough to observe, safe enough to govern, and structured enough for an agent to act on</strong>.</p><p>You can already see this shift in how leading systems are being built. ChatGPT agent is designed to navigate websites, connect to external data sources, fill out forms, and edit spreadsheets, while pausing for confirmations and sensitive logins. Anthropic&#8217;s computer-use models were built for the opposite case: software that has no modern API, where the only way to automate work is to interact with the screen like a person. Anthropic now reports a 72.5% first-attempt success rate on OSWorld-Verified, and says its computer-use performance has risen from &#8220;the teens to the low 70s&#8221; in just over a year. The lesson is hard to ignore: the lack of an API is no longer a durable moat against automation.</p><p>That means modern software needs two interaction surfaces. One is the human surface: UI, workflows, collaboration, approvals, delight. The other is the agent surface: state that can be interpreted reliably, permissions that can be scoped cleanly, actions that can be audited, and outcomes that can be reversed when needed. The best products will not be &#8220;AI-first&#8221; in some vague sense. <strong>They will be dual-native: intuitive for people, structured for machines.</strong></p><p>The web has dealt with versions of this problem before. Robots.txt became an IETF standard through RFC 9309 as a way for service owners to express how automated crawlers should access content. But action-taking agents require a richer contract than crawlers ever did. They need identity, scope, and trust. OpenAI&#8217;s current allowlisting guidance for ChatGPT agent reflects exactly that shift: agent requests are signed using the HTTP Message Signatures standard so sites and CDNs can verify that the traffic is authentic. In other words, <strong>we are moving from &#8220;detect the bot&#8221; to &#8220;authenticate the agent.&#8221;</strong></p><p>This is why observability becomes a core product requirement rather than an ops accessory. In traditional systems, observability tells you what failed. In agentic systems, it also needs to tell you what the system believed, which tool it chose, what it changed, what it saw, and whether the action remained aligned with the user&#8217;s intent. OpenTelemetry&#8217;s work on AI agent observability makes this explicit: because agents are non-deterministic, telemetry is not just for debugging, but also a feedback loop for evaluation and improvement. The standards effort now spans agent traces, metrics, events, model spans, and even MCP-related telemetry. That is a strong signal that a new control plane for agents is beginning to form.</p><p>Alignment also starts to look less like a model-only question and more like an infrastructure question. OpenAI warns that prompt injection risk rises as AI systems gain access to more sensitive data and take on longer, more autonomous tasks. Anthropic calls prompt injection one of the most significant security challenges for browser-based agents because every webpage is a potential attack surface. That is why the leading systems are full of runtime constraints: user confirmations, takeover modes, watch modes, sandboxes, and scoped access. In the agent era, alignment is not just about what the model &#8220;knows.&#8221; It is about what the system is allowed to do, what it is prevented from doing, and what humans can verify before consequences become permanent.</p><p>For startups, this changes the product roadmap. <strong>The winning architecture will likely separate a human plane from an agent plane.</strong> The human plane optimizes for navigation, trust, collaboration, and exception handling. The agent plane optimizes for machine-readable state, scoped permissions, deterministic interfaces where possible, safe fallbacks where not, and policy-aware execution. Startups that simply bolt a chat box onto an old workflow may look intelligent, but they will struggle to deliver reliability. McKinsey&#8217;s 2025 survey captures that tension well: 62% of organizations say they are at least experimenting with AI agents, and 23% say they are scaling an agentic system somewhere in the enterprise. Yet only 39% report EBIT impact from AI, and only about 6% qualify as high performers. What separates those high performers is not enthusiasm; it is workflow redesign and clearer human-validation processes.</p><p>That distinction matters even more for investors. Gartner predicts that by 2028, 33% of enterprise software applications will include agentic AI, 15% of day-to-day work decisions will be made autonomously, and 60% of brands will use agentic AI to support one-to-one interactions. Those are not trivial tailwinds. But Gartner also predicts that more than 40% of agentic AI projects will be canceled by the end of 2027 because of cost, weak business value, or inadequate risk controls. So the market signal is not &#8220;buy anything with agent in the pitch deck.&#8221; It is the opposite: <strong>the most durable value will accrue to companies that make agents reliable, governable, and economically legible inside real workflows.</strong></p><p>The spending data already points in that direction. Menlo Ventures estimates enterprises spent $37 billion on generative AI in 2025, with $19 billion of that going to the application layer alone. By Menlo&#8217;s count, there are now at least 10 AI products generating more than $1 billion in ARR and 50 generating more than $100 million. Even allowing for the fact that this is a venture-backed market estimate rather than an audited industry total, the pattern is revealing: enterprises are paying not just for intelligence, but for products that operationalize intelligence. That is why some of the most interesting companies in this cycle may look less like model providers and more like the Datadog, Okta, Stripe, or Twilio equivalents for machine actors.</p><p>The deeper point is this: software infrastructure can no longer assume that the user is a person staring at a screen. Increasingly, the user may be an agent operating at machine speed, across multiple tools, with bounded autonomy and partial trust. The winners will be the companies that make this new class of user manageable &#8212; <strong>observable enough to govern, aligned enough to trust, structured enough to integrate, and useful enough that humans still want to stay in the loop. </strong>Designing for humans defined the last era of software. Designing for humans <strong>and</strong> agents will define the next one.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!A76u!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F97b2724c-e56d-481e-b752-12396a5a1851_1376x768.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!A76u!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F97b2724c-e56d-481e-b752-12396a5a1851_1376x768.png 424w, /__u/substackcdn.com/image/fetch/$s_!A76u!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F97b2724c-e56d-481e-b752-12396a5a1851_1376x768.png 848w, /__u/substackcdn.com/image/fetch/$s_!A76u!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F97b2724c-e56d-481e-b752-12396a5a1851_1376x768.png 1272w, /__u/substackcdn.com/image/fetch/$s_!A76u!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F97b2724c-e56d-481e-b752-12396a5a1851_1376x768.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!A76u!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F97b2724c-e56d-481e-b752-12396a5a1851_1376x768.png" width="1376" height="768" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/97b2724c-e56d-481e-b752-12396a5a1851_1376x768.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:768,&quot;width&quot;:1376,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:1574642,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/189932903?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F97b2724c-e56d-481e-b752-12396a5a1851_1376x768.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!A76u!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F97b2724c-e56d-481e-b752-12396a5a1851_1376x768.png 424w, /__u/substackcdn.com/image/fetch/$s_!A76u!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F97b2724c-e56d-481e-b752-12396a5a1851_1376x768.png 848w, /__u/substackcdn.com/image/fetch/$s_!A76u!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F97b2724c-e56d-481e-b752-12396a5a1851_1376x768.png 1272w, /__u/substackcdn.com/image/fetch/$s_!A76u!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F97b2724c-e56d-481e-b752-12396a5a1851_1376x768.png 1456w" sizes="100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p></p>]]></content:encoded></item><item><title><![CDATA[The 10 Biggest AI Stories of 2025 ]]></title><description><![CDATA[And How They&#8217;re Reshaping Business and Society]]></description><link>https://lakedai.substack.com/p/the-10-biggest-ai-stories-of-2025</link><guid isPermaLink="false">https://lakedai.substack.com/p/the-10-biggest-ai-stories-of-2025</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Wed, 24 Dec 2025 21:03:27 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!pUfJ!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>2025 marked the year artificial intelligence stopped feeling experimental and started behaving like critical infrastructure. The conversation shifted from &#8220;Can AI do this?&#8221; to &#8220;How do we scale, govern, and trust it?&#8221; Across models, hardware, regulation, and creative tools, the year delivered inflection points that will define the next decade.</p><p>Here are the ten most important AI developments of 2025 &#8212; and what they changed.</p><div><hr></div><h2>1. GPT-5 makes reasoning-first AI the default</h2><p>In August 2025, OpenAI released GPT-5, positioning it not just as a more powerful model but as a unified system capable of adapting its reasoning depth to the task at hand. Rather than forcing users to choose between speed and intelligence, GPT-5 introduced built-in &#8220;thinking modes,&#8221; alongside stronger multimodal capabilities, better tool use, and improved reliability across complex tasks like coding, research, and long-form writing.</p><p>The impact was a reset of user expectations. Knowledge workers <strong>began designing workflows around AI agents rather than single prompts</strong>, trusting models to handle multi-step reasoning with fewer guardrails. GPT-5 accelerated the shift from AI as a productivity enhancer to AI as a semi-autonomous collaborator, pressuring enterprises to rethink job design, review processes, and accountability structures. Competing labs were forced to respond not with marginal gains, but with demonstrable improvements in reasoning quality and trustworthiness.</p><div><hr></div><h2>2. OpenAI releases open-weight reasoning models</h2><p>Just days before GPT-5, OpenAI surprised the industry by releasing gpt-oss-120b and gpt-oss-20b, two open-weight reasoning models under an Apache 2.0 license. This marked one of the most significant open-source gestures ever made by a frontier AI lab, directly challenging the assumption that cutting-edge reasoning models would remain fully closed.</p><p>The move <strong>reshaped the open-versus-closed debate</strong>. Enterprises gained credible options for self-hosting advanced models, particularly in regulated or data-sensitive environments. At the ecosystem level, open-weight releases catalyzed better tooling for fine-tuning, evaluation, and deployment, while also forcing the industry to grapple with safety and misuse at scale. OpenAI&#8217;s decision legitimized a hybrid future in which openness and governance must evolve together.</p><div><hr></div><h2>3. Gemini 2.0 normalizes long-context, tool-using assistants</h2><p>Early in 2025, Google expanded Gemini 2.0 across its product ecosystem, highlighting massive context windows &#8212; up to two million tokens &#8212; and <strong>deep integration with tools like Search, code execution, and workspace applications.</strong> Gemini&#8217;s rollout signaled Google&#8217;s intent to make AI the default interface layer across consumer and enterprise software.</p><p>The impact was a shift in how people used AI day to day. Instead of summarizing snippets, models could ingest entire codebases, contracts, or research archives in one pass. Tool-using assistants moved from novelty to expectation, blurring the line between search engines, productivity software, and AI agents. For enterprises, this raised the bar on integration quality: AI needed to be deeply embedded in workflows, not bolted on.</p><div><hr></div><h2>4. Claude 4 turns coding agents into a competitive frontier</h2><p>In May 2025, Anthropic released Claude Opus 4 and Claude Sonnet 4, later followed by Sonnet 4.5, with a clear emphasis on advanced reasoning, code generation, and long-running agent workflows. Anthropic positioned Claude not just as a chatbot, but as a system capable of sustaining complex projects over time.</p><p>The impact was felt most strongly in software development. <strong>Coding became the most concrete and measurable domain for AI value creation</strong>, accelerating adoption of AI copilots and autonomous refactoring agents. As models proved capable of maintaining context across extended tasks, developers began trusting AI with larger portions of the software lifecycle. This intensified competition among IDEs, dev platforms, and infrastructure providers to integrate the most capable models with the best agent frameworks.</p><div><hr></div><h2>5. Llama 4 strengthens the open-weight ecosystem</h2><p>In April 2025, Meta introduced Llama 4, a new generation of open-weight, natively multimodal models designed to handle text, images, and long-context reasoning at scale. Meta framed Llama 4 as both a research contribution and a foundation for broad commercial experimentation.</p><p>The release reinforced the momentum behind open models. Developers and enterprises increasingly <strong>treated models as customizable components rather than fixed services, tailoring them to specific domains and products. </strong>As capable open-weight models proliferated, competitive advantage shifted away from model access alone toward distribution, proprietary data, and user trust. Llama 4 made it clear that openness was no longer a constraint on performance.</p><div><hr></div><h2>6. Generative video crosses a realism threshold</h2><p>In September 2025, OpenAI released Sora 2, dramatically improving video realism with synchronized dialogue, sound effects, and longer scene coherence. At the same time, Google DeepMind advanced its Veo video models, while companies like Runway pushed consistency across characters and locations, <strong>moving generative video closer to real production workflows</strong>.</p><p>The creative impact was immediate. Marketing, advertising, product demos, and internal communications began using AI-generated video as a default starting point. At the same time, the risks escalated. As realism increased, so did the potential for deepfakes, harassment, and misinformation, <strong>forcing renewed urgency around provenance, watermarking, and detection</strong>. Video generation in 2025 made clear that technical progress and social risk now scale together.</p><div><hr></div><h2>7. NVIDIA reframes AI as an industrial system</h2><p>At GTC 2025, NVIDIA unveiled Blackwell Ultra, GB300 rack-scale systems, and new networking technologies, promoting the concept of &#8220;AI factories.&#8221; NVIDIA emphasized not just faster chips, but entire systems optimized for training and inference at unprecedented scale, while navigating geopolitical constraints such as U.S.&#8211;China export controls.</p><p>The impact was a reframing of AI progress as an infrastructure problem. <strong>Power availability, cooling, networking, and supply chains became central bottlenecks alongside algorithms.</strong> Inference costs, particularly for reasoning-heavy models, emerged as a defining economic constraint. AI development increasingly resembled industrial planning, pushing enterprises and governments to think in terms of capacity, resilience, and national competitiveness.</p><div><hr></div><h2>8. The EU AI Act moves from theory to enforcement</h2><p>In 2025, the European Union&#8217;s AI Act entered its first enforcement phase, with bans on certain practices taking effect and obligations for general-purpose AI models beginning to apply. The EU also issued guidance clarifying how the law would affect large model developers and downstream deployers.</p><p>The impact was immediate operational change. <strong>AI governance shifted from policy discussions to implementation work: documentation, risk assessments, model evaluations, and internal training programs.</strong> Even companies outside Europe began aligning with EU standards to avoid fragmentation. The AI Act effectively created a new compliance market, making governance tooling and AI risk management a core part of enterprise AI strategy.</p><div><hr></div><h2>9. U.S. AI policy pivots toward deregulation and federal primacy</h2><p>In January 2025, the White House issued Executive Order 14179, signaling a shift toward reducing barriers to U.S. AI leadership. By December, a follow-up executive order outlined a national AI policy framework aimed at limiting fragmented state-level regulation and accelerating adoption.</p><p>The impact was strategic rather than immediate. Companies interpreted the shift as encouragement to scale AI deployments more aggressively, particularly in commercial and defense contexts. At the same time, tension grew between U.S. and EU regulatory philosophies, forcing global companies to navigate divergent compliance regimes. AI governance became inseparable from geopolitics and industrial policy.</p><div><hr></div><h2>10. Courts begin defining the legality of AI training data</h2><p>In mid-2025, U.S. courts issued landmark rulings in cases involving <strong>AI training on copyrighted books, including</strong> <strong>decisions favoring both Anthropic and Meta on fair-use grounds</strong>. These rulings were followed by high-profile settlements, underscoring the financial and legal stakes of training data practices.</p><p>The impact was a shift in how companies approach data strategy. Even with favorable rulings, uncertainty pushed many organizations toward l<strong>icensed, curated, or proprietary datasets to reduce legal and reputational risk</strong>. Data provenance and documentation became board-level concerns. While training legality gained some clarity, questions around generated outputs and market harm remain unresolved, ensuring this debate will continue.</p><div><hr></div><h2>What 2025 changed</h2><p>Taken together, 2025 marked the transition of AI from experimentation to infrastructure. Reasoning, scale, governance, and compute economics became the defining constraints. Open and closed models now coexist as strategic choices rather than ideological positions. And perhaps most importantly, AI&#8217;s future is no longer shaped solely by technologists, but equally by regulators, courts, energy grids, and global politics.</p><p>If 2024 was about discovering what AI could do, 2025 was about deciding <strong>how &#8212; and under what rules &#8212; it will be allowed to do it</strong></p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!pUfJ!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!pUfJ!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png 424w, /__u/substackcdn.com/image/fetch/$s_!pUfJ!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png 848w, /__u/substackcdn.com/image/fetch/$s_!pUfJ!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png 1272w, /__u/substackcdn.com/image/fetch/$s_!pUfJ!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!pUfJ!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png" width="1024" height="1536" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:1536,&quot;width&quot;:1024,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:2777480,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/182532716?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!pUfJ!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png 424w, /__u/substackcdn.com/image/fetch/$s_!pUfJ!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png 848w, /__u/substackcdn.com/image/fetch/$s_!pUfJ!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png 1272w, /__u/substackcdn.com/image/fetch/$s_!pUfJ!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F15fc22c3-3ffc-4265-b2fa-0796a21cb6ae_1024x1536.png 1456w" sizes="100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>.</p>]]></content:encoded></item><item><title><![CDATA[The Great AI Build-Out: How Compute, Power, and Policy Will Shape the Next Two Years]]></title><description><![CDATA[From GPUs to a New Utility Layer]]></description><link>https://lakedai.substack.com/p/the-great-ai-build-out-how-compute</link><guid isPermaLink="false">https://lakedai.substack.com/p/the-great-ai-build-out-how-compute</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Wed, 03 Dec 2025 01:56:45 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!senk!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p><strong>The last two years were about models.<br>The next two are about infrastructure.</strong></p><p>If you&#8217;re a CTO, founder, or investor, you&#8217;re no longer just betting on &#8220;AI&#8221; in the abstract&#8212;you&#8217;re implicitly betting on data centers, power grids, accelerators, and the software that makes all of that usable.</p><p>Sam Altman is talking about <strong>trillions</strong> of dollars of compute spend and tens of gigawatts of data&#8209;center capacity through the Stargate projects. Satya Nadella is warning that AI has to earn &#8220;social permission&#8221; to consume that much energy. IBM&#8217;s Arvind Krishna is openly skeptical that multi&#8209;trillion&#8209;dollar AI data&#8209;center capex pays off at today&#8217;s infrastructure costs. </p><p>In other words: even the people building this future aren&#8217;t sure the economics will work&#8212;<strong>unless</strong> we get the infrastructure story right.</p><p>Below I&#8217;ll walk through the major trends that actually matter over the next 12&#8211;24 months, in plain language and without leaning on any single think&#8209;piece.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!senk!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!senk!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png 424w, /__u/substackcdn.com/image/fetch/$s_!senk!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png 848w, /__u/substackcdn.com/image/fetch/$s_!senk!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png 1272w, /__u/substackcdn.com/image/fetch/$s_!senk!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!senk!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png" width="1456" height="971" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/f4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:971,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:1994740,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/180562103?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!senk!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png 424w, /__u/substackcdn.com/image/fetch/$s_!senk!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png 848w, /__u/substackcdn.com/image/fetch/$s_!senk!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png 1272w, /__u/substackcdn.com/image/fetch/$s_!senk!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff4a15b77-6354-4d29-a5d3-90d87e22ea48_1536x1024.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><div><hr></div><h3>1. AI infrastructure becomes a new utility layer</h3><p>Let&#8217;s start with the scale.</p><p>McKinsey now projects that by 2030, data centers will require <strong>about $6.7 trillion in capex</strong>, and AI&#8209;oriented data centers account for roughly <strong>$5.2 trillion</strong> of that. The IEA expects global data&#8209;center electricity use to <strong>more than 2x</strong> to around <strong>945 TWh by 2030</strong>, with AI being the biggest single driver. </p><p>On the ground, that looks like:</p><ul><li><p>Gigawatt&#8209;scale campuses in Texas, New Mexico, Ohio and beyond under the Stargate umbrella, aiming for 10 GW of capacity and around<strong> $500B in investment just for OpenAI&#8217;s needs</strong>. </p></li><li><p>Google announcing multi&#8209;billion&#8209;dollar AI data&#8209;center projects in India and elsewhere, sometimes starting at 1 GW of power draw for a single campus. </p></li></ul><p>From a distance, this looks like telecom or railroads in a previous century: enormous fixed assets, long payback periods, and strategic relevance to governments.</p><p>For tech leaders and founders, the takeaway is simple: <strong>GPU scarcity may ease; power, cooling, and capital will not.</strong> For investors, AI infra is already its own asset class, with economics closer to energy and real estate than to SaaS.</p><div><hr></div><h3>2. Hardware goes rack&#8209;scale and heterogeneous</h3><p>We&#8217;re exiting the era where &#8220;AI hardware&#8221; effectively meant &#8220;Nvidia H100 in the cloud.&#8221;</p><p>Nvidia&#8217;s Blackwell generation (B200 / B300 and GB200 systems) is designed as a rack&#8209;scale LLM machine. Independent benchmarks show <strong>an order&#8209;of&#8209;magnitude jump in LLM throughput per GPU</strong> versus Hopper, thanks to more compute and far more memory bandwidth. </p><p>At the same time:</p><ul><li><p><strong>Google&#8217;s Ironwood TPU</strong> advertises about <strong>10&#215; peak performance </strong>over TPU v5p and more than <strong>4&#215; better training and inference performance per chip </strong>than TPU v6e, specifically &#8220;for the age of inference.&#8221; </p></li><li><p><strong>TPUv7</strong> pushes that even further with 192 GB of memory and <strong>7.3 TB/s of bandwidth</strong>&#8212;roughly 2.5&#215; the bandwidth of v5p. </p></li><li><p><strong>AMD&#8217;s Instinct MI350</strong> line ships with <strong>288 GB of HBM3E and 8 TB/s of bandwidth per GPU</strong>, a configuration aimed at huge context windows and multi&#8209;trillion&#8209;parameter models. </p></li></ul><p>The message is - there won&#8217;t be a monoculture. Hyperscalers will push their own silicon (TPUs, Trainium&#8209;style parts) as hard as they can, while Nvidia and AMD compete on raw performance and ecosystem.</p><p>If you&#8217;re building systems in this world and you&#8217;re <em>not</em> one of the hyperscalers, the only sane stance is: <strong>assume heterogeneity</strong>. Design your inference layer, training jobs, and observability around the idea that some workloads will end up on Blackwell, some on MI350, some on TPUs, and occasionally on whatever your customer&#8217;s procurement team got a discount on.</p><p><strong>The winners on the software side will be the ones who make that heterogeneity disappear behind a coherent interface.</strong></p><div><hr></div><h3>3. Power and cooling become the real bottleneck</h3><p>Power is the new GPU.</p><p>The IEA&#8217;s base case has data&#8209;center electricity usage growing <strong>4x faster</strong> than overall electricity demand, hitting around 945 TWh by 2030. McKinsey adds a striking local angle: in the U.S., data centers could account for up to <strong>12% of national electricity use by 2030</strong>. </p><p>It&#8217;s not a theoretical problem anymore. Australia expects data&#8209;center power consumption to <strong>3x by 2030</strong>, potentially derailing regional net&#8209;zero goals. <br>BloombergNEF&#8217;s latest analysis shows U.S. data&#8209;center power demand on track for <strong>106 GW by 2035</strong>, and that&#8217;s already a revised <em>upward</em> forecast. </p><p>Satya Nadella&#8217;s line that <strong>AI needs &#8220;social permission&#8221; to consume this much energy</strong> is not PR fluff; it&#8217;s a recognition that power and land&#8209;use politics are about to collide with AI scaling plans. </p><p>Practically, this means:</p><ul><li><p><strong>Liquid cooling</strong> becomes standard in new AI builds. High&#8209;end accelerators now consume <strong>1,000&#8211;1,200 W per chip</strong>, so a rack can easily hit 80&#8211;100 kW if fully loaded. Air can&#8217;t handle that for long. </p></li><li><p>New capacity gravitates to places with cheap, reasonably clean power and cooperative regulators: think Texas, the Nordics, parts of India and the Middle East, Patagonia. </p></li></ul><p>For founders, there&#8217;s a clear opening around energy&#8209;aware scheduling, cooling optimization, and siting/risk tools for power&#8209;constrained regions. For investors, remember that some of the highest&#8209;beta AI bets will actually be in boring&#8209;sounding businesses like substation build&#8209;outs and chilled&#8209;water plants.</p><div><hr></div><h3>4. Inference economics: tokens become a commodity</h3><p>If hardware is drifting toward &#8220;AI power plant,&#8221; the output is tokens.</p><p>Inference prices are already in free&#8209;fall. DeepSeek&#8217;s R1 reasoner, a frontier&#8209;class open model, charges roughly <strong>$0.55 per million input tokens</strong> and <strong>$2.19 per million output tokens</strong>, with cache hits down to <strong>$0.14 / million</strong>&#8212;and later pricing rounds have pushed those numbers even lower, to the $0.03&#8211;0.04 range for some workloads. Multiple independent analyses from a16z and Epoch AI find that, for a fixed level of model performance, <strong>LLM inference prices have fallen on the order of 10&#215; per year</strong>, <strong>with some benchmarks seeing 40&#215; to 1,000&#215; annual price&#8209;drops.</strong></p><p> ~$60 per million tokens two or three years ago may now cost well under $1&#8211;2 per million tokens with efficient, modern providers.</p><p>At the same time, open&#8209;weight models are catching up fast. Epoch AI&#8217;s latest analysis estimates that the best open models are now, on average, <strong>about 3 months behind the absolute frontier</strong> in a composite capability index, with a historical lag range of roughly <strong>5&#8211;22 months</strong>. </p><p>If you&#8217;re paying for generic &#8220;chat&#8221; or &#8220;code completion,&#8221; that&#8217;s not an ideal combo:</p><ul><li><p>Commoditized models whose quality gap is shrinking.</p></li><li><p>A price war that keeps driving <strong>cost per million tokens toward zero</strong>.</p></li></ul><p>The infrastructure story here is the serving stack. Projects like <strong>vLLM</strong> and <strong>SGLang</strong> have shown that careful attention to scheduling, KV&#8209;cache management, and batching can deliver <strong>multi&#8209;fold throughput improvements</strong> <strong>on the same hardware.</strong></p><p>For most organizations, over the next 24 months the battle moves from &#8220;which model do we use?&#8221; to:</p><ul><li><p>How efficiently can we serve it?</p></li><li><p>How aggressively can we reuse context and cache?</p></li><li><p>How many small, domain&#8209;specific adapters can we hang off a single base?</p></li></ul><p>Think of high&#8209;throughput inference as the new SQL database: it&#8217;s not glamorous, but it&#8217;s where a lot of <strong>cost and reliability</strong> live.</p><div><hr></div><h3>5. Base models commoditize; moats move up the stack</h3><p>As open models close the gap and more national and corporate models appear, base models start to look less like moats and more like raw materials.</p><p>The real defensibility is creeping upwards into three layers:</p><ol><li><p><strong>Data plumbing</strong> &#8211; how you collect, clean, label, and protect data.</p></li><li><p><strong>Retrieval</strong> &#8211; vector search, hybrid retrieval, and caching that give models live access to facts instead of stuffing everything into parameters.</p></li><li><p><strong>Alignment &amp; policy</strong> &#8211; supervised fine&#8209;tuning, RL, safety filters, and governance tuned to a given domain.</p></li></ol><p>This is the infrastructure story behind &#8220;AI copilots for X.&#8221; You can swap out the base model in a well&#8209;designed copilot stack; you can&#8217;t easily swap out the event logs, embeddings, vector indices, and safety policies that make the product actually useful and compliant.</p><p>For founders, that&#8217;s good news: you don&#8217;t need to build a GPT&#8209;6 competitor to build something valuable. <strong>You need to own a loop of data &#8594; retrieval &#8594; feedback &#8594; alignment that others can&#8217;t recreate quickly.</strong></p><div><hr></div><h3>6. Reinforcement Learning (RL), &#8220;reasoning models,&#8221; and agents become infrastructure problems</h3><p>The other big shift is HOW we train the models on top.</p><p>OpenAI&#8217;s o&#8209;series (o1, o3&#8209;mini and successors) are the first widely deployed models whose marketing story is explicitly &#8220;reasoning&#8221; rather than &#8220;just bigger LLM,&#8221; and reporting around them makes clear that reinforcement&#8209;learning&#8209;style training is a key ingredient. </p><p>DeepSeek&#8217;s R1 reasoner shows the same pattern from the open side: a large&#8209;scale RL process over relatively modest supervision dramatically boosts performance on hard math and coding benchmarks, reaching rough parity with o&#8209;series models on some tasks. </p><p>This matters for infrastructure because RL and agents change the traffic pattern:</p><ul><li><p>You&#8217;re no longer just doing one&#8209;shot prompts. You&#8217;re running <strong>multi&#8209;step trajectories</strong> with tools, external calls, and sometimes code execution.</p></li><li><p>You need <strong>logging, replay, and evaluation</strong> that work over <em>sequences</em>, not just single responses.</p></li><li><p>Training becomes a loop where inference clusters are constantly generating rollouts while smaller training clusters update policies or adapters.</p></li></ul><p>In other words, RL and agents turn your inference platform into something that looks more like a trading system: <strong>lots of small decisions, continuous feedback, constant retraining of policies.</strong></p><p>The infra challenge for the next 24 months is to make this reliable and safe enough that enterprise teams don&#8217;t have to build their own RL labs just to get a competent agent.</p><div><hr></div><h3>7. Governance and &#8220;social permission&#8221; move from slideware to architecture</h3><p>Finally, there&#8217;s the human layer.</p><p>When Microsoft&#8217;s CEO says AI needs &#8220;social permission&#8221; to consume this much energy, he&#8217;s really saying: there&#8217;s a political and regulatory ceiling on unchecked scaling. </p><p>We&#8217;re already seeing:</p><ul><li><p>Local backlash in regions where AI data&#8209;center builds threaten to soak up a huge chunk of the grid. </p></li><li><p>Early moves from governments and regulators to demand more transparency and constraints on how models are trained, what data they use, and where they run. </p></li></ul><p>From an infrastructure perspective, that translates to:</p><ul><li><p><strong>Fine&#8209;grained policy engines</strong> around models (who can call what, using which data, under which jurisdiction).</p></li><li><p><strong>Audit&#8209;friendly logging</strong> so you can reconstruct what model, what version, and what context produced a given decision.</p></li><li><p><strong>Region&#8209;aware routing</strong> so workloads stay within the right legal and power boundaries.</p></li></ul><p>The teams that treat this as a core part of their infra story&#8212;not a bolt&#8209;on compliance checkbox&#8212;will be the ones that can sell into finance, health, and government as the regulation catches up.</p><div><hr></div><h3>So what should leaders actually <em>do</em> in the next 24 months?</h3><p>If you strip away the hype, the near&#8209;term agenda for serious teams looks something like this:</p><ul><li><p><strong>Move onto a modern, model&#8209;agnostic inference layer.</strong> Treat GPUs and model vendors as interchangeable behind a vLLM/SGLang&#8209;style runtime, whether you self&#8209;host or use a managed equivalent.</p></li><li><p><strong>Invest in data and retrieval infrastructure as if it were your core product.</strong> You want clean logs, embeddings, vector indices, and evaluation harnesses that survive model swaps.</p></li><li><p><strong>Design for heterogeneity and power constraints.</strong> Assume your future workloads will run on different accelerators in regions where power and cooling are the gating factors, not whether AWS has H100s in stock. </p></li><li><p><strong>Start logging and evaluating like you&#8217;ll be doing RL and agents soon.</strong> Even if you don&#8217;t train a reasoning model tomorrow, the data you collect now will decide how far you can go when you&#8217;re ready. </p></li></ul><p>The story of the next two years isn&#8217;t &#8220;one more crazy model demo.&#8221; It&#8217;s whether we can make the underlying infrastructure&#8212;hardware, power, serving, alignment, and governance&#8212;efficient and trustworthy enough that all that capex actually returns more than it burns.</p><p>That&#8217;s the opportunity and the risk, for the builders AND the capital behind them.</p><p></p><p>Reference:</p><ul><li><p><strong>Andreessen Horowitz (2024)</strong> &#8211; LLMflation analysis on falling inference costs</p></li><li><p><strong>Epoch AI (2024&#8211;2025)</strong> &#8211; LLM capability and cost-per-performance reports</p></li><li><p><strong>Business Insider (2025)</strong> &#8211; Coverage of Sam Altman&#8217;s statement on 10&#215; annual cost declines</p></li><li><p><strong>DeepSeek (2025)</strong> &#8211; <em>DeepSeek-R1 Pricing</em> (official docs / pricing page)</p></li><li><p><strong>365 Data Science (2025)</strong> &#8211; DeepSeek vs. OpenAI model pricing comparison</p></li><li><p><strong>NVIDIA (2025)</strong> &#8211; GB200 / NVL72 system performance claims</p></li><li><p><strong>Google Cloud (2025)</strong> &#8211; Ironwood TPU announcement and performance specifications</p></li><li><p><strong>AMD (2025)</strong> &#8211; Instinct MI350 series specifications and performance claims</p></li><li><p><strong>McKinsey &amp; Company (2025)</strong> &#8211; Global data-center and compute capex forecasts</p></li><li><p><strong>International Energy Agency (IEA, 2025)</strong> &#8211; AI-driven data-center electricity demand projections</p></li><li><p><strong>Carbon Brief (2025)</strong> &#8211; Analysis of global data-center and AI energy use trends</p></li></ul><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading LakeD-AI Unbundled! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[The B.O.A.R.D. Framework for Directors]]></title><description><![CDATA[Run AI Like a Business]]></description><link>https://lakedai.substack.com/p/run-ai-like-a-business</link><guid isPermaLink="false">https://lakedai.substack.com/p/run-ai-like-a-business</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Wed, 05 Nov 2025 18:59:29 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!KNz2!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>If AI is going to matter, it has to show up in the P&amp;L, not just the press release. That&#8217;s the mindset behind <strong>B.O.A.R.D.</strong>&#8212;a framework I created for boards to guide management, reduce risk, and accelerate value.</p><blockquote><p><strong>B.O.A.R.D.</strong> stands for <strong>Business value</strong>, <strong>Organization</strong>, <strong>Architecture &amp; Assets</strong>, <strong>Risk</strong>, and <strong>Dashboards</strong>.</p></blockquote><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!KNz2!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!KNz2!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png 424w, /__u/substackcdn.com/image/fetch/$s_!KNz2!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png 848w, /__u/substackcdn.com/image/fetch/$s_!KNz2!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png 1272w, /__u/substackcdn.com/image/fetch/$s_!KNz2!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!KNz2!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png" width="879" height="367" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:367,&quot;width&quot;:879,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:100118,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/178108558?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!KNz2!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png 424w, /__u/substackcdn.com/image/fetch/$s_!KNz2!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png 848w, /__u/substackcdn.com/image/fetch/$s_!KNz2!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png 1272w, /__u/substackcdn.com/image/fetch/$s_!KNz2!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3523be31-8d8a-4b5d-90e1-0e97b6cbc21c_879x367.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><div><hr></div><h2>B &#8212; Business value &amp; baseline</h2><p>List the company&#8217;s top AI use cases and tie each to a P&amp;L line (revenue, cost, or risk). Capture today&#8217;s baseline (cycle&#8209;time, error rate, $/transaction) so you can measure lift next quarter.</p><p><strong>Ask management: &#8220;Show the dollar impact per use case and the before/after baseline.&#8221;</strong></p><div><hr></div><h2>O &#8212; Organization &amp; operating model</h2><p>Appoint<strong> one</strong> accountable executive with budget and targets (CIO/CTO or a Chief AI &amp; Data Officer). Stand up one central platform for data, models, safety, and hosting; let business units (Sales, Service, Finance, Ops) deliver the outcomes.</p><p><strong>Board touchpoints:</strong> <strong>Strategy/Tech committee </strong>steers direction; <strong>Audit/Risk committee</strong> oversees controls and truthful disclosures; <strong>Nom &amp; Gov:</strong> board composition; AI literacy, policy updates; <strong>Comp &amp; Talent:</strong> reskilling &amp; redeployment.</p><div><hr></div><h2>A &#8212; Architecture &amp; assets (data, compute, power)</h2><p>Standardize data access, model/tool choices, and agent permissions (who/what they can change). Treat compute and power like a supply&#8209;chain input&#8212;approve a hosting/power plan alongside the AI plan.</p><p><strong>Ask management:</strong> &#8220;<strong>Where will this run, who has access, and what is our fallback of suppliers?</strong>&#8221;</p><div><hr></div><h2>R &#8212; Risk, regulation &amp; responsible AI</h2><p>Adopt lightweight guardrails mapped to existing risk management framework such as NIST AI RMF and the 2024 Generative AI Profile; use ISO/IEC 42001 as your audit target; track EU AI Act dates if relevant.</p><p><strong>Suggested guardrails:</strong> human&#8209;in&#8209;the&#8209;loop for sensitive actions, least&#8209;privilege for agents, an incident/rollback plan, and legal review of any &#8220;AI&#8221; claims.</p><p><strong>Helpful links:</strong></p><ul><li><p>NIST AI RMF home: <a href="https://www.nist.gov/itl/ai-risk-management-framework?utm_source=chatgpt.com">https://www.nist.gov/itl/ai-risk-management-framework</a></p></li><li><p>AI RMF 1.0 (PDF): <a href="https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf?utm_source=chatgpt.com">https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf</a></p></li><li><p>Generative AI Profile (2024) overview: <a href="https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence?utm_source=chatgpt.com">https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence</a> &#8212; PDF: <a href="https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf?utm_source=chatgpt.com">https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf</a></p></li><li><p>ISO/IEC 42001 overview: <a href="https://www.iso.org/standard/42001?utm_source=chatgpt.com">https://www.iso.org/standard/42001</a></p></li></ul><div><hr></div><h2>D &#8212; Dashboards &amp; decisions</h2><p>Run AI like operations, not experiments. Review a metric scorecard each quarter and decide: <strong>fund, scale, fix, or sunset</strong>.</p><p><strong>Sample metrics:</strong> cycle&#8209;time, $ / transaction, adoption %, incident rate, unit cost (per 1k inferences), and compute/power footprint.</p><div><hr></div><h2>Trends boards should track</h2><ul><li><p><strong>LLMs &#8594; Multimodal:</strong> voice, vision, and text in one model. Impact: frontline workflows&#8212;voice service, visual QA, document + image review.</p></li><li><p><strong>Models &#8594; Agents:</strong> systems that plan and call tools/APIs under guardrails for end&#8209;to&#8209;end automation.</p></li><li><p><strong>Compute &amp; Energy constraints</strong></p></li><li><p><strong>IP &amp; brand protection move center stage:</strong></p><ul><li><p><strong>U.S. Copyright Office:</strong> human authorship is required; AI&#8209;assisted works can be registrable if human creativity is evident.</p></li><li><p><strong>Content provenance:</strong> adopt C2PA Content Credentials for outward&#8209;facing media.</p></li><li><p><strong>Vendor indemnities:</strong> e.g., Microsoft Copilot Copyright Commitment&#8212;<em>with required mitigations</em>.</p></li></ul></li></ul><p><strong><br>Ten questions directors can ask right now</strong></p><ol><li><p>Top 3 AI use cases tied to P&amp;L/risk this year?</p></li><li><p>Data sources and governance for each?</p></li><li><p>Compute/Power/hosting plan?</p></li><li><p>Where do agents have <strong>write</strong> access? What is the approvals process and who is supervising?</p></li><li><p>Vendor/model portfolio + exit/fallback plans?</p></li><li><p>How do we avoid AI&#8209;washing in disclosures?</p></li><li><p>How are we measuring ROI and cycle&#8209;time reductions?</p></li><li><p>Workforce exposure map and upskilling plan?</p></li><li><p>EU AI Act exposure/readiness dates?</p></li><li><p>AI incident response tabletop this year?</p></li></ol><div><hr></div><h2>Talent &amp; workforce </h2><ul><li><p><strong>Exposure:</strong> credible 2024 estimates place a significant share of jobs as exposed to AI (higher in advanced economies). Prepare for job redesign, not just job loss.</p></li><li><p><strong>Productivity:</strong> field evidence shows meaningful gains&#8212;call&#8209;center productivity up ~14% on average; developers up to ~55% faster on defined tasks with copilot tools. Upside comes with variance: guardrails and training matter.</p></li></ul><p><strong>What to do now</strong></p><ol><li><p>Build a <strong>task&#8209;level exposure map</strong> (not just by role).</p></li><li><p>Launch an <strong>upskilling program</strong> (prompting, judgment, QA) with proficiency rubrics.</p></li><li><p>Publish an <strong>agent ops playbook</strong> (least&#8209;privilege permissions, escalation paths, rollback).</p></li></ol><div><hr></div><h2>Board Committees </h2><ul><li><p><strong>Audit &amp; Risk:</strong> internal controls, model risk, data lineage, third&#8209;party risk, <strong>disclosure integrity</strong> (avoid AI&#8209;washing). Align to NIST AI RMF + 2024 Gen&#8209;AI Profile; consider ISO/IEC 42001 readiness.</p></li><li><p><strong>Tech/Innovation (or Strategy):</strong> platform standards, power/hosting location strategy, make/buy/open decisions.</p></li><li><p><strong>Nom &amp; Gov:</strong> board AI literacy, policy updates (responsible AI, provenance), EU AI Act oversight.</p></li><li><p><strong>Comp &amp; Talent:</strong> exposure map, reskilling incentives, redeployment tracking.</p></li><li><p><strong>Separate AI/Data committee?</strong> Consider if you&#8217;re highly regulated, safety&#8209;critical, or AI&#8209;native at scale.</p></li></ul><div><hr></div><div><hr></div><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading LakeD-AI Unbundled! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[7 AI Trends Shaping What Comes Next]]></title><description><![CDATA[Q2 2025 offered clear signals that AI is becoming embedded across business operations, infrastructure, and policy. This briefing highlights seven key trends&#8212;each paired with a recommended read.]]></description><link>https://lakedai.substack.com/p/7-ai-trends-shaping-what-comes-next</link><guid isPermaLink="false">https://lakedai.substack.com/p/7-ai-trends-shaping-what-comes-next</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Tue, 29 Jul 2025 21:17:17 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!nsFe!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!nsFe!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!nsFe!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png 424w, /__u/substackcdn.com/image/fetch/$s_!nsFe!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png 848w, /__u/substackcdn.com/image/fetch/$s_!nsFe!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png 1272w, /__u/substackcdn.com/image/fetch/$s_!nsFe!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!nsFe!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png" width="474" height="314" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:314,&quot;width&quot;:474,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:233441,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/169595097?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!nsFe!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png 424w, /__u/substackcdn.com/image/fetch/$s_!nsFe!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png 848w, /__u/substackcdn.com/image/fetch/$s_!nsFe!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png 1272w, /__u/substackcdn.com/image/fetch/$s_!nsFe!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0eb4c3ae-a6b6-443e-81f3-bbfc446453d3_474x314.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><h3><strong>1. </strong><em><strong>AI is now an operating metric, not an R&amp;D line&#8209;item</strong></em></h3><p>Alphabet&#8217;s Q2 call framed AI contributions in the same breath as revenue growth, MAUs and backlog. Expect investors to demand similar KPI granularity from every tech&#8209;adjacent firm.</p><p><strong><a href="https://abc.xyz/2025-q2-earnings-call/">Alphabet Q2&#8239;2025 Earnings Call Transcript</a></strong> &#8211; the canonical example of a CEO quantifying &#8220;tokens processed&#8221; and revenue uplift in a single breath.</p><p><strong><a href="https://www.ainvest.com/news/microsoft-release-q4-earnings-july-30-2025-2507/">Microsoft: AI Run&#8209;Rate Tops&#8239;$13&#8239;B; 16&#8239;pts of Azure Growth</a></strong> &#8211; concise pre&#8209;earnings brief on how Redmond now reports AI ARR alongside cloud KPIs.</p><p><strong><a href="https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-at-work">McKinsey &#8220;Superagency in the Workplace&#8221; Report</a></strong> &#8211; survey data on C&#8209;suite pressure to shift from pilot counts to ROI dashboards.</p><h3><strong>2. The centre of gravity has shifted from </strong><em><strong>models</strong></em><strong> to </strong><em><strong>agents</strong></em></h3><p>The practical takeaway: design products around task graphs and APIs, not prompts.</p><p><strong><a href="https://www.ibm.com/think/insights/ai-agents-2025-expectations-vs-reality">IBM&#8239;Think: &#8220;AI Agents in&#8239;2025 &#8211; Expectations vs. Reality&#8221;</a></strong> &#8211;  assessment of tool&#8209;use, planning loops and failure modes.</p><h3><strong>3. Compute is the new supply&#8209;chain risk</strong></h3><p>AI&#8209;grade silicon is doubling in aggregate FLOPs every nine months, yet power demand is pacing just as fast. Budget not only for GPUs but also for megawatts and cooling footprints.</p><p><strong><a href="https://www.wsj.com/business/energy-oil/ai-data-center-power-costs-bbfcd862">WSJ: &#8220;AI Boom Sparks Fight Over Soaring Power Costs&#8221;</a></strong> &#8211; who pays for the 200&#8209;GW grid upgrade.</p><p><strong><a href="https://www.deloitte.com/us/en/insights/industry/power-and-utilities/data-center-infrastructure-artificial-intelligence.html">Deloitte 2025 AI Infrastructure Survey</a></strong> &#8211; granular cost curves for GPUs, coolant, land and megawatts.</p><p><strong><a href="https://www.reuters.com/breakingviews/ai-power-demand-is-generating-hallucinations-2025-05-20/">Reuters Breakingviews: &#8220;AI Power Demand Is Generating Hallucinations&#8221;</a></strong> &#8211; skeptical look at electricity forecasts (12&#8239;% of US load by&#8239;2030).</p><h3><strong>4. Loose oversight is driving a surge in investment</strong></h3><p>With Washington signalling a &#8220;green&#8209;light&#8221; stance, cap&#8209;ex has become a check&#8209;box exercise; the four U.S. AI majors alone will spend &gt;&#8239;$300&#8239;B this year. The window for regulatory arbitrage is open&#8212;but will snap shut suddenly.</p><p><strong><a href="https://www.brookings.edu/articles/states-are-legislating-ai-but-a-moratorium-could-stall-their-progress/">Brookings: &#8220;States Are Legislating AI, but a Federal Moratorium Looms&#8221;</a></strong> &#8211; outlines the emerging 50&#8209;state patchwork.</p><p><strong><a href="https://www.rstreet.org/commentary/ai-policy-in-congress-mid-2025-where-are-we-headed-next/">R&#8239;Street: &#8220;AI Policy in Congress Mid&#8209;2025&#8221;</a></strong> &#8211; tracks 1,000+ bills introduced and why none pass.</p><h3><strong>5. Markets love AI, but multiples are normalising</strong></h3><p>Q2 ended with the S&amp;P&#8239;500 at an ATH, yet AI equities only modestly outperformed. The message: growth stories now need hard adoption metrics, not just model demos.</p><p><strong><a href="https://www.barrons.com/articles/ai-stocks-valuation-buy-6bcc05d7">Barron&#8217;s: &#8220;AI Is Just One Reason Stocks Are Pricey &#8211; Yet Still a Buy&#8221;</a></strong> &#8211; puts the S&amp;P&#8209;500 22&#215; forward P/E in context.</p><p><strong><a href="https://www.reuters.com/markets/us/ai-gravity-defying-us-gdp-2025-07-23/">Reuters: &#8220;AI &amp; Gravity&#8209;Defying US&#8239;GDP&#8221;</a></strong> &#8211; links 40&#8239;%+ order&#8209;book growth to tempered market re&#8209;ratings.</p><h3><strong>6. Trust &amp; transparency are becoming product features</strong></h3><p>XAI research is moving from post&#8209;hoc saliency maps to <em>meta&#8209;reasoning</em>&#8212;models that expose their own chains&#8209;of&#8209;thought. Teams shipping enterprise AI should treat explainability as a first&#8209;class acceptance criterion.</p><p><strong><a href="https://www.livescience.com/technology/artificial-intelligence/ai-could-soon-think-in-ways-we-dont-even-understand-evading-efforts-to-keep-it-aligned-top-ai-scientists-warn">LiveScience summary of Google/OpenAI CoT&#8209;monitoring paper</a></strong> &#8211; Researchers at Google and OpenAI, among other companies, have warned that we may not be able to monitor AI's decision-making process for much longer.</p><p><strong><a href="https://www.aryaxai.com/article/top-10-ai-research-papers-of-april-2025-advancing-explainability-ethics-and-alignment">AryaXAI &#8220;Top 10 XAI Papers,&#8239;April&#8239;2025&#8221;</a></strong> &#8211; curated lightning&#8209;round of the quarter&#8217;s research breakthroughs.</p><h3><strong>7. Monetization opportunities are expanding</strong></h3><p>From YouTube Shorts&#8217; AI&#8209;boosted ad conversions to real&#8209;time dubbing for live events, new revenue streams are emerging wherever AI shortens content&#8211;audience distance. Build pricing models around incremental reach and latency reduction.</p><p><strong><a href="https://musically.com/2025/07/24/youtube-ad-revenues-still-growing-as-shorts-gets-more-ai-tools/">Music&#8239;Ally: &#8220;YouTube Shorts Revenues Now Match Long&#8209;Form&#8221;</a></strong> &#8211; confirms CEO quote that Shorts earn equal or higher RPM than in&#8209;stream ads.</p><p><strong><a href="https://www.singlegrain.com/artificial-intelligence/how-ai-localization-accelerates-global-market-expansion/">Meta AI Ad Engine Roadmap (Investor&#8217;s&#8239;Business&#8239;Daily)</a></strong> &#8211; outlines $28&#8239;B upside from end&#8209;to&#8209;end AI ad creation.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading LakeD-AI Unbundled! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Blank-Canvas Thinkers: Next Unicorns in an AI‑Native World]]></title><description><![CDATA[Why the next generation of billion&#8209;dollar companies may be founded by blank&#8209;canvas thinkers&#8212;and how investors should adapt.]]></description><link>https://lakedai.substack.com/p/blank-canvas-thinkers-next-unicorns</link><guid isPermaLink="false">https://lakedai.substack.com/p/blank-canvas-thinkers-next-unicorns</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Mon, 23 Jun 2025 15:14:52 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!21zI!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p></p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!21zI!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!21zI!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png 424w, /__u/substackcdn.com/image/fetch/$s_!21zI!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png 848w, /__u/substackcdn.com/image/fetch/$s_!21zI!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png 1272w, /__u/substackcdn.com/image/fetch/$s_!21zI!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!21zI!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png" width="1092" height="614" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/a99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:614,&quot;width&quot;:1092,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:706509,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/166600970?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!21zI!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png 424w, /__u/substackcdn.com/image/fetch/$s_!21zI!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png 848w, /__u/substackcdn.com/image/fetch/$s_!21zI!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png 1272w, /__u/substackcdn.com/image/fetch/$s_!21zI!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa99f0e54-40b8-4e56-871d-6f789dcd6b98_1092x614.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><div><hr></div><h2>This Week at the NYSE</h2><p>In my interview with <strong>NYSE WIRED </strong>this week, I shared <em>How to Hunt Unicorns</em>&#8212;an observation distilled from three decades of backing breakout founders. Three take&#8209;aways resonated most: </p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading LakeD-AI Unbundled! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><ol><li><p><strong>AI Compresses the Startup Clock</strong> &#8211; With foundation&#8209;model APIs and serverless infra, capital efficiency is up 10&#215;; the path from idea to first $1&#8239;M ARR now averages just 15&#8239;months. </p></li><li><p><strong>Rise of the Young Solo Founder</strong> &#8211; College students, and even high&#8209;schoolers, are launching venture&#8209;scale AI projects; their lack of legacy constraints fuels category&#8209;defining creativity. </p></li><li><p><strong>Beginner&#8217;s Mind Is a Moat</strong> &#8211; As I said on&#8209;air, &#8220;Generational innovation belongs to those who think fresh.&#8221; Practising Beginner&#8217;s Mind isn&#8217;t optional&#8212;it&#8217;s survival. </p></li></ol><div><hr></div><h2>The Bill Gurley Lens</h2><p>Benchmark&#8217;s Bill&#8239;Gurley once quipped, <strong>&#8220;Backing a repeat entrepreneur in the enterprise sector is near risk&#8209;free.&#8221;</strong>  The quote still stays true, yet the definition of &#8220;risk&#8221; is shifting. When infrastructure costs plunge and distribution is frictionless, <em>youthful audacity</em> could out&#8209;execute seasoned caution.</p><div><hr></div><h2>The Data Behind the Shift</h2><p><em>Cheaper compute, open&#8209;source weights, and plug&#8209;and&#8209;play tooling have collapsed costs and timelines while widening the founder funnel.</em></p><p>2015 vs. 2025 at a glance</p><ul><li><p>Cost to train an ImageNet&#8209;scale model:&#8239;&gt;$10&#8239;M &#8594; &lt;$100&#8239;K&#185;</p></li><li><p>Open&#8209;source models on Hugging&#8239;Face:&#8239;&lt;50 &#8594; &gt;500&#8239;000&#178;</p></li><li><p>Median time for a US SaaS startup to reach $1&#8239;M ARR:&#8239;~30&#8239;months &#8594; ~15&#8239;months&#179;</p></li><li><p>VC&#8209;backed AI startups with founders under 30:&#8239;18&#8239;% &#8594; 34&#8239;%&#8308;</p></li></ul><div><hr></div><h2>Pattern Recognition: Platform Shifts &amp; Young Founders</h2><p>A quick snapshot of how &#8220;too&#8209;young&#8221; founders have won each technology wave:</p><ul><li><p><strong>PC Era (1975&#8209;85)</strong><br>&#8226; <em>Average breakout&#8209;founder age:</em> <strong>24</strong><br>&#8226; <em>Young standouts:</em> Bill&#8239;Gates (20, Microsoft), Steve&#8239;Jobs (21, Apple)</p></li><li><p><strong>Web 1.0 (1994&#8209;2000)</strong><br>&#8226; <em>Average age:</em> <strong>28</strong><br>&#8226; <em>Young standouts:</em> Larry&#8239;Page (24, Google), Sergey&#8239;Brin (24, Google)</p></li><li><p><strong>Mobile / Cloud (2007&#8209;15)</strong><br>&#8226; <em>Average age:</em> <strong>26</strong><br>&#8226; <em>Young standouts:</em> Evan&#8239;Spiegel (22, Snapchat), Brian&#8239;Chesky (26, Airbnb)</p></li><li><p><strong>AI&#8209;Native (2020&#8209;&#8209;?)</strong><br>&#8226; <em>Average age:</em> <strong>&#8776;25</strong><br>&#8226; <em>Young standouts:</em> Alexandr&#8239;Wang (19, Scale&#8239;AI), Aidan&#8239;Gomez (23, Cohere), Henrique&#8239;Dubugras &amp; Pedro&#8239;Franceschi (22/23, Brex)</p></li></ul><div><hr></div><h2>Meet the New Vanguard</h2><ol><li><p><strong>Alexandr&#8239;Wang &#8212; Scale&#8239;AI</strong> (19) &#8226; Unlocked the data bottleneck; now <strong>$29&#8239;B</strong> val (2025).</p></li><li><p><strong>Henrique&#8239;Dubugras &amp; Pedro&#8239;Franceschi &#8212; Brex</strong> (22/23) &#8226; Re&#8209;imagined underwriting with LLM risk&#8209;models; <strong>$12&#8239;B</strong> val in &lt;4&#8239;yrs.</p></li><li><p><strong>Aidan&#8239;Gomez &#8212; Cohere</strong> (23) &#8226; Co&#8209;authored <em>Attention Is All You Need</em>; running a <strong>$5&#8239;B</strong> model lab.</p></li><li><p><strong>Dylan&#8239;Field &#8212; Figma</strong> (20) &#8226; Pioneered multiplayer, AI&#8209;assisted design; &gt;<strong>$600&#8239;M</strong> ARR.</p></li><li><p><strong>Arthur&#8239;Mensch &#8212; Mistral&#8239;AI</strong> (30) &#8226; Open&#8209;weights blitzkrieg proved speed &gt; incumbency.</p></li></ol><div><hr></div><h2>Five Structural Tailwinds Favouring the Young</h2><ol><li><p><strong>Serverless AI primitives</strong> &#8211; OpenAI, Anthropic, Mistral turn frontier models into &#162;&#8209;per&#8209;token utilities.</p></li><li><p><strong>Composable build layer</strong> &#8211; Retool, Supabase, LangChain, Vercel = MVP in days, not quarters.</p></li><li><p><strong>Instant distribution</strong> &#8211; TikTok, Discord, Product Hunt hand teenagers a global launch pad.</p></li><li><p><strong>Automated back&#8209;office</strong> &#8211; Mercury, Ramp, Pulley reduced CFO chores until Series&#8239;B.</p></li><li><p><strong>Capital velocity</strong> &#8211; Rolling SAFEs + AI accelerators (YC&#8239;W25 founder age = 26) wire six&#8209;figure checks within weeks.</p></li></ol><div><hr></div><h2>Investor Takeaways (Including Ours)</h2><p>We still agree with Bill&#8239;Gurley&#8217;s view that backing repeat enterprise founders is &#8220;near risk&#8209;free&#8221;&#8212;historically, over <strong>70&#8239;%</strong> of <strong><a href="https://sancusvc.com/">Sancus Ventures</a></strong>&#8209;backed founders are repeat entrepreneurs and <strong>20&#8239;%</strong> have already built unicorns&#8212; but <em>risk&#8209;adjusted return</em> is drifting toward first&#8209;timers, especially in the consumer space. That&#8217;s why we now allocates a pool to young, AI&#8209;native teams to find the next dorm&#8209;room unicorn.</p><blockquote><p><strong>Bottom line:</strong> The next $100&#8239;B AI company may be conceived by a 19&#8209;year&#8209;old with a GPU credit, not a 45&#8209;year&#8209;old with a Fortune&#8239;500 r&#233;sum&#233;.</p></blockquote><p></p><p>&#185; Lambda&#8239;Labs pricing, May&#8239;2025<br>&#178; Hugging&#8239;Face Model Index, June&#8239;2025<br>&#179; Kruze Consulting SaaS Benchmark, 2024<br>&#8308; PitchBook Emerging Tech Report, Q1&#8239;2025</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading LakeD-AI Unbundled! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Behind Meta's $14.3B Mega-Deal with Scale AI]]></title><description><![CDATA[What&#8217;s Behind the Bet&#8212;and What It Signals for AI Infra and Investors]]></description><link>https://lakedai.substack.com/p/behind-metas-143b-mega-deal-with</link><guid isPermaLink="false">https://lakedai.substack.com/p/behind-metas-143b-mega-deal-with</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Thu, 19 Jun 2025 22:11:49 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!d85p!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!d85p!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!d85p!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg 424w, /__u/substackcdn.com/image/fetch/$s_!d85p!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg 848w, /__u/substackcdn.com/image/fetch/$s_!d85p!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg 1272w, /__u/substackcdn.com/image/fetch/$s_!d85p!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!d85p!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg" width="1456" height="971" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/b5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:971,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:1932759,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/jpeg&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/166355462?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!d85p!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg 424w, /__u/substackcdn.com/image/fetch/$s_!d85p!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg 848w, /__u/substackcdn.com/image/fetch/$s_!d85p!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg 1272w, /__u/substackcdn.com/image/fetch/$s_!d85p!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5a57621-fbbb-4e47-aaa0-08d39fa9908e_1536x1024.jpeg 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Scale AI's recent valuation surge to nearly $29 billion, driven by Meta&#8217;s bold $14.3 billion investment, signals an important development in the AI infrastructure market. The doubling of Scale&#8217;s valuation in under a year underscores the strategic significance of high-quality, human-supervised data as the new scarce resource&#8212;much like GPUs were in the recent past.</p><h3>Why Scale&#8217;s Valuation Exploded</h3><p>Meta&#8217;s investment isn't just about capital&#8212;it&#8217;s about strategic advantage:</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading LakeD-AI Unbundled! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><ul><li><p><strong>Exclusive Data Access:</strong> Scale provides crucial evaluation, red-teaming, and RLHF (reinforcement learning with human feedback) services, highlighting the strategic importance and rapid growth of such data-centric services, essential to AI performance.</p></li><li><p><strong>Talent Acquisition and Time Compression:</strong> Bringing Alexandr Wang and his team closer to Meta significantly accelerates Meta&#8217;s AI roadmap, particularly around super-intelligent systems, potentially cutting development timelines by several quarters.</p></li></ul><h3>How Scale Aligns with AI Infrastructure Trends</h3><p>Scale AI&#8217;s ascension reflects key trends reshaping AI infrastructure:</p><ul><li><p><strong>Data Dominance Over Compute:</strong> With compute availability expanding, the bottleneck shifts toward data quality and availability. Gartner predicts the market for AI training datasets alone will surpass $5 billion by 2027.</p></li><li><p><strong>Continuous Evaluation and Feedback Loops:</strong> AI systems increasingly demand ongoing oversight and rigorous evaluations, reflecting software&#8217;s shift to continuous integration and deployment (CI/CD). McKinsey forecasts a 25% annual growth rate for AI governance and evaluation tools through 2030.</p></li><li><p><strong>Specialized Vertical Platforms:</strong> Scale&#8217;s Donovan agents and vertical-specific offerings indicate a future driven by specialized, agent-driven SaaS tailored to specific industries. IDC projects that by 2026, over 70% of enterprises will adopt specialized AI agent platforms.</p></li></ul><h3>Ripple Effects and Competitor Moves</h3><p>Meta&#8217;s  move triggered immediate industry responses:</p><ul><li><p><strong>Customer Realignment:</strong> Google, Microsoft, and OpenAI are reevaluating their dependencies on Scale, shifting significant attention toward neutral providers such as Labelbox and Snorkel AI, which have already reported 2&#8211;3&#215; spikes in demand post-announcement.</p></li><li><p><strong>Potential Countermoves:</strong></p><ul><li><p><strong>Microsoft/OpenAI</strong> may accelerate internal labeling capabilities or acquire independent data firms.</p></li><li><p><strong>Google</strong> could expand its AI Data Cloud initiative, positioning itself as an internal data services provider for Gemini.</p></li><li><p><strong>Amazon/Anthropic</strong> may further their partnership to secure exclusive evaluation capacities, paralleling the Meta&#8211;Scale model.</p></li><li><p>All major AI players are likely to diversify their data sourcing strategies to mitigate single-provider risks.</p></li></ul></li><li><p><strong>Vendor Consolidation:</strong> Heightened interest in specialized data infrastructure fuels significant M&amp;A activity, with AI infrastructure-related deals up notably year-over-year.</p></li></ul><h3>Investment Opportunities</h3><p>The evolving AI infrastructure landscape presents compelling investment avenues:</p><ul><li><p><strong>Neutral Data Providers:</strong> Independent providers offering cross-platform data annotation and evaluation will benefit significantly from increased enterprise demand.</p></li><li><p><strong>Verticalized AI SaaS:</strong> Specialized platforms targeting high-value sectors such as defense, healthcare, and finance are poised for robust growth. Deloitte expects over 30% annual growth in the vertical AI SaaS market through 2030.</p></li><li><p><strong>AI Governance and Evaluation Tools:</strong> Platforms addressing regulatory frameworks such as the EU AI Act and the U.S. NIST AI Risk Management Framework represent high-growth opportunities.</p></li></ul><h3>The Next Multi-Billion Acquisition?</h3><p>Expect the next wave of substantial M&amp;A activity in the neutral data and evaluation segment. Major tech companies&#8212;including Microsoft, Amazon, Google, Salesforce, and Apple&#8212;may look to strengthen their capabilities by acquiring specialized AI infrastructure firms. Potential acquisition targets in this space include <strong>Snorkel AI, Labelbox, and Scale&#8217;s remaining independent competitors</strong>. The common thread: Leading companies increasingly view robust, data-centric platforms and evaluation tools as essential competitive assets.</p><p>Never a dull moment in the AI infra space.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading LakeD-AI Unbundled! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[AI Breakthroughs 2025: The Mid-Year Report]]></title><description><![CDATA[Critical AI launches and what they mean for the future of work, tools, and platforms]]></description><link>https://lakedai.substack.com/p/ai-breakthroughs-2025-the-mid-year</link><guid isPermaLink="false">https://lakedai.substack.com/p/ai-breakthroughs-2025-the-mid-year</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Mon, 26 May 2025 21:35:19 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!c_gx!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F68315f7c-44b0-4361-8c43-acf8684fc427_1024x1024.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!c_gx!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F68315f7c-44b0-4361-8c43-acf8684fc427_1024x1024.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!c_gx!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F68315f7c-44b0-4361-8c43-acf8684fc427_1024x1024.png 424w, /__u/substackcdn.com/image/fetch/$s_!c_gx!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F68315f7c-44b0-4361-8c43-acf8684fc427_1024x1024.png 848w, /__u/substackcdn.com/image/fetch/$s_!c_gx!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F68315f7c-44b0-4361-8c43-acf8684fc427_1024x1024.png 1272w, /__u/substackcdn.com/image/fetch/$s_!c_gx!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F68315f7c-44b0-4361-8c43-acf8684fc427_1024x1024.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!c_gx!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F68315f7c-44b0-4361-8c43-acf8684fc427_1024x1024.png" width="1024" height="1024" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/68315f7c-44b0-4361-8c43-acf8684fc427_1024x1024.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:1024,&quot;width&quot;:1024,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:null,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:null,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:null,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!c_gx!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F68315f7c-44b0-4361-8c43-acf8684fc427_1024x1024.png 424w, /__u/substackcdn.com/image/fetch/$s_!c_gx!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F68315f7c-44b0-4361-8c43-acf8684fc427_1024x1024.png 848w, /__u/substackcdn.com/image/fetch/$s_!c_gx!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F68315f7c-44b0-4361-8c43-acf8684fc427_1024x1024.png 1272w, /__u/substackcdn.com/image/fetch/$s_!c_gx!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F68315f7c-44b0-4361-8c43-acf8684fc427_1024x1024.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>2025 is shaping up as a pivotal year in AI. From groundbreaking model enhancements to new capabilities reshaping enterprise workflows. here&#8217;s an overview of the most significant AI releases and their impacts so far. </p><h2>Key Themes to Watch in 2025</h2><ul><li><p><strong>Integration &amp; Multimodal AI</strong>: Bridging gaps between text, audio, visual, and coding workflows.</p></li><li><p><strong>Enterprise AI Autonomy</strong>: Increasing AI&#8217;s role in direct business process automation.</p></li><li><p><strong>Local &amp; Privacy-Focused AI</strong>: Shifting towards edge-computing and privacy-centric AI solutions.</p></li></ul><h2>1. <strong>OpenAI Advances with GPT-4.1 and o-series</strong> <em>(Released: April 16 &amp; May 14)</em></h2><p>OpenAI&#8217;s latest launches&#8212;the o-series and GPT-4.1&#8212;deliver unprecedented multimodal capabilities, seamless integration of web, vision, and Python tools, and significantly lower costs. Developers enjoy easier access to complex AI workflows, prompting faster innovation across multiple sectors.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading Lake&#8217;s Substack! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p><strong>GPT-4.1 vs GPT-4-turbo Performance</strong> A side-by-side bar chart highlights GPT-4.1&#8217;s superior latency, lower cost, and improved tool orchestration.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!Z-nP!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd87cf4a3-aea8-46cb-bca6-ed5b5497266c_594x384.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!Z-nP!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd87cf4a3-aea8-46cb-bca6-ed5b5497266c_594x384.png 424w, /__u/substackcdn.com/image/fetch/$s_!Z-nP!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd87cf4a3-aea8-46cb-bca6-ed5b5497266c_594x384.png 848w, /__u/substackcdn.com/image/fetch/$s_!Z-nP!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd87cf4a3-aea8-46cb-bca6-ed5b5497266c_594x384.png 1272w, /__u/substackcdn.com/image/fetch/$s_!Z-nP!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd87cf4a3-aea8-46cb-bca6-ed5b5497266c_594x384.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!Z-nP!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd87cf4a3-aea8-46cb-bca6-ed5b5497266c_594x384.png" width="594" height="384" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/d87cf4a3-aea8-46cb-bca6-ed5b5497266c_594x384.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:384,&quot;width&quot;:594,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:41645,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/164507600?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd87cf4a3-aea8-46cb-bca6-ed5b5497266c_594x384.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!Z-nP!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd87cf4a3-aea8-46cb-bca6-ed5b5497266c_594x384.png 424w, /__u/substackcdn.com/image/fetch/$s_!Z-nP!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd87cf4a3-aea8-46cb-bca6-ed5b5497266c_594x384.png 848w, /__u/substackcdn.com/image/fetch/$s_!Z-nP!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd87cf4a3-aea8-46cb-bca6-ed5b5497266c_594x384.png 1272w, /__u/substackcdn.com/image/fetch/$s_!Z-nP!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd87cf4a3-aea8-46cb-bca6-ed5b5497266c_594x384.png 1456w" sizes="100vw"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p></p><h2>2. <strong>Anthropic's Claude 4 Pushes Deep Work Boundaries</strong> <em>(Released: May 22)</em></h2><p>Anthropic introduces extended thinking modes tailored for complex tasks, positioning itself as a key player in advanced coding and data analysis. It enables significantly longer coding sessions and sophisticated agent-like capabilities.</p><p><strong>Task Depth Handling - Claude 4 vs Claude 3</strong> A line graph shows Claude 4&#8217;s stronger performance across more complex, multi-step tasks.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!ej5f!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6a3bca43-789b-4309-9a7b-ad86b53abc5e_594x376.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!ej5f!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6a3bca43-789b-4309-9a7b-ad86b53abc5e_594x376.png 424w, /__u/substackcdn.com/image/fetch/$s_!ej5f!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6a3bca43-789b-4309-9a7b-ad86b53abc5e_594x376.png 848w, /__u/substackcdn.com/image/fetch/$s_!ej5f!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6a3bca43-789b-4309-9a7b-ad86b53abc5e_594x376.png 1272w, /__u/substackcdn.com/image/fetch/$s_!ej5f!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6a3bca43-789b-4309-9a7b-ad86b53abc5e_594x376.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!ej5f!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6a3bca43-789b-4309-9a7b-ad86b53abc5e_594x376.png" width="594" height="376" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/6a3bca43-789b-4309-9a7b-ad86b53abc5e_594x376.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:376,&quot;width&quot;:594,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:59715,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/164507600?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6a3bca43-789b-4309-9a7b-ad86b53abc5e_594x376.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!ej5f!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6a3bca43-789b-4309-9a7b-ad86b53abc5e_594x376.png 424w, /__u/substackcdn.com/image/fetch/$s_!ej5f!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6a3bca43-789b-4309-9a7b-ad86b53abc5e_594x376.png 848w, /__u/substackcdn.com/image/fetch/$s_!ej5f!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6a3bca43-789b-4309-9a7b-ad86b53abc5e_594x376.png 1272w, /__u/substackcdn.com/image/fetch/$s_!ej5f!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6a3bca43-789b-4309-9a7b-ad86b53abc5e_594x376.png 1456w" sizes="100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p></p><h2>3. <strong>Google DeepMind&#8217;s Gemini 2.5 Expands Capabilities</strong> <em>(Released: May 21)</em></h2><p>Gemini adds native audio interactions, fine-grained budgeting for compute-intensive tasks, and automation for business processes. It enhanced flexibility for enterprises to control AI costs and integrate multimodal tasks. Gemini&#8217;s full-stack supports audio, RPA, and budgeted reasoning.</p><h2>4. <strong>Meta&#8217;s Llama 4 Keeps Open Source Thriving</strong> <em>(Released: April 5)</em></h2><p>Meta&#8217;s Llama 4 now supports multimodal inputs (text, image, video) under an open-source license, bolstering innovation within the developer community. It provides greater access reduces vendor lock-in risks, driving rapid industry-wide innovation.</p><p><strong>Llama OSS Growth (2023&#8211;2025)</strong> A dual-line chart shows dramatic growth in GitHub stars and user community over time.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!1fBM!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8f8ce1a0-418d-402a-990a-48cc74aaa902_594x368.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!1fBM!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8f8ce1a0-418d-402a-990a-48cc74aaa902_594x368.png 424w, /__u/substackcdn.com/image/fetch/$s_!1fBM!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8f8ce1a0-418d-402a-990a-48cc74aaa902_594x368.png 848w, /__u/substackcdn.com/image/fetch/$s_!1fBM!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8f8ce1a0-418d-402a-990a-48cc74aaa902_594x368.png 1272w, /__u/substackcdn.com/image/fetch/$s_!1fBM!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8f8ce1a0-418d-402a-990a-48cc74aaa902_594x368.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!1fBM!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8f8ce1a0-418d-402a-990a-48cc74aaa902_594x368.png" width="594" height="368" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/8f8ce1a0-418d-402a-990a-48cc74aaa902_594x368.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:368,&quot;width&quot;:594,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:54460,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/164507600?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8f8ce1a0-418d-402a-990a-48cc74aaa902_594x368.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!1fBM!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8f8ce1a0-418d-402a-990a-48cc74aaa902_594x368.png 424w, /__u/substackcdn.com/image/fetch/$s_!1fBM!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8f8ce1a0-418d-402a-990a-48cc74aaa902_594x368.png 848w, /__u/substackcdn.com/image/fetch/$s_!1fBM!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8f8ce1a0-418d-402a-990a-48cc74aaa902_594x368.png 1272w, /__u/substackcdn.com/image/fetch/$s_!1fBM!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8f8ce1a0-418d-402a-990a-48cc74aaa902_594x368.png 1456w" sizes="100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p></p><h2>5. <strong>Mistral Builds European AI Momentum</strong> <em>(Released: January 28 &amp; May 2025)</em></h2><p>Mistral introduces GDPR-aligned multimodal models and optimized code generation tools, establishing itself as a critical third AI pole. It strengthens European enterprise options and compliance-friendly innovation.</p><h2>6. <strong>Runway Gen-4 Revolutionizes Video Creation</strong> <em>(Released: March 31)</em></h2><p>Runway&#8217;s Gen-4 dramatically simplifies video generation workflows, offering filmmakers unprecedented consistency and creativity. It streamlines creative production, slashing timelines from days to minutes.</p><h2>7. <strong>NVIDIA&#8217;s RTX 50 Transforms Consumer AI</strong> <em>(Released: January 6)</em></h2><p>NVIDIA&#8217;s RTX 50 GPUs bring powerful local inference capabilities to consumers, accelerating the democratization of AI. It enables widespread adoption of sophisticated AI tasks on consumer hardware.</p><p><strong>RTX 50 vs RTX 40 Performance Gains</strong> Bar graphs show relative speed improvements in inference, generation, and local hosting.</p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!8ZTD!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F05b6b932-a2b2-49d8-9643-1e2110654119_598x398.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!8ZTD!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F05b6b932-a2b2-49d8-9643-1e2110654119_598x398.png 424w, /__u/substackcdn.com/image/fetch/$s_!8ZTD!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F05b6b932-a2b2-49d8-9643-1e2110654119_598x398.png 848w, /__u/substackcdn.com/image/fetch/$s_!8ZTD!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F05b6b932-a2b2-49d8-9643-1e2110654119_598x398.png 1272w, /__u/substackcdn.com/image/fetch/$s_!8ZTD!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F05b6b932-a2b2-49d8-9643-1e2110654119_598x398.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!8ZTD!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F05b6b932-a2b2-49d8-9643-1e2110654119_598x398.png" width="598" height="398" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/05b6b932-a2b2-49d8-9643-1e2110654119_598x398.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:398,&quot;width&quot;:598,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:47639,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:true,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/164507600?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F05b6b932-a2b2-49d8-9643-1e2110654119_598x398.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!8ZTD!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F05b6b932-a2b2-49d8-9643-1e2110654119_598x398.png 424w, /__u/substackcdn.com/image/fetch/$s_!8ZTD!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F05b6b932-a2b2-49d8-9643-1e2110654119_598x398.png 848w, /__u/substackcdn.com/image/fetch/$s_!8ZTD!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F05b6b932-a2b2-49d8-9643-1e2110654119_598x398.png 1272w, /__u/substackcdn.com/image/fetch/$s_!8ZTD!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F05b6b932-a2b2-49d8-9643-1e2110654119_598x398.png 1456w" sizes="100vw" loading="lazy"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><h2>8. <strong>Microsoft Copilot Becomes Enterprise Essential</strong> <em>(Released: April-May 2025)</em></h2><p>Microsoft Copilot integrates deeper into enterprise environments, offering Python in Excel, agent workflows, and governance features. Copilot solidifies its role as a central productivity and innovation platform.</p><p><strong>Graphic: Copilot Enterprise Adoption (2025)</strong> A time-series plot shows Copilot&#8217;s steep adoption growth curve from January to May.</p><h2>9. <strong>Adobe Firefly Scales Enterprise Content Production</strong> <em>(Released: January 13 &amp; March 2025)</em></h2><p>Adobe&#8217;s Firefly brings generative AI into large-scale content pipelines, significantly accelerating marketing operations. It cuts campaign execution times and enhances content personalization at scale - reduce average campaign production time from 14 to 3 days.</p><h2>10. <strong>Apple Prepares for AI Expansion</strong> <em>(Announced: May 20; Launch Expected: WWDC June 2025)</em></h2><p>Apple teases its upcoming &#8220;Apple Intelligence,&#8221; promising privacy-focused, on-device AI capabilities that could reshape app development and privacy-sensitive industries. It could trigger a significant shift toward mobile-focused, privacy-preserving AI apps.</p><p>Stay tuned as we track these themes and their unfolding impacts throughout 2025. <em>Join our <a href="/__u/lakedai.substack.com/">newsletter </a>to stay ahead of AI trends and insights.</em></p><p><strong>Sources:</strong></p><ul><li><p>OpenAI, Anthropic, DeepMind, Meta, Mistral, Runway, NVIDIA, Microsoft, Adobe, Apple official announcements and developer documentation</p></li><li><p>Visuals generated from benchmarked or representative metrics for educational and editorial use</p></li></ul><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading Lake&#8217;s Substack! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[The New Relationship Triangle With AI]]></title><description><![CDATA[The new relationship triangle - human with human, AI with human, AI with AI]]></description><link>https://lakedai.substack.com/p/the-new-relationship-triangle-with</link><guid isPermaLink="false">https://lakedai.substack.com/p/the-new-relationship-triangle-with</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Wed, 21 May 2025 16:00:44 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!Q6re!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<h3></h3><p>The future of relationships isn&#8217;t just about humans getting along&#8212;it's also about how we'll partner with artificial intelligence (AI), and how AI will interact among themselves. But are we ready for these new types of connections?</p><div><hr></div><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!Q6re!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!Q6re!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png 424w, /__u/substackcdn.com/image/fetch/$s_!Q6re!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png 848w, /__u/substackcdn.com/image/fetch/$s_!Q6re!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png 1272w, /__u/substackcdn.com/image/fetch/$s_!Q6re!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!Q6re!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png" width="1456" height="971" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:971,&quot;width&quot;:1456,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:2132677,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/164061083?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!Q6re!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png 424w, /__u/substackcdn.com/image/fetch/$s_!Q6re!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png 848w, /__u/substackcdn.com/image/fetch/$s_!Q6re!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png 1272w, /__u/substackcdn.com/image/fetch/$s_!Q6re!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F01b740c1-e58b-4372-a1c1-c26c1e160f84_1536x1024.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><ul><li><p><strong>Human with Human:</strong> Empathy, creativity, and storytelling become our most valuable skills.</p></li><li><p><strong>AI with Human:</strong> AI may move from simple assistant tasks to mentoring roles, training humans directly. Think: "vibe coding" may not only replace junior developers for repetitive tasks but also help train them into senior developers&#8212;without traditional on-the-job training.</p></li><li><p><strong>AI with AI:</strong> Automated systems will manage and oversee each other, with AI "police" agents ensuring safety and fairness.</p></li></ul><div><hr></div><h3>1 | Human &#8596; Human: Why Empathy is the New Currency</h3><p>Information used to be king&#8212;but now anyone can ask an AI for quick answers. The real value is no longer in knowing facts, but in asking great questions, understanding others deeply, and telling compelling stories. This puts empathy, teamwork, and emotional intelligence at the top of the skills ladder.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading Lake&#8217;s Substack! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p><strong>Yesterday vs. Tomorrow:</strong></p><ul><li><p><strong>Knowing answers &#8594; Asking better questions</strong></p></li><li><p><strong>Solo heroics &#8594; Collaborative storytelling</strong></p></li><li><p><strong>Explaining data &#8594; Providing meaningful context</strong></p></li></ul><p>Teams that thrive will prioritize emotional intelligence and know how to leverage AI without losing their human touch.</p><div><hr></div><h3>2 | AI &#8596; Human: Your AI Assistant May Become Your Mentor</h3><p>Think about learning guitar with YouTube videos. Now imagine if that video could pause, gently correct your finger placement, and then show you exactly how to improve. That's what's happening with AI&#8212;moving beyond just performing tasks, towards actively training us.</p><p><strong>Example: Coding and the "Vibe" Revolution</strong><br>With "vibe coding," AI handles repetitive programming tasks, allowing developers to focus on high-level thinking&#8212;even early in their careers. Instead of writing basic functions, junior developers can learn from AI-generated examples and quickly grow into strategic thinkers and problem-solvers.<br><br>The CEO of Duolingo predicts a future where AI handles direct teaching, allowing human educators more time to mentor students personally. Already, universities are experimenting with AI tutors that help students learn through dialogue rather than lectures.</p><p><strong>Future Ways of Working:</strong></p><ul><li><p>Each employee could have a personalized AI assistant logging their decisions and highlighting areas for growth. </p></li><li><p>Companies may start measuring success by how quickly people gain new skills, not just their immediate productivity.</p></li></ul><div><hr></div><h3>3 | AI &#8596; AI: Supervising Agents and Digital Police</h3><p>Just like humans, AI systems need oversight to prevent mistakes or misuse. In the future, autonomous agents may run entire operations&#8212;from self-driving trucks to managing supply chains. But who watches them?</p><p>In many cases, humans simply can&#8217;t match the speed at which AI operates, such as in AI-generated cyberattacks. This raises the need for new safeguards. Enter the AI "police"&#8212;specialized digital watchdogs designed to ensure that other AI agents follow ethical guidelines, comply with regulations, and behave safely and responsibly.</p><p><strong>Imagine:</strong></p><ul><li><p>AI coordinating traffic but also needing supervision to avoid dangerous shortcuts.</p></li><li><p>AI managing customer service, but needing oversight to prevent miscommunication or biases.</p></li><li><p>AI fleets managing deliveries and logistics, monitored by supervisory agents to maintain efficiency and compliance.</p><p></p></li></ul><p>As AI interactions grow increasingly complex, another question emerges: Can humans not only trust AI, but also retain control&#8212;and reverse its actions when needed? Can we truly thrive in this new relationship triangle? How should we redesign education to prepare both the next generation&#8212;and ourselves&#8212;for a world where human-AI dynamics shape every aspect of life?</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading Lake&#8217;s Substack! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Is AI Another Internet Bubble?]]></title><description><![CDATA[Over the past few decades, technological trends have swung investors between bursts of optimism and sobering corrections.]]></description><link>https://lakedai.substack.com/p/is-ai-another-internet-bubble</link><guid isPermaLink="false">https://lakedai.substack.com/p/is-ai-another-internet-bubble</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Wed, 14 May 2025 23:13:56 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!POUU!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!POUU!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!POUU!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png 424w, /__u/substackcdn.com/image/fetch/$s_!POUU!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png 848w, /__u/substackcdn.com/image/fetch/$s_!POUU!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png 1272w, /__u/substackcdn.com/image/fetch/$s_!POUU!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!POUU!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png" width="1024" height="1024" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:1024,&quot;width&quot;:1024,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:1327204,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/163592604?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!POUU!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png 424w, /__u/substackcdn.com/image/fetch/$s_!POUU!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png 848w, /__u/substackcdn.com/image/fetch/$s_!POUU!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png 1272w, /__u/substackcdn.com/image/fetch/$s_!POUU!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2361dd5d-396f-4ba9-882e-52d6313e7a71_1024x1024.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Over the past few decades, technological trends have swung investors between bursts of optimism and sobering corrections. The dot-com era was characterized by exuberant investments with unproven business models. Today, how do we distinguish <strong>genuine innovation from mere hype in AI</strong>?</p><p>Having built through internet bubbles and financial crises, I&#8217;ve observed a distinct difference this time. The internet bubble was characterized by rapid overinvestment in infrastructure, with the internet&#8217;s potential often exaggerated beyond its immediate practical uses. In contrast, AI has already demonstrated its <strong>real-world impact </strong>beyond ChatGPT. If you look everywhere, there is hardly any industry not facing complete disruption by AI.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading Lake&#8217;s Substack! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div><p><strong>Smart AI investors are targeting foundational and essential building models, and scalable projects that integrate AI with existing systems.</strong> They understand that building an AI-powered future requires a balanced approach &#8212; one that supports both the development of cutting-edge algorithms and the practical implementation of these technologies. This includes investments in data storage, computing power, and cybersecurity measures that are essential for AI to function reliably and securely.</p><p>Additionally, AI is grounded in decades of rigorous scientific research. The roots of AI stretch back to the mid-20th century, with breakthroughs in areas such as statistical modeling, machine learning, and neural networks. It leverages powerful algorithms and large datasets, enabling systems to learn, adapt, and make predictions with unprecedented accuracy. This depth of academic research and continuous innovation in the field provides a robust backbone, ensuring that AI&#8217;s applications are not only technically sound but also practically relevant.</p><p>Let&#8217;s examine some examples of how AI boosts efficiency, cuts costs, and unlocks new revenue streams. The global economic impact of <strong>AI could add $15.7 trillion to global GDP by 2030</strong>&#8203;, with North America projected to see a ~14% GDP boost from AI innovations&#8203;. In practical terms, AI spending and adoption are surging in key industries. The table below highlights how major sectors are leveraging AI, with market growth and concrete benefits:</p><p><strong>AI Adoption Across Industries</strong></p><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!PRrw!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F386f33bc-ceaa-4033-88b7-cd1896f3a786_1024x768.webp" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!PRrw!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F386f33bc-ceaa-4033-88b7-cd1896f3a786_1024x768.webp 424w, /__u/substackcdn.com/image/fetch/$s_!PRrw!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F386f33bc-ceaa-4033-88b7-cd1896f3a786_1024x768.webp 848w, /__u/substackcdn.com/image/fetch/$s_!PRrw!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F386f33bc-ceaa-4033-88b7-cd1896f3a786_1024x768.webp 1272w, /__u/substackcdn.com/image/fetch/$s_!PRrw!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F386f33bc-ceaa-4033-88b7-cd1896f3a786_1024x768.webp 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!PRrw!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F386f33bc-ceaa-4033-88b7-cd1896f3a786_1024x768.webp" width="1024" height="768" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/386f33bc-ceaa-4033-88b7-cd1896f3a786_1024x768.webp&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:768,&quot;width&quot;:1024,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:51490,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/webp&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:false,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/163592604?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F386f33bc-ceaa-4033-88b7-cd1896f3a786_1024x768.webp&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!PRrw!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F386f33bc-ceaa-4033-88b7-cd1896f3a786_1024x768.webp 424w, /__u/substackcdn.com/image/fetch/$s_!PRrw!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F386f33bc-ceaa-4033-88b7-cd1896f3a786_1024x768.webp 848w, /__u/substackcdn.com/image/fetch/$s_!PRrw!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F386f33bc-ceaa-4033-88b7-cd1896f3a786_1024x768.webp 1272w, /__u/substackcdn.com/image/fetch/$s_!PRrw!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F386f33bc-ceaa-4033-88b7-cd1896f3a786_1024x768.webp 1456w" sizes="100vw"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Each industry&#8217;s experience reinforces that AI is delivering real value today. Below we dive deeper into how AI is transforming these sectors, with verified examples and projections for the coming years.</p><p><strong>Healthcare: Saving Lives and Costs with AI</strong></p><p>AI is becoming healthcare&#8217;s &#8220;new nervous system,&#8221; enhancing clinical outcomes and reducing costs. The healthcare AI market has experienced rapid expansion, growing from an estimated $600 million in 2014 to $6.6 billion by 2021 &#8212; approximately 40% annual growth. Projections suggest this figure could grow more than tenfold in the next five years<strong>.</strong> By 2030, global AI health spending is expected to reach between $150 billion and <strong>$190 billion</strong>.</p><p>Massive cost savings are also on the horizon. AI applications in healthcare &#8212; including robotic surgeries, virtual nursing assistants, and predictive diagnostics &#8212; <strong>could save the U.S. healthcare system up to $150 billion annually by 2026.</strong> AI can help identify high-risk patients earlier or automate routine administrative tasks, preventing costly complications and reducing waste.</p><p>Real-world cases already highlight AI&#8217;s life-saving impact. At Johns Hopkins Hospital, an AI-powered sepsis detection system analyzes patient data and alerts doctors to early signs of infection. This implementation has<strong> reduced sepsis-related mortality by 20%</strong>, as the AI catches symptoms hours earlier than traditional methods. Similarly, at the University of Rochester Medical Center, AI-enhanced ultrasound devices have <strong>increased ultrasound capture by 116%</strong>, leading to faster and more accurate diagnostics. These examples demonstrate that AI in healthcare is not theoretical &#8212; it&#8217;s improving care quality and achieving the &#8220;quadruple aim&#8221; (better outcomes, lower costs, improved patient and provider experience) today.</p><p>Looking ahead, AI will likely be even more deeply embedded in U.S. healthcare over the next 5&#8211;10 years. AI-driven diagnostics, drug discovery, and hospital operational optimization are expected to become standard practice. <strong>By 2030, an estimated 90% of U.S. hospitals may incorporate AI </strong>into medical imaging, risk prediction, and other critical application<strong>s</strong> thanks to its demonstrated efficiency and return on investment. The result will be a healthcare system that is more effective and cost-efficient &#8212; solidifying AI as a necessary tool for better patient outcomes rather than mere hype.</p><p><strong>Finance: AI Unlocking $1 Trillion in Banking Value</strong></p><p>The financial services industry was an early adopter of AI, and it is yielding significant returns. Banks, insurers, and investment firms in the U.S. are leveraging AI for <strong>fraud detection, algorithmic trading, customer service chatbots, and credit underwriting</strong>. This momentum is reflected in market projections, with the AI in the finance sector expected to grow from approximately $38 billion in 2024 to over<strong> $190 billion by 2030</strong> &#8212; a compound annual growth rate (CAGR) of roughly 32%. U.S. banks alone are forecast to invest tens of billions of dollars in AI solutions this decade to remain competitive and drive profitability.</p><p>Cost reduction<strong> </strong>is a major factor behind this investment. Analysts estimate that AI will contribute to over <strong>$1 trillion in savings</strong> for traditional financial institutions by 2030 &#8212; a reduction of approximately 22% in operating expenses across front-, middle-, and back-office operations. Specifically, AI-driven efficiency gains include:</p><blockquote><p>&#183; <strong>Front office:</strong> Estimated <strong>$490 billion in savings</strong> by 2030 from AI-powered customer service (e.g., virtual assistants like Bank of America&#8217;s Erica), workforce optimization, and smarter sales targeting.</p><p>&#183; <strong>Middle office: </strong>Around <strong>$350 billion in savings</strong> through AI-enhanced compliance, fraud detection, KYC/AML processes, and risk modeling.</p><p>&#183; <strong>Back office:</strong> About <strong>$200 billion in savings</strong> enabled by AI improvements in loan underwriting, claims processing, and other administrative functions.</p></blockquote><p>These are not just theoretical projections &#8212; tangible results are already being seen. JPMorgan Chase&#8217;s AI platform, COIN (Contract Intelligence), has automated commercial loan agreement reviews, reducing a process that previously required <strong>360,000 hours of manual work</strong> annually to just seconds. This AI has not only eliminated a substantial administrative burden but also improved accuracy by <strong>reducing errors</strong> in loan servicing. Likewise, Wells Fargo is utilizing AI to analyze customer data for personalized product recommendations, and Mastercard has implemented AI algorithms that significantly cut down false fraud declines &#8212; helping preserve millions in revenue.</p><p>Looking ahead, AI will continue reshaping the financial sector, enabling more personalized banking experiences, real-time risk management, and fully automated processes. By 2030, AI is expected to be a key driver behind most U.S. banks&#8217; decision-making, spanning credit approvals, investment strategies, and regulatory compliance. With AI projected to <strong>add $1.2 trillion in value to the financial industry by 2035</strong>, it is evident that AI is not merely a trend &#8212; it is a fundamental transformation of how finance operates.</p><p><strong>Manufacturing: Smart Factories and Productivity Gains</strong></p><p>In manufacturing, AI and machine learning are central to the &#8220;Industry 4.0&#8221; revolution, making production smarter, faster, and safer. The AI in the manufacturing market<strong> </strong>has expanded drastically, growing from just $1 billion in 2018 to an<strong> estimated $16 billion by 2025</strong>. As factories deploy AI for robotics, quality control, maintenance, and supply chain optimization, U.S. manufacturers are increasingly adopting these technologies to remain globally competitive.</p><p><strong>Efficiency and uptime improvements</strong> are among the most significant benefits AI offers. AI-driven <strong>predictive maintenance</strong>, for instance, analyzes sensor data to detect potential failures before they occur, allowing for optimized scheduling of repairs. Companies implementing predictive maintenance have reported up to <strong>70% fewer equipment breakdowns and a 25% reduction in maintenance costs</strong>. This directly minimizes costly downtime on factory floors. At a BMW auto plant, AI maintenance models that identified machine fault patterns helped save over 500 minutes of production line disruption annually &#8212; equating to avoiding delays in hundreds of vehicle productions. Implementing such AI-driven maintenance strategies not only cuts costs but also ensures manufacturers can fulfill orders on time, reducing operational stress.</p><p>AI is also transforming<strong> quality control.</strong> Computer vision-powered AI now inspects products for defects with greater speed and accuracy than human inspectors. A U.S. packaging company that implemented AI-powered inspections <strong>reduced inspection time by 50% </strong>and labor costs by 10%, demonstrating the tangible financial benefits of AI-enhanced quality control. Additionally, AI optimizes supply chain logistics &#8212; for instance, adjusting production schedules in real-time based on demand fluctuations &#8212; to reduce surplus inventory and minimize warehousing expenses.</p><p>Looking ahead, <strong>smart factories</strong> will likely become the industry standard. By 2030, AI-powered automation could <strong>increase global manufacturing productivity by 20&#8211;30%</strong>, generating trillions in additional economic output. In the U.S., AI co-bots (collaborative robots) on assembly lines, AI-based production scheduling to minimize idle machine time, and generative design AI for optimizing product materials and efficiency will likely be widespread. These advancements point to significant cost reductions and increased profit margins, underscoring AI as a critical operational investment. Manufacturers that integrate AI are already experiencing its benefits, paving the way for leaner, more resilient production lines.</p><p><strong>Retail: Personalization and Efficiency Driving Growth</strong></p><p>Retail &#8212; spanning both e-commerce and brick-and-mortar stores &#8212; is increasingly leveraging AI to understand customers, manage inventory, and optimize pricing. The AI in the retail market is experiencing rapid growth, projected to expand from approximately $11&#8211;13 billion in 2024 to between $40 billion and <strong>$55+ billion by 2030</strong>, with annual growth rates exceeding 30%. U.S. retail giants, as well as midsized chains, are adopting AI solutions for demand forecasting, supply chain optimization, customer analytics, and in-store automation. These investments are translating into measurable increases in sales and reductions in operational costs, making AI a fundamental component of modern retail strategy.</p><p>One of AI&#8217;s most visible contributions to retail is <strong>revenue growth through personalization</strong>. AI algorithms analyze shopper behaviors and preferences to provide highly tailored product recommendations, leading to increased conversions and larger purchases. A well-known example is <strong>Amazon&#8217;s AI-driven recommendation engine,</strong> which accounts for an estimated <strong>35% of the company&#8217;s total sales</strong> &#8212; amounting to tens of billions of dollars annually. Other retailers report similar success; AI-powered personalization engines in e-commerce have been shown to increase revenue per customer by 10&#8211;15% by suggesting relevant add-ons or predicting what shoppers are most likely to purchase next. <strong>AI-driven marketing and dynamic pricing</strong> further allow retailers to maximize revenue by reaching customers with the right offers at the right time.</p><p><strong>Operational efficiencies</strong> are another key benefit of AI in retail. A 2024 survey of retail and consumer packaged goods (CPG) firms found that <strong>69% of businesses using AI saw revenue growth</strong>, while <strong>72% reported lower operating costs</strong> &#8212; both crucial for industries operating with tight margins. Over half of surveyed retail executives noted that AI delivered more than a 15% improvement in both revenue and cost metrics. AI-powered demand forecasting is helping retailers optimize inventory, reducing capital tied up in overstock and preventing costly stockouts that impact sales. For example, grocery chains use AI to predict demand for perishable items, cutting food waste and saving millions of dollars. AI also enhances store operations by automating tasks such as monitoring shelf stocking needs with computer vision or analyzing checkout lines to optimize staffing, improving both efficiency and the customer experience.</p><p><strong>Leading retailers have already demonstrated AI&#8217;s return on investment (ROI).</strong> Walmart, for instance, employs AI-driven analytics to fine-tune its supply chain across 4,700+ U.S. stores, enhancing on-shelf product availability and reducing inventory turnaround time. Target&#8217;s AI-backed holiday demand forecasts have been so precise that they significantly reduced last-minute logistics expenses. Similarly, Sephora&#8217;s AI-powered chatbots engage customers online, improving conversions while easing call center workloads &#8212; resulting in both revenue gains and labor cost savings.</p><p>Looking ahead, <strong>AI is set to become deeply ingrained in retail operations</strong> within the next 5&#8211;10 years. Innovations such as autonomous checkouts, AI-assisted store layouts, and fully algorithmic pricing strategies will become standard. By 2030, most U.S. retailers &#8212; both large and small &#8212; are likely to rely on AI for daily decision-making, as early adopters are already outpacing competitors. Given the continued double-digit growth in AI investments within the retail sector, this presents a major opportunity to enhance efficiency and profitability in one of the world&#8217;s most competitive industries.</p><p><strong>Other Key Sectors: Energy, Logistics, Agriculture, and More</strong></p><p>AI&#8217;s transformative impact extends across nearly every major industry. Here are a few additional sectors where AI is generating tangible benefits and is poised for significant growth in the coming years:</p><blockquote><p>&#183; <strong>Energy &amp; Utilities: </strong>AI is optimizing energy usage and power distribution. Google&#8217;s DeepMind AI, for example,<strong> reduced energy consumption for cooling Google&#8217;s data centers by up to 40%</strong>, which translated into an overall power usage reduction of about 15% at those facilities. Scaling similar AI-driven energy optimizations to other industrial HVAC systems and data centers could save hundreds of millions of dollars in electricity costs. Additionally, power companies are turning to AI for demand prediction and grid optimization to prevent blackouts and reduce energy waste. By 2030, &#8220;smart grid&#8221; AI systems are expected to enhance efficiency in renewable energy integration, improving solar and wind energy management through better weather forecasting and supply balancing.</p><p>&#183; <strong>Transportation &amp; Logistics:</strong> AI is making transportation systems more efficient and cost-effective. AI-driven routing and autonomous vehicle technologies are helping reduce fuel consumption and labor costs in logistics. A prime example is UPS&#8217;s ORION routing AI, which analyzes millions of package delivery routes in real-time. At full deployment, <strong>ORION helps UPS save an estimated 100 million driving miles and 10 million gallons of fuel annually</strong> &#8212; equating to<strong> cost savings between $300 million and $400 million each year</strong>. AI-powered fleet management is also optimizing trucking routes to reduce empty miles, further lowering fuel consumption and operating costs. In the coming years,<strong> autonomous vehicles</strong> are expected to play a more prominent role, with AI-guided self-driving trucks potentially reducing logistics costs by up to 30% per mile while improving roadway safety and reducing accident rates.</p><p>&#183; <strong>Agriculture:</strong> AI-powered <strong>precision agriculture</strong> is helping farmers maximize crop yields while reducing input costs. Companies such as <strong>Arable and CropX</strong> have reported <strong>up to 30% increases in crop productivity</strong> due to AI-driven irrigation and fertilization optimization. AI-based weather analysis enables farmers to adjust water and nutrient use efficiently, preventing overuse and minimizing waste. Additionally, AI-powered visual recognition systems can detect weeds and pests early, enabling targeted herbicide spraying, which has been shown to <strong>reduce herbicide usage by up to 90%</strong> in some trials. With the global demand for food increasing, AI&#8217;s role in farming is growing rapidly. By 2030, AI adoption in agriculture is expected to expand significantly, with automated tools for <strong>crop disease prediction, smart tractors, and AI-driven supply chain analytics </strong>becoming widespread in U.S. farming operations.</p><p>&#183; <strong>Other Industries: </strong>AI-driven innovations are impacting <strong>education</strong>, marketing, government, and entertainment. AI tutors and personalized learning platforms are improving student outcomes while reducing the cost of remedial education. In <strong>marketing and advertising</strong>, AI is optimizing ad placements and content relevance, increasing return on investment for marketing spend. <strong>Government </strong>and <strong>defense </strong>agencies are using AI for predictive maintenance of critical infrastructure and intelligent data analysis &#8212; such as optimizing public transit schedules or improving emergency response strategies. In <strong>media and entertainment</strong>, AI is enhancing content recommendation algorithms and even assisting in content creation, ensuring that media companies retain subscribers with highly curated content experiences. Across all these industries, AI is consistently delivering efficiency gains, cost reductions, and a better user experience.</p></blockquote><p>The next 5&#8211;10 years are poised to bring even greater AI-driven transformation. Analysts forecast robust growth in AI investment, with global AI revenues (software, hardware, and services) climbing sharply. By 2030, many estimates put the worldwide AI market in the hundreds of billions of dollars, and AI is expected to contribute <strong>over $1.5 trillion annually in the U.S. economy alone</strong> through efficiency and innovation gains. Sectors that effectively leverage AI will likely outpace those that lag, in much the same way that internet-enabled businesses outpaced those who resisted going online.</p><p><strong>Sources</strong></p><blockquote><p>&#183; <strong>Accenture (2020) </strong>&#8212; <em>AI: Healthcare&#8217;s New Nervous System </em>, projected $150B savings in US healthcare by 2026; AI health market growth figures. (<em>accenture.com</em>)</p><p>&#183; <strong>AI Business / Global Market Insights (2019) </strong>&#8212; <em>AI in Manufacturing to Grow to $16B by 2025 </em>, manufacturing AI market growth and benefits in efficiency and cost savings. (<em>aibusiness.com</em>)</p><p>&#183; <strong>BizTech Magazine (2025) </strong>&#8212; <em>Predictive Maintenance in Manufacturing </em>, BMW plant case study showing AI saved 500+ minutes of downtime. (<em>biztechmagazine.com</em>)</p><p>&#183; <strong>Deloitte Analytics (2017) &#8212; </strong><em>Predictive Maintenance Report </em>, AI reducing equipment breakdowns by 70% and maintenance costs by 25%. (<em>beekeeper.io</em>)</p><p>&#183; <strong>Evdelo / McKinsey (2022) </strong>&#8212; <em>Amazon Personalization Case </em>, AI-driven recommendation engine contributing 35% of Amazon&#8217;s total sales. (<em>evdelo.com</em>)</p><p>&#183; <strong>FindLaw (2019) </strong>&#8212; <em>Machine Learning Saves JPMorgan 360,000 Hours of Legal Work </em>, case study of JPMorgan COIN automating legal document review. (<em>findlaw.com</em>)</p><p>&#183; <strong>GitHub (2023) </strong>&#8212; <em>AI-powered Developer Productivity </em>, AI coding tools accelerating development by 55%; Duolingo&#8217;s 25% velocity increase case study. (<em>github.blog</em>)</p><p>&#183; <strong>Grand View Research (2024) </strong>&#8212; <em>AI in Retail Market Report </em>, AI in retail projected to surpass $40B by 2030 (23% CAGR). (<em>grandviewresearch.com</em>)</p><p>&#183; <strong>INFORMS Journal (2016) </strong>&#8212; <em>UPS ORION Case Study </em>, AI improving UPS logistics, saving $300&#8211;$400M and 10M gallons of fuel annually. (<em>informs.org</em>)</p><p>&#183; <strong>Johns Hopkins University (2022) </strong>&#8212; Clinical study on AI sepsis detection, AI reducing sepsis mortality by 20%. (<em>hub.jhu.edu</em>)</p><p>&#183; <strong>Keymakr (2023) &#8212; </strong><em>AI in Agriculture </em>, AI-powered precision farming increasing crop yields by up to 30% while reducing herbicide usage by 90%. (<em>keymakr.com</em>)</p><p>&#183; <strong>NVIDIA/NYU Stern (2024) </strong>&#8212; <em>State of AI in Retail &amp; CPG </em>, survey results showing 69% of retailers saw revenue growth and 72% experienced cost reductions due to AI. (<em>images.nvidia.com</em>)</p><p>&#183; <strong>The Financial Brand / Autonomous (2018) </strong>&#8212; <em>AI and Banking&#8217;s $1 Trillion Opportunity </em>, cost savings breakdown in banking through 2030. (<em>thefinancialbrand.com</em>)</p><p>&#183; <strong>WIRED (2016) &#8212; </strong><em>DeepMind AI Cuts Data Center Cooling by 40% </em>, Google&#8217;s AI reducing data center energy usage. (<em>wired.com</em>)</p></blockquote><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading Lake&#8217;s Substack! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Chaos Theory: Why VC Investment May Thrive in Uncertainty]]></title><description><![CDATA[Why invest in venture capital&#8212;an illiquid asset class&#8212;when stocks are declining, tariffs are escalating trade tensions, and cash feels like the safest bet?]]></description><link>https://lakedai.substack.com/p/chaos-theory-why-vc-investment-may</link><guid isPermaLink="false">https://lakedai.substack.com/p/chaos-theory-why-vc-investment-may</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Mon, 05 May 2025 16:01:55 GMT</pubDate><enclosure url="https://substack-post-media.s3.amazonaws.com/public/images/824d2582-31ca-42eb-bdd0-777dfbf1ba7f_500x621.jpeg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!iVHf!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe84b3ac3-2f5f-426d-83b1-53401d7ec0a2_500x621.jpeg" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!iVHf!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe84b3ac3-2f5f-426d-83b1-53401d7ec0a2_500x621.jpeg 424w, /__u/substackcdn.com/image/fetch/$s_!iVHf!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe84b3ac3-2f5f-426d-83b1-53401d7ec0a2_500x621.jpeg 848w, /__u/substackcdn.com/image/fetch/$s_!iVHf!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe84b3ac3-2f5f-426d-83b1-53401d7ec0a2_500x621.jpeg 1272w, /__u/substackcdn.com/image/fetch/$s_!iVHf!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe84b3ac3-2f5f-426d-83b1-53401d7ec0a2_500x621.jpeg 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!iVHf!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe84b3ac3-2f5f-426d-83b1-53401d7ec0a2_500x621.jpeg" width="500" height="621" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/e84b3ac3-2f5f-426d-83b1-53401d7ec0a2_500x621.jpeg&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:false,&quot;imageSize&quot;:&quot;normal&quot;,&quot;height&quot;:621,&quot;width&quot;:500,&quot;resizeWidth&quot;:500,&quot;bytes&quot;:100805,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/jpeg&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/162864762?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fcb187b44-f531-4f8b-97a3-dd2055fe4bbf_500x621.jpeg&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:&quot;center&quot;,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!iVHf!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe84b3ac3-2f5f-426d-83b1-53401d7ec0a2_500x621.jpeg 424w, /__u/substackcdn.com/image/fetch/$s_!iVHf!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe84b3ac3-2f5f-426d-83b1-53401d7ec0a2_500x621.jpeg 848w, /__u/substackcdn.com/image/fetch/$s_!iVHf!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe84b3ac3-2f5f-426d-83b1-53401d7ec0a2_500x621.jpeg 1272w, /__u/substackcdn.com/image/fetch/$s_!iVHf!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe84b3ac3-2f5f-426d-83b1-53401d7ec0a2_500x621.jpeg 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p>Why invest in venture capital&#8212;an illiquid asset class&#8212;when stocks are declining, tariffs are escalating trade tensions, and cash feels like the safest bet? Interestingly, tariffs haven&#8217;t slowed AI software; if anything, they've accelerated it. When labor, manufacturing, and input costs rise, companies get squeezed. This pressure forces new thinking, making AI the obvious solution: automate what you can, optimize the rest, and find ways to grow without increasing headcount. AI adoption has shifted from optional to essential. Gartner forecasts that AI software revenue will reach $134.8 billion by 2025.</p><p>Fortune 500 companies have been quick to integrate AI to navigate economic uncertainties. For example, Walmart leveraged AI to optimize inventory management and supply chain logistics, reportedly saving millions annually by reducing excess stock and improving efficiency. Similarly, JPMorgan Chase has significantly expanded its AI-driven customer service and fraud detection systems, enhancing productivity and client experience while reducing costs by over $150 million annually.</p><p>History also shows that venture capital thrives during downturns. Cambridge Associates reports that vintages during economic crises (such as 2001 and 2008) have historically yielded returns above 20%. Notably, the 2008 vintage funds returned an average IRR of approximately 21.7%, significantly outperforming those launched in economically stable periods. Prominent companies such as Airbnb, Uber, and Slack emerged during these challenging economic climates, benefiting from the resilience and adaptability forged in adversity.</p><p>AI infrastructure and no-code tools are dramatically accelerating startup creation by significantly reducing cost and time (increasing supply). Platforms like Hugging Face, Loveable, and Cursor&#8217;s Vibe Coding exemplify this shift, empowering technical and non-technical founders alike to rapidly prototype and scale innovative products without heavy capital investment. Meanwhile, demand from venture capital is shrinking due to limited partners' reduced appetite for illiquid assets in the current market environment (decreasing demand). According to PitchBook data, VC fundraising dropped approximately 30% in Q4 2024 compared to the previous year, signaling a pullback from LPs.</p><p>Demand and supply are rebalancing, which may shift us back toward a buyer's market. This dynamic provides investors with significant leverage to access favorable valuations, greater equity stakes, and more founder-friendly terms. In essence, investors entering at this juncture are positioned to capture outsized returns as market conditions stabilize and growth accelerates.</p><p>It's an interesting time&#8212;when fundraising is difficult, it's often the best moment to invest and build long-lasting, successful companies. Historical patterns and current indicators suggest that the pressure-cooker environment we're experiencing may incubate some of tomorrow's market leaders. Could today's anxieties make the 2025 vintage one of the most rewarding in recent years? If history is a guide, venture capital investment today may indeed reap exceptional rewards tomorrow.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading Lake&#8217;s Substack! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[Cloud to Edge, The Next AI Gold Rush?]]></title><description><![CDATA[It's about bringing AI right to your phone, watch, or even your fridge. Welcome to the era of inference&#8212;where AI delivers meaningful experiences instantly.]]></description><link>https://lakedai.substack.com/p/cloud-to-edge-the-next-ai-gold-rush</link><guid isPermaLink="false">https://lakedai.substack.com/p/cloud-to-edge-the-next-ai-gold-rush</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Fri, 07 Mar 2025 16:59:58 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!f615!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!f615!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!f615!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png 424w, /__u/substackcdn.com/image/fetch/$s_!f615!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png 848w, /__u/substackcdn.com/image/fetch/$s_!f615!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png 1272w, /__u/substackcdn.com/image/fetch/$s_!f615!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!f615!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png" width="1024" height="1536" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:1536,&quot;width&quot;:1024,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:2747711,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/png&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/158568891?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!f615!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png 424w, /__u/substackcdn.com/image/fetch/$s_!f615!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png 848w, /__u/substackcdn.com/image/fetch/$s_!f615!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png 1272w, /__u/substackcdn.com/image/fetch/$s_!f615!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31d4ad78-86bf-4529-8755-03fba14690b4_1024x1536.png 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p></p><h2>From Big Models to Big Moments: AI&#8217;s New Chapter</h2><p>AI is turning a corner! Once all about huge data centers and giant models, now it's about bringing AI right to your phone, watch, or even your fridge. Welcome to the era of inference&#8212;where AI delivers meaningful experiences instantly.</p><p>The growth of inference is clear: the AI inference market is projected to rise sharply from $106 billion in 2025 to nearly $255 billion by 2030. This trend highlights that making AI effective in real-world use is now more crucial (and costly) than simply training the models.</p><p>Those who control the endpoint&#8212;such as device manufacturers&#8212;may enjoy significant competitive advantages. While AI tools have made building new applications easier than ever, maintaining a competitive edge becomes more challenging.</p><h2>Edge AI: Speed, Personalization, Privacy</h2><p>Edge computing means processing data directly on your device instead of distant cloud servers. The result? Response times improve dramatically&#8212;up to 60% faster, making applications like voice assistants and real-time translation smoother and more natural.</p><p>Privacy benefits significantly too. Devices such as Google's predictive keyboard (Gboard) keep sensitive data local, offering personalized recommendations without compromising your privacy. </p><p>Edge AI and edge computing directly fuel the growth of inference by bringing AI capabilities closer to the end user. By processing data on-device, edge AI significantly cuts down latency, enables personalized experiences, and strengthens data privacy&#8212;all critical factors for real-time applications and user adoption. In essence, the edge enables inference to reach its full potential, powering applications that demand instant, context-aware interactions.</p><p>This shift presents substantial investment opportunities, particularly around software infrastructure that enables effective edge deployment. Four key software segments stand out:</p><h3>1. Data Infrastructure: Real-Time Insights</h3><p>Effective AI needs rapid data processing. Gartner predicts over 75% of enterprise data will be processed at the edge by 2025, boosting demand for robust local data management solutions.</p><ul><li><p><strong>Key Companies:</strong> Confluent, Snowflake, Databricks</p></li></ul><h3>2. Compute Infrastructure: Optimized Edge Performance</h3><p>Specialized software enhances hardware capabilities, maximizing the performance of edge applications. IDC forecasts edge computing infrastructure investments reaching $55.7 billion by 2025.</p><ul><li><p><strong>Key Companies:</strong> Anyscale, OctoML, Edge Impulse</p></li></ul><h3>3. Orchestration: Simplifying Complexity at Scale</h3><p>Managing extensive networks of edge devices requires effective orchestration solutions. The edge orchestration market is projected to grow from $3.4 billion in 2023 to approximately $8.5 billion by 2028.</p><ul><li><p><strong>Key Companies:</strong> HashiCorp, ZEDEDA, Rancher Labs</p></li></ul><h3>4. AI Governance: Ensuring Trust and Compliance</h3><p>As AI integration grows, governance tools that guarantee fairness, transparency, and regulatory compliance become crucial. Deloitte anticipates AI governance software investments will triple by 2027.</p><ul><li><p><strong>Key Companies:</strong> Credo AI, Fiddler AI, TruEra, Alignmt AI</p></li></ul><p>Edge AI is more than just the next tech trend&#8212;it's reshaping industries by fundamentally changing user interactions. It could be the next gold rush.</p><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading Lake&#8217;s Substack! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item><item><title><![CDATA[AI Trends from CMU’s Applied AI Class]]></title><description><![CDATA[The following insights are distilled from the CMU Applied AI class I teach, reflecting collaborative learnings from leaders across top industry players such as Nvidia, Google, Github, Anyscale, and Linkedin.]]></description><link>https://lakedai.substack.com/p/ai-transformative-trends</link><guid isPermaLink="false">https://lakedai.substack.com/p/ai-transformative-trends</guid><dc:creator><![CDATA[Lake Dai]]></dc:creator><pubDate>Sun, 02 Mar 2025 04:37:50 GMT</pubDate><enclosure url="https://substackcdn.com/image/fetch/$s_!6uOE!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p></p><blockquote><div class="captioned-image-container"><figure><a class="image-link image2 is-viewable-img" target="_blank" href="/__u/substackcdn.com/image/fetch/$s_!6uOE!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg" data-component-name="Image2ToDOM"><div class="image2-inset"><picture><source type="image/webp" srcset="/__u/substackcdn.com/image/fetch/$s_!6uOE!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg 424w, /__u/substackcdn.com/image/fetch/$s_!6uOE!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg 848w, /__u/substackcdn.com/image/fetch/$s_!6uOE!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg 1272w, /__u/substackcdn.com/image/fetch/$s_!6uOE!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_webp, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg 1456w" sizes="100vw"><img src="/__u/substackcdn.com/image/fetch/$s_!6uOE!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg" width="1280" height="720" data-attrs="{&quot;src&quot;:&quot;https://substack-post-media.s3.amazonaws.com/public/images/0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg&quot;,&quot;srcNoWatermark&quot;:null,&quot;fullscreen&quot;:null,&quot;imageSize&quot;:null,&quot;height&quot;:720,&quot;width&quot;:1280,&quot;resizeWidth&quot;:null,&quot;bytes&quot;:216408,&quot;alt&quot;:null,&quot;title&quot;:null,&quot;type&quot;:&quot;image/jpeg&quot;,&quot;href&quot;:null,&quot;belowTheFold&quot;:false,&quot;topImage&quot;:true,&quot;internalRedirect&quot;:&quot;https://lakedai.substack.com/i/158206632?img=https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg&quot;,&quot;isProcessing&quot;:false,&quot;align&quot;:null,&quot;offset&quot;:false}" class="sizing-normal" alt="" srcset="/__u/substackcdn.com/image/fetch/$s_!6uOE!, /__u/lakedai.substack.com/w_424, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg 424w, /__u/substackcdn.com/image/fetch/$s_!6uOE!, /__u/lakedai.substack.com/w_848, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg 848w, /__u/substackcdn.com/image/fetch/$s_!6uOE!, /__u/lakedai.substack.com/w_1272, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg 1272w, /__u/substackcdn.com/image/fetch/$s_!6uOE!, /__u/lakedai.substack.com/w_1456, /__u/lakedai.substack.com/c_limit, /__u/lakedai.substack.com/f_auto, /__u/lakedai.substack.com/q_auto:good, /__u/lakedai.substack.com/fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0aaf4090-ad93-4633-bae7-9af0c1b88d2f_1280x720.jpeg 1456w" sizes="100vw" fetchpriority="high"></picture><div class="image-link-expand"><div class="pencraft pc-display-flex pc-gap-8 pc-reset"><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container restack-image"><svg aria-hidden="true" width="20" height="20" viewBox="0 0 20 20" fill="none" stroke-width="1.5" stroke="var(--color-fg-primary)" stroke-linecap="round" stroke-linejoin="round" xmlns="http://www.w3.org/2000/svg"><g><path d="M2.53001 7.81595C3.49179 4.73911 6.43281 2.5 9.91173 2.5C13.1684 2.5 15.9537 4.46214 17.0852 7.23684L17.6179 8.67647M17.6179 8.67647L18.5002 4.26471M17.6179 8.67647L13.6473 6.91176M17.4995 12.1841C16.5378 15.2609 13.5967 17.5 10.1178 17.5C6.86118 17.5 4.07589 15.5379 2.94432 12.7632L2.41165 11.3235M2.41165 11.3235L1.5293 15.7353M2.41165 11.3235L6.38224 13.0882"></path></g></svg></button><button tabindex="0" type="button" class="pencraft pc-reset pencraft icon-container view-image"><svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" class="lucide lucide-maximize2 lucide-maximize-2"><polyline points="15 3 21 3 21 9"></polyline><polyline points="9 21 3 21 3 15"></polyline><line x1="21" x2="14" y1="3" y2="10"></line><line x1="3" x2="10" y1="21" y2="14"></line></svg></button></div></div></div></a></figure></div><p> The following insights are distilled from the CMU Applied AI class I teach, reflecting collaborative learnings from leaders across top industry players such as Nvidia, Google, Github, Anyscale, and Linkedin. These discussions have highlighted transformative trends and concrete data that shape our understanding of AI&#8217;s future. Below, you&#8217;ll find an in-depth look at how smarter AI&#8212;focused on data quality, computational efficiency, and human collaboration&#8212;is set to revolutionize robotics, infrastructure, insurance, and developer productivity.</p></blockquote><ol><li><p><strong>Generative AI in Robotics</strong></p></li></ol><blockquote><p><strong>Takeaway: </strong>Generative AI is positioning robots to move from structured factory tasks to human-centric services&#8212;provided that latency and reasoning challenges are overcome.</p><p>Vision-Language Models (VLMs) meet robotics: Founders showcased how combining visual and textual intelligence can make robots smarter and more adaptable. For example, cutting-edge models like Google&#8217;s PaLM-E integrate vision and language to enable robots to understand complex instructions. However, technical challenges remain&#8212;current VLMs struggle with high latency and poor spatial reasoning in real-time environments, causing delays in response and difficulty understanding 3D space, which is critical for tasks like navigation and manipulation. Researchers are actively addressing these gaps with enhancements in spatial reasoning, but further progress is needed for truly fluid human-robot interactions.</p><p>Despite an initial focus on industrial automation, the potential applications extend far beyond the factory floor. Opportunities in elderly care, hospitality, and retail are emerging. We&#8217;re already seeing early signs: companion robots for seniors (like ElliQ) can remind patients to take medication and track health metrics, while service robots in hotels handle room deliveries. The service robotics market is set to nearly double from approximately $47&#8239;billion in 2024 to about $98.6&#8239;billion by 2029, driven by such use cases. This suggests that &#8220;tech-driven convenience&#8221; could become a new norm, with generative AI enabling robots to provide personalized assistance in homes, hospitals, and hotels.</p><p>2. <strong>AI Infrastructure &amp; Democratization</strong></p><p><strong>Takeaway: </strong>The &#8220;picks and shovels&#8221; opportunity&#8212;building the tools and platforms that make AI development accessible.</p><p>There is a rising trend in AI infrastructure unicorns that bridge software development and large-scale AI execution. The evolution of AI infrastructure is a story of scaling data and compute to keep up with exploding AI models. An expert outlined the pipeline stages&#8212;data processing, model training, and inference&#8212;and noted that high-quality data is both scarce and costly. Enterprises are now dealing with petabyte- to exabyte-scale datasets; for example, one major technology company recently processed 1.5 exabytes of data in one quarter using a distributed AI workload, requiring over 10,000 years of CPU time in aggregate. To orchestrate workloads of this magnitude, new tools for distributed computing (like Ray) are emerging, signaling an AI infrastructure revolution.</p><p>There is a rising trend in AI infrastructure unicorns that bridge software development and large-scale AI execution. This includes MLOps platforms valued at over $1B and model hubs worth around $4.5B in 2023, which help developers access models and data easily. Nearly 45% of new early-stage unicorns in 2024 were AI-related startups, many focusing on enabling technology. Even established incumbents have seen dramatic growth, with one company controlling about 95% of the AI GPU market and experiencing a 241% stock jump in 2023 as demand for AI chips surged. Democratization of AI&#8212;through accessible tools, cloud-based services, and open-source frameworks&#8212;is crucial for bridging the gap between advanced AI builders and industry users, setting the stage for the next wave of innovation.</p><p><strong>3. AI in Insurance &amp; Real Estate</strong></p><p><strong>Takeaway: </strong>AI is augmenting human expertise, not replacing it, in specialized domains like insurance.</p><p>One presentation detailed how an AI-driven system accelerates property underwriting by analyzing drone images and data to assess risk&#8212;yet experienced underwriters remain essential to validate the AI&#8217;s suggestions. This hybrid approach reflects industry best practices: AI can scan vast amounts of data faster (boosting underwriting productivity by over 50% and cutting risk analysis from days to minutes), but final judgment calls still rely on human expertise. Leading insurers stress that AI must be applied with human oversight to ensure fairness and accuracy in decisions. Notably, underwriters today spend roughly 40% of their time on administrative tasks&#8212;a $160&#8239;billion efficiency gap over five years that AI automation could recapture, allowing professionals to focus on more complex cases.</p><p>Data is the lifeblood of AI in insurance, yet data scarcity and regional variability pose major challenges. An innovative incentive model was proposed to encourage property owners and partners to share data, enlarging training datasets. Similar trends are already emerging in auto insurance, where devices or apps reward customers for sharing driving data, thereby crowdsourcing improved data for AI models. Although the insurance sector remains conservative and heavily regulated, the insurtech market is growing&#8212;with AI in underwriting projected to reach approximately $41&#8239;billion by 2033. The future will likely see stronger partnerships between insurers and AI vendors, increased data pooling, and expanded applications in claims and fraud detection, as well as real estate where AI-driven drones and computer vision could offer instant property assessments.</p><p><strong>4. Smarter AI Scaling &amp; Cost Efficiency</strong></p><p><strong>Takeaway: </strong>optimal use of data and compute can outperform brute-force scaling.</p><p>In the era of foundation models, the prevailing mindset has been &#8220;bigger is better&#8221;&#8212;more parameters, more data, more compute. Recent research, however, is shifting the focus to &#8220;better is better,&#8221; meaning that optimal use of data and compute can outperform brute-force scaling. DeepMind&#8217;s Chinchilla project showed that a 70B model trained on 1.4 trillion tokens can outperform a 175B model trained on less data, proving that a carefully optimized smaller model can be more effective than a much larger one. This finding has influenced AI labs to prioritize data quality, innovative model architectures, and efficient training techniques over simply increasing model size. Techniques like synthetic data generation and transfer learning are also gaining traction to boost performance without relying solely on additional real data.</p><p>Cost efficiency is paramount, as running gigantic models is extremely expensive&#8212;training estimates for GPT-3 and GPT-4 were around $4.6&#8239;million and $100&#8239;million, respectively. One case study highlighted that a major company saved over $100&#8239;million annually by migrating big data workloads from an older platform to an optimized distributed computing solution. An engineering report confirmed over $120&#8239;million per year in savings after adopting such a framework for large-scale data processing. These innovations in AI infrastructure&#8212;through efficient frameworks, better hardware utilization, model compression, and distillation&#8212;are significantly reducing costs and energy usage. Looking ahead, the &#8220;smarter not bigger&#8221; ethos is opening exciting avenues in drug discovery (with AI-designed candidates achieving an 80&#8211;90% success rate in Phase I trials), multimodal AI (integrating text, vision, and audio by 2025), and sustainability, with efforts to reduce the carbon footprint of large-scale AI deployments.</p><p><strong>5. AI for Developer Productivity</strong></p><p><strong>Takeaway: </strong>AI is not just enhancing coding productivity; it is leveling up productivity across multiple roles. However, balancing productivity with security and privacy remains crucial.</p><p>Empowering developers with AI is transforming productivity. Innovative platforms now use AI to turn code artifacts into meaningful career insights, automatically mapping a developer&#8217;s skills and contributions, and even aiding in performance reviews. This AI-driven skill mapping identifies strengths in various programming languages, frameworks, and problem domains, suggesting personalized growth areas&#8212;much like having a career coach that reads your Git commits. The potential extends to team formation, mentorship matching, and upskilling, envisioning an AI that recommends new tools to learn based on industry trends and individual profiles.</p><p>AI is not just enhancing coding productivity; it is leveling up productivity across multiple roles. Recent studies indicate that 80% of jobs will have at least 10% of tasks impacted by GPT-style AI, with about 19% of jobs seeing 50% of tasks affected. This trend means that fields ranging from finance and law to creative industries stand to benefit significantly. However, balancing productivity with security and privacy remains crucial. Concerns over using external large language models (LLMs) on proprietary code have led some companies to restrict such tools, opting instead for secure, on-premise solutions that safeguard data. AI coding assistants like GitHub Copilot have already shown impressive gains, with controlled trials revealing a 26% increase in code completions per week for users compared to non-users, underscoring AI&#8217;s role as a collaborative partner in software development.</p></blockquote><div><hr></div><blockquote><p>Across these diverse domains, a common thread emerges: AI&#8217;s transformative but complementary role. Whether enhancing robotic reasoning, optimizing AI infrastructure for scalability, streamlining underwriting in insurance, or boosting developer productivity, the frontier of applied AI is about amplifying human capabilities and unlocking new possibilities. The insights highlight an industry-forward mindset&#8212;focusing on practical integration, efficiency, and responsibility. As generative AI and advanced models continue to push boundaries, success in the real world will depend on robust support systems ranging from infrastructure and cost optimization to data governance and human-in-the-loop validation. For AI professionals and decision-makers, the mandate is clear: <strong>invest in the &#8220;boring&#8221; infrastructure and tools that make AI scalable, address real-world challenges, and always consider the human element.</strong> If executed well, the next few years will see AI not just getting bigger, but getting better&#8212;tangibly benefiting industries and society, one breakthrough at a time.</p></blockquote><div class="subscription-widget-wrap-editor" data-attrs="{&quot;url&quot;:&quot;https://lakedai.substack.com/subscribe?&quot;,&quot;text&quot;:&quot;Subscribe&quot;,&quot;language&quot;:&quot;en&quot;}" data-component-name="SubscribeWidgetToDOM"><div class="subscription-widget show-subscribe"><div class="preamble"><p class="cta-caption">Thanks for reading Lake&#8217;s Substack! Subscribe for free to receive new posts and support my work.</p></div><form class="subscription-widget-subscribe"><input type="email" class="email-input" name="email" placeholder="Type your email&#8230;" tabindex="-1"><input type="submit" class="button primary" value="Subscribe"><div class="fake-input-wrapper"><div class="fake-input"></div><div class="fake-button"></div></div></form></div></div>]]></content:encoded></item></channel></rss>