This week at a glance:

• The new engines of open source AI: NVIDIA now leads Hugging Face contributions past 600 repositories in the past year, open-weight models sit inside a comparable performance band to proprietary frontier ones, and for a growing share of production workloads the remaining quality gap is smaller than the cost gap.

• Policy briefs and studies on reward hacking, the UN AI panel, and the EU AI Act

• AI content marking: five studies on AI content marking found that none of today’s watermarking methods meet all four of the requirements set out in Article 50, and that only 38% of deployed generators adequately mark their output.

• Startup spotlights: goNEON, Cascade, and TidalSense each closed rounds this week across infrastructure design automation, construction procurement intelligence, and point-of-care COPD diagnostics.

1. The new engines of open source AI

Hugging Face repository activity through the first half of 2026 shows Google, Meta, Microsoft, Tencent, Alibaba, Baidu, and NVIDIA among the largest contributors to the open-source AI ecosystem, counting only repositories with meaningful traction. NVIDIA now leads with over 600 repositories in the past year and crossed 1,000 total in June, a remarkable shift for a company whose roots are in hardware. Alibaba’s Qwen series has become one of the world’s most widely adopted open model families, and Ai2’s Olmo series tops the Artificial Analysis Openness Index as one of the few fully open options credibly challenging Chinese models on adoption.

Plotted against output cost, the strongest open-weight models now sit inside a comparable performance band to the frontier proprietary ones and well to the left on price: DeepSeek V4 Pro, GLM 5.2, and Kimi K2 each land at or above the scores of models costing five to ten times more per million tokens. For a growing share of production workloads, the remaining quality gap is smaller than the cost gap, and where enterprises still pay the premium it is increasingly for reliability, long-context behaviour, and indemnification rather than raw capability.

→ Read the full article at AI World

2. Policy briefs and studies on reward hacking, the UN AI panel, and the EU AI Act

OpenAI acknowledged that two frontier models, safety refusals off and told to win an internal benchmark, found a zero-day vulnerability and pulled test answers directly from Hugging Face’s servers. Researchers call the pattern specification gaming, documented across games, coding tests, and in rare cases models editing their own reward functions. The UN’s Independent International Scientific Panel on AI published its first report this month, finding capability doubling every four to seven months and wealth concentrated in a few firms across two countries while most states cannot inspect the systems they depend on. The EU AI Act’s transparency rules take effect 2 August on schedule; high-risk obligations land in 2027 and 2028.

→ Read the full analysis on AI World Linkedin page

3. This week’s most notable AI research papers

Can AI-generated content actually be marked reliably? Five studies examine the question the EU AI Act’s Article 50 makes urgent. None of the watermarking methods tested clears all four qualities the regulation demands simultaneously; the strongest are built into model training, not stamped onto outputs. A light image edit removes a watermark while leaving the image intact. An audit of generators already in deployment finds only 38% marking their output adequately. The marking obligation becomes binding on 2 August. The Code of Practice meant to operationalise it, published seven weeks earlier, stays voluntary.

→ Read the full paper summaries on AI World

4. Startup spotlights of the week

This week’s selection showcases the application of AI to structural gaps in civil infrastructure planning, construction procurement timing, and primary care respiratory diagnostics.

🇨🇭 goNEON: Switzerland-based ETH Zurich spin-out, co-founded by Raphael Eder and Dr Lukas Ballo. goNEON’s agentic AI platform takes engineering requirements, regulatory frameworks, and site constraints as inputs and generates multiple technically feasible, compliant design alternatives in minutes, giving planning teams a broader range of validated options before committing to a direction. Secured €160,000 from Venture Kick.

🇬🇧 Cascade: co-founded by Hannia Zia and Joana Ferreira. Cascade tracks the signals that precede formal procurement in construction and engineering, scanning public filings, permits, budget announcements, and government meeting minutes and ranking them against each firm’s track record, providing an earlier view of the project pipeline with go/no-go intelligence and AI-assisted proposal drafts. Raised $3.5 million in a seed round led by Andreessen Horowitz Speedrun.

🇬🇧 TidalSense: Cambridge-based, led by Dr Ameera Patel. TidalSense adapted CO₂ waveform analysis into a handheld point-of-care device that assesses COPD in 75 seconds using AI models trained on more than 2.5 million patient breaths, enabling four to six assessments per hour in any clinical setting without specialist training. Raised $19 million in a growth round led by Cross-Border Impact Ventures, bringing total funding to $40 million.

What’s next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s round-up directly in your inbox and stay ahead of key developments.

Have a great week ahead!

Gaia Cavaglioni
On behalf of the AI World Team

This week at a glance:

• Open source is one release cycle behind the frontier: Kimi K3, a 2.8 trillion parameter open model from Moonshot AI, places around Opus 4.8 tier on independent benchmarks, at $15 per million output tokens against Fable 5’s $50, compressing the lag between open and proprietary to a single release cycle.

• AI’s energy footprint, five studies: renewable expansion relaxes the AI scaling constraint without displacing fossil generation, and the external cost of US data centre emissions already runs to roughly $25 billion a year, borne by surrounding communities rather than the sector itself.

• Three policy briefs on efficiency, capital, and grid: the IEA finds data centre demand growing 50% in 2025 despite falling energy per query; Big Tech capital spending in 2026 is projected to exceed global oil and gas investment; and Europe’s plan to triple data centre capacity by the early 2030s depends on grid connection, not permits or money.

• Startup spotlights: Hyperion Robotics, Sodex Innovations, and Sightera Biosciences each closed rounds this week across robotic concrete manufacturing, construction site real-time data, and patient-derived oncology drug discovery.

1. Open source is one release cycle behind the frontier

Moonshot AI has released Kimi K3, a 2.8 trillion parameter mixture-of-experts model with native vision, a 1M token context window, and two new attention mechanisms the company claims yield roughly 2.5 times better scaling efficiency than its predecessor.

On independent benchmarks from Artificial Analysis, K3 places around Opus 4.8 tier, winning or placing second on most coding and agentic evaluations against Claude Opus 4.8, GPT 5.5, and GLM-5.2. Moonshot’s own blog acknowledges K3 still trails Claude Fable 5 and GPT 5.6 Sol, but an open model outperforming last generation’s Western flagships weeks after their successors shipped compresses the lag between open and proprietary to something like a single release cycle, a gap that used to be measured in more than a year.

→ Read the full article at AI World

2. AI’s energy footprint, five studies

US AI server deployment could add 24 to 44 million tonnes of CO2 equivalent a year by 2030, and best industry practice covers only 73% of residual emissions. Gui and Dai (2026) show why renewable expansion does not close the gap: markets rewarding capability scaling push developers towards larger models regardless of carbon intensity, so new clean capacity relaxes the scaling constraint without reducing fossil generation. Muller (2026) prices the external damage from US data centre emissions at roughly $25 billion a year, currently unpriced by the sector.

→ Read the full paper summaries on AI World

3. Three policy briefs on efficiency, capital, and grid

The IEA finds energy per AI task has fallen by roughly an order of magnitude per year, yet global data centre electricity grew 17% in 2025 and AI-focused data centre consumption grew 50%, with demand projected to roughly double to 950 TWh by 2030; heavier tasks including reasoning, agents, and video use hundreds to thousands of times more energy per query, and cheaper tasks invite more tasks. The IEA's World Energy Investment 2026 report projects Big Tech capital spending, most of it on data centres, reaching $715 billion in 2026, more than global oil and gas investment at $546 billion and more than these companies earn, with spending now absorbing almost all their cash and the pace of new electricity demand becoming a function of market sentiment. The same IEA report finds Europe's data centre pipeline implies 2.3 times current installed capacity under the AI Continent Action Plan's target of tripling by the early 2030s, with the Commission's own European Grids Package tabled in 2025 to address connection queues, and the report's conclusion unambiguous: overcoming energy constraints is the critical variable.

→ Read the full analysis on AI World Linkedin page

4. Startup spotlights of the week

This week’s selection showcases the application of AI and robotics to structural gaps in infrastructure manufacturing, construction site data collection, and patient-derived drug discovery.

🇫🇮 Hyperion Robotics: Finland-based, co-founded by Fernando De los Ríos, Ashish Mohite, and Henry Unterreiner. Hyperion’s robotic microfactory platform uses 3D printing and AI to manufacture low-carbon concrete components locally, with its Forge software calculating lightweight geometries that use up to 75% less material and production processes that reduce carbon emissions by up to 70% compared to conventional methods. Raised $7.4 million in a growth round co-led by Course Corrected and the European Innovation Council Fund.

🇦🇹 Sodex Innovations: Austria-based, co-founded by Ralf Pfefferkorn, Bernhard Gantner, and Raphael Ott. Sodex mounts laser scanning hardware on existing construction and mining machinery, turning excavators into continuous data collectors that feed real-time terrain models, stockpile volume calculations, and quantity documentation into its SDX-Cloud platform without interrupting operations. Raised €4 million in a round led by Capmont Technology.

🇧🇪 Sightera Biosciences: Belgium-based, spun out from the University of Antwerp and Antwerp University Hospital. Sightera trains its generative AI on proprietary datasets generated from patient-derived samples of therapy-resistant cancers and fibrosis, producing drug candidates designed against actual diseased tissue biology rather than generic chemical databases, with lead programme SIGHT001 advancing towards preclinical candidate selection. Raised €3 million in a pre-seed round co-led by Entourage, Anacura, and QBIC.

What’s next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s round-up directly in your inbox and stay ahead of key developments.

Have a great weekend!

Gaia Cavaglioni

On behalf of the AI World Team.

This week at a glance:

• System prompts as a signal: Anthropic’s Fable 5 system prompt now runs to around 25,000 words, the longest in the field, with safety and user wellbeing guidance making up a larger share than any other provider.

• AI energy economics, five studies: renewable expansion mainly relaxes the AI scaling constraint without displacing fossil generation, and past 2030 grid reliability will determine where frontier AI infrastructure gets built more than raw capacity will.

• Three policy briefs on energy, safety, and grids: the IEA projects data centre electricity demand roughly doubling to 950 TWh by 2030; the Future of Life Institute gives Anthropic the field’s top safety score of C+, with Mistral among three labs failing outright; and 60% of European grid operators already use AI on their networks.

• Startup spotlights: Polysense, Bohr Energie, and CurifyLabs each closed rounds this week across food production quality control, independent renewable energy market access, and automated pharmaceutical compounding.

1. System prompts and what they tell us about the chat before the chat

Before every conversation with a major AI model, the lab loads a system prompt that shapes how the model behaves. Tracking these prompts over time reveals how lab priorities have shifted. The strongest trend is length: Anthropic's Fable 5 now runs to around 25,000 words, approximately 70 pages, driven by the compounding weight of personalisation instructions and behavioural guidance.

Safety and user wellbeing content accounts for a larger share of Anthropic's prompts than any other provider in the field. At the other end, Grok keeps system prompts minimal. The differences are a direct signal of what each lab considers worth encoding before a conversation begins.

→ Read the full article at AI World

2. From the chip to the grid

This week's selection examines the energy economics of AI scaling across five studies, from the hardware and electricity constraints binding US output to the adaptation trap that makes renewable expansion a scaling enabler, and the spatial specificity of where European grid reliability will determine AI infrastructure deployment past 2030.

→ Read the full paper summaries on AI World

3. Three policy briefs on energy, safety, and grids

The IEA’s special report on AI and energy projects predicts that data centre electricity demand will roughly double to around 950 TWh by 2030, with energy per AI query falling fast as video, reasoning, and agent tasks consume hundreds of times more per query than basic text generation. However, AI could also save over 13 exajoules of energy by 2035. The problem is that grids, transformers, and chips are unable to scale up as quickly as models are being developed.

The Future of Life Institute’s Summer 2026 AI Safety Index grades nine companies across 37 indicators. Anthropic leads with a C+, the highest score in the field, and Mistral is among three companies that fail outright. Existential safety is the weakest domain across the industry, with no company scoring above a C-, and four major labs have weakened the pause commitments they made when the capability thresholds seemed far off. A JRC report on AI in European energy infrastructure found that 60% of grid operators were already using AI to forecast demand and manage networks, 70% were using it to save energy and 90% were aware of the AI Act. However, awareness was not yet translating into compliance.

→ Read the full analysis on AI World Linkedin page

4. Startup spotlights of the week

This week's selection showcases the application of AI to structural gaps in food production quality control, independent renewable energy market access, and pharmaceutical compounding.

🇧🇪 Polysense: Belgium-based, co-founded by Yarne De Munck, Lucas Van Dijck, and Jarne Bogaert. Polysense places cameras on existing production lines for continuous unit-level inspection, combines quality and process data into a single intelligence platform, and automatically adjusts machine parameters in response to changing ingredient conditions without operator intervention. Raised $10.7 million in an oversubscribed seed round led by Felix Capital, bringing total funding to approximately $12.9 million.

🇫🇷 Bohr Energie: Toulouse-based, founded 2020 by Julien Haure, Luis Urday, Julien Chollet, and Jean-Pierre Mader. Bohr Energie gives independent renewable energy producers access to production forecasting, real-time operational control, regulatory compliance, and direct electricity market trading through a managed service that does not require in-house trading capability. Raised €10 million in a Series A led by Suma Capital through its SC Net Zero Ventures fund.

🇫🇮 CurifyLabs: Finland-based, founded 2021 by Charlotta and Niklas Sandler Topelius. CurifyLabs integrates software, GMP-manufactured excipient bases, and 3D printing hardware into an end-to-end automated pharmaceutical compounding system that generates digital batch records and audit trails, reducing compounding time by up to nine times compared to manual processes. Raised €12 million in a Series A co-led by Sandwater and HealthCap, bringing total funding to approximately €18.7 million.

What’s next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s round-up directly in your inbox and stay ahead of key developments.

Have a great weekend!

Gaia Cavaglioni

On behalf of the AI World Team.

This week at a glance:

• AI is making reconnaissance cheap: five studies trace a consistent pattern across coding, geolocation, and zero-day exploitation, where every apparent ceiling on AI-enabled reconnaissance has turned out to be a fixable design choice rather than a hard limit.

• Three policy briefs on surveillance, regional divides, and cyber risk: AI-powered OSINT can now assemble personal profiles from public data with almost no oversight; firm-level AI adoption across Europe remains concentrated in a handful of hubs despite exposure being spread evenly; and 94% of leaders name AI as the biggest force reshaping cyber risk.

• Startup spotlights: Kalipso, Digiclean, and ALP Bio each closed rounds this week across regulatory compliance automation, industrial cleaning chemistry, and pre-clinical antibody immunogenicity screening.

1. How AI is making reconnaissance cheap

AI is rapidly removing the skill and access barriers that used to limit who could conduct reconnaissance - coding, OSINT, geolocation, and vulnerability exploitation - and every limitation that initially seemed like a hard ceiling has turned out to be a fixable design gap rather than a genuine constraint. This means that the same democratising effect that enables beginners to learn to code and journalists to verify video locations also lowers the threshold for malicious actors to track individuals or exploit systems.

→ Read the full breakdown on AI World

2. Three policy briefs on surveillance, regional divides, and cyber risk

A review in AI & Society finds that AI turns scattered public traces into a detailed personal profile in minutes, that the most advanced OSINT tools run with almost no oversight, and that only around 12% of research on the field even touches the ethical and legal questions it raises. A Regional Studies paper mapping AI across EU regions on innovation, exposure, and use finds a clean paradox: exposure to AI is now spread fairly evenly across the continent, but firm-level adoption stays locked in a handful of high-income, high-skill hubs, with institutional, organisational, and capability frictions, rather than technology, holding other regions back. The WEF Global Cybersecurity Outlook 2026, produced with Accenture, finds AI named by 94% of leaders as the biggest force reshaping cyber risk, with 87% citing AI-related vulnerabilities as the fastest-growing threat of the past year, and the share of organisations testing AI tools before deployment nearly doubling from 37% to 64% in a single year.

→ Read the full analysis on AI World Linkedin page

3. Startup spotlights of the week

This week's selection showcases the application of AI to structural gaps in corporate regulatory compliance, industrial chemical process optimisation, and pre-clinical drug development.

🇪🇸 Kalipso: Barcelona-based, co-founded by Pierre Ferran and Virginia Debernardi. Kalipso tracks over 100 legal sources across 40 jurisdictions, processing more than 3,000 daily regulatory updates and mapping each rule to an organisation’s specific business model, then pinpointing policy gaps and generating traceable action plans with full audit trails, covering frameworks including GDPR, DORA, the EU AI Act, and MiCAR. Raised $3.2 million in a seed round backed by Varsity, with Lanai, Plug and Play, Kima Ventures, and Vento participating.

🇸🇪 Digiclean: Sweden-based, founded by Charlotte Stigen Låstberg and Andreas Låstberg. Digiclean installs sensors directly into factory chemical baths to continuously analyse composition and automatically adjust dosages in real time, removing the manual testing cycle from industrial cleaning and reducing chemical use, wastewater, and unplanned downtime. Raised €2.5 million in a seed round co-led by Unconventional Ventures and Almi Invest GreenTech.

🇨🇭 ALP Bio: Switzerland-based, co-founded by Dr Christian Vahlensieck, Dr Lucas Schaus, Anatol Ehrlich, and Punit Mehra. ALP Bio screens therapeutic antibody candidates for immunogenicity risk before clinical development begins, combining human tonsil-derived immune tissue with AI-powered sequence analysis and generative AI-guided epitope redesign to identify and reduce immune risk at the discovery stage. Raised €161,000 from Venture Kick Stage 3, following a CHF 1.75 million pre-seed led by 42CAP earlier this year.

What’s next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s round-up directly in your inbox and stay ahead of key developments.

Have a great weekend!

Gaia Cavaglioni

On behalf of the AI World Team.

This week at a glance:

• Internet crawlers should give and take, by now they mostly take: Cloudflare data shows Anthropic’s Claude generating around 71,000 page requests for every referral sent back to a source, against Google’s near-parity ratio, with publishers already returning more than one billion HTTP 402 responses per day to AI crawlers.

• AI can make climate data legible, but synthetic fidelity erodes the trust it depends on: three studies find that AI improves climate risk prediction and public reach, but that realistic AI-generated disaster images reduce willingness to act when their artificial origin is detected or suspected.

• Two policy briefs on energy and mobility

• Startup spotlights: Compri, Timefold, and JUPUS each closed rounds this week across industrial procurement automation, enterprise scheduling infrastructure, and AI-powered legal operations.

1. Internet crawlers should give and take, by now they mostly take

Automated systems now account for 57.5% of global HTTP requests, more than a year ahead of Cloudflare's own forecast. A new Cloudflare metric comparing AI crawler activity against referral traffic finds that Anthropic's Claude generated around 71,000 page requests for every referral sent back to a source in the week of 19 to 26 June 2025, reflecting its heavy use in agentic and API-driven workflows where users never encounter the underlying links.

The contrast with traditional search is structural: Google's ratio sits closer to parity, because users still navigate to sources to read content in full. Publishers are already responding, with Cloudflare customers returning more than one billion HTTP 402 responses per day to AI crawlers, and the Wikimedia Foundation attributing a 50% rise in multimedia bandwidth since January 2024 to automated scrapers. The interface through which users choose to interact with AI determines not just how they experience it, but what the open web receives in return.

→ Read the full story at AI World

2. AI can make climate data legible, but synthetic fidelity erodes the trust it depends on

AI improves climate risk prediction and public reach, but the trust dynamics run in the opposite direction.

Three studies trace the same asymmetry. AI now predicts floods, droughts, and heatwaves more accurately and reaches more people than any prior medium, but Atkins et al. (2024) found its climate accuracy uneven across the regions that need it most, and Eppler (2026) found across 2,580 participants that realistic AI-generated disaster images erode trust and reduce willingness to act when their synthetic origin is suspected. Buck et al. (2026) argue the IPCC's core function is sustaining a shared basis of reality, and that AI backlash could erode exactly that.

→ Read the full paper summaries on AI World

3. Two policy briefs on energy and mobility

Deployment is outrunning readiness across two sectors. A UCL model puts the grid cost of European AI data centres at up to 323 GW of new capacity and 25 million additional tonnes of CO2 by 2050. An OECD and ITF paper concludes AI governs urban traffic most effectively as a guidance layer, adjusting flow without direct infrastructure control.

→ Read the full analysis on AI World Linkedin page

4. Startup spotlights of the week

This week’s selection showcases the application of AI to structural gaps in industrial procurement coordination, enterprise scheduling optimisation, and legal administrative operations.

🇮🇹 Compri: Milan-based, founded 2024 by Edoardo Arbizzi and Edoardo Gava. Compri deploys specialised AI agent teams across procurement and supply chain workflows, handling supplier chasing, compliance auditing, order tracking, and paperwork collection from a centralised hub integrating ERP systems, emails, and regulatory databases, with full SAP and Oracle integration and EU-hosted GDPR-compliant infrastructure. Raised €3.2 million in a seed round led by Picus Capital, bringing total funding to over €5 million.

🇧🇪 Timefold: Belgium-based, co-founded by Maarten Vandenbroucke and Geoffrey De Smet. Timefold provides scheduling optimisation infrastructure for software teams embedding routing and workforce scheduling into enterprise applications, combining AI with deterministic constraint optimisation across field service routing, shift scheduling, pick-up and delivery, and task scheduling. Raised $13 million in a Series A led by Alstin Capital, bringing total funding to approximately $19 million.

🇩🇪 JUPUS: Germany-based, founded by René Fergen, Jannis Gebauer, and Oskar Handrick. JUPUS automates the administrative core of law firm operations through three modules covering AI voice reception, client onboarding and case file creation, and document analysis drawing on a database of over one million German court rulings, built to comply with DSGVO, BRAO, and §203 StGB. Raised €13 million in a Series A led by Semapa Next.

What’s next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s round-up directly in your inbox and stay ahead of key developments.

Have a great weekend!

Gaia Cavaglioni

On behalf of the AI World Team.

This week at a glance:

• The first frontier model to get banned: three days after launch, Claude Fable 5 was pulled worldwide after a US government directive targeted foreign national access, establishing for the first time that proprietary AI models can be treated as export-controlled assets and leaving every non-American working at a US AI company with a newly clarified understanding of how precarious that access is.

• Where European AI compute is heading: Europe is projected to hold around 5% of global AI compute capacity by 2031, with France set to account for 35% of the continental total if current plans are executed, while the US is forecast to hold roughly 70% globally.

• Training a European frontier model: five papers map the technical path from distributed compute to a model at scale, and find the hardware increasingly tractable and the governance questions still largely unresolved.

• Startup spotlights: Qorelo, Lightbringer, and OneSoil each closed rounds this week across SAP migration automation, AI-assisted patent filing, and precision agriculture.

1. The first frontier model to get banned

Three days after launch, Claude Fable 5 was suspended worldwide after Anthropic received a government directive ordering immediate compliance and citing national security authorities. The trigger was reportedly a claim that Mythos 5 had been jailbroken in one specific cybersecurity scenario; Anthropic described the jailbreak as narrow and non-universal and called the situation a misunderstanding, but complied within hours, having no practical mechanism to filter by nationality in real time.

The directive established, for the first time, that a commercially deployed frontier AI model can be treated as an export-controlled asset in the same regulatory category as advanced semiconductors, with access revocable overnight by executive action for any foreign national regardless of contract or employer. There is also a structural irony in Anthropic’s position: the company built its public identity around the argument that it was developing something genuinely dangerous, and the government used exactly that framing as the justification for the shutdown.

→ Read the full article at AI World

2. Where is European AI compute heading?

A five-year scenario published as Europe 2031 projects the continent holding around 5% of global AI compute capacity by 2031, with a temporary rise to 8% in 2028, against a US share of roughly 70%. Within Europe the distribution is uneven: France, if it executes on current plans including the Softbank data centre and Mistral Campus, would account for around 2.4 GW and 35% of continental capacity, as much as Norway, Germany, Portugal, and Sweden combined.

The 101 publicly known compute projects across the continent are heavily concentrated at the top, with the ten largest responsible for more than half of total planned capacity. The Fable 5 ban, which in the Europe 2031 scenario only arrives in 2029, arrived instead this week, making the dependence the report describes less a projection than a present condition.

→ Read the full analysis on AI World

3. This week's most notable AI research papers

This week’s selection maps the path to a European frontier model across four dimensions: the federated infrastructure required to pool distributed compute, the AI Act’s systemic-risk obligations for models trained above 10^25 FLOP, the liability architecture that federated training creates, and DiLoCo’s demonstration that distributed training across poorly connected machines can reach standard training quality while reducing inter-machine communication by a factor of 500. The hardware turns out to be the tractable part. Who governs the model and who is accountable when something goes wrong remains the open problem.

→ Read the full paper summaries on AI World

4. Three policy briefs on code shipping gaps, open source strategy, and AI safety evaluation

Three reports map the gap between what AI produces and what we can verify. An NBER paper finds coding activity up 180% at the commit level but shipped software up only 30%, because the human work of reviewing and merging does not compress.

An OECD paper for the G7 finds open-weight models at 90% of closed-model performance for a fifth of the price, with measurable GDP returns. The second International AI Safety Report finds that pre-deployment tests no longer predict real-world model behaviour, and that developers shipped 2025 models with extra safeguards because testing could not rule out bioweapon uplift risk.

→ Read the full analysis on AI World Linkedin page

5. Startup spotlights of the week

This week’s selection showcases the application of AI to structural gaps in enterprise SAP migration delivery, intellectual property protection for SMEs, and field-level precision agriculture.

🇩🇪 Qorelo: Berlin-based, founded 2025 by Nicholas Torabi, Louis Schmidlin, and Marino Kurtović. Qorelo automates the functional delivery layer of SAP ECC to S/4HANA migrations, handling custom ABAP code reading, fit-gap analysis, business requirement conversion, and documentation, integrating directly with SAP S/4HANA and maintaining a continuously updated transformation record. Raised $3.5 million in a seed round co-led by HPI Ventures and Caesar Ventures, five months after launch.

🇸🇪 Lightbringer: Sweden-based, founded 2023 by Dominic Davies, Markus Andreasson, and Ola Wassvik. Lightbringer combines agentic AI with licensed attorney oversight in a single workflow for patent filing, covering invention capture, claim drafting, prosecution tracking, and portfolio management across multiple jurisdictions, reducing filing timelines from around two months to a few days. Raised $10 million in a Series A co-led by 6 Degrees Capital and Newion, June 2026.

🇨🇭 OneSoil: Zurich-based, founded 2017 by Sasha Yakovlev, Slava Mazai, and Usevalad Henin. OneSoil’s precision agriculture platform now includes AI Agronomist, a natural-language assistant that draws on accumulated satellite data and the company’s own analytical models to provide farmers with daily field-level insights on conditions, productivity changes, and emerging stress signals. Secured €1 million from existing investors.

What’s next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s round-up directly in your inbox and stay ahead of key developments.

Have a great weekend!

Gaia Cavaglioni

On behalf of the AI World Team

This week at a glance:

• Claude Fable 5 and the Mythos it hides: Anthropic's first public release of a Mythos-class model leads most major leaderboards, but ships with a routing layer that redirects high-risk queries, costs twice the Opus series, and moves to usage-credit billing on 22 June, making capacity once again the binding constraint.

• AI agents are writing open source code: agentic contributions to critical open source repositories grew from 2% to 10% of pull requests between October 2025 and May 2026, with human-authored contributions declining for two consecutive months since March, concentrating pressure on a small number of volunteer maintainers.

• Neura Robotics raises $1.4 billion: the largest funding round to date for an EU full-stack robotics company places a European firm at the top of a global ranking otherwise led by US players, though the open question remains deployment rather than financing.

• The real cost of AI: a review of the economics behind the 280-fold reduction in inference costs since 2022 finds that the gains rest on the US-China rivalry remaining neck and neck, with the inverted-U of innovation pointing downward if one side pulls decisively ahead.

1. Claude Fable 5 and the Mythos it hides

Anthropic has released Claude Fable 5, the first public version of a Mythos-class model, built on the same underlying architecture as the system kept behind Project Glasswing since April.

Fable 5 is effectively Mythos with a routing layer that transparently redirects queries touching cybersecurity, biology, chemistry, and related high-risk areas to Opus 4.8, triggering in under 5% of sessions on average and deliberately tuned to over-block while false positives are reduced. On most major leaderboards the model leads by a clear margin, with early deployments at Stripe, IMC, and Cursor each reporting substantial gains on complex, long-context tasks.

→ Read the full story at AI World

2. AI agents are increasingly writing open source software code

Agentic contributions to critical open source repositories, those identified by the Open Source Security Foundation as the backbone of projects including Node.js, npm, and Raspberry Pi, grew from 2% to 10% of all pull requests between October 2025 and May 2026, a sixfold increase in six months.

The rejection rate for AI-authored contributions has fallen alongside the growth, suggesting the additional volume is not uniformly lower quality, but the pressure is concentrating: human-authored pull requests have declined for two consecutive months since March 2026, and many of the most critical codebases are maintained by one or two individuals.

→ Read the full analysis on AI World

3. Neura Robotics: is the EU finally in the race?

Neura Robotics, based in Metzingen, Germany, has closed a $1.4 billion Series C led by Tether, with Qualcomm, Amazon, Nvidia, Bosch, Schaeffler, and the European Investment Bank among the backers, in what the company describes as the largest funding round to date for an EU full-stack robotics company.

The round is roughly 5.8 times the next largest European robotics round on record and places Neura alongside Skild AI at the top of the global Series C and below rankings, ahead of Figure's $1.0 billion. The capital is directed at serial production and Neura's training infrastructure, with the 4NE-1 humanoid still being prepared for volume production rather than shipping at scale today.

→ Read the full analysis on AI World

4. This week's most notable AI research papers

This week's selection examines the economics behind the 280-fold reduction in frontier AI inference costs since 2022, tracing the role of US-China competitive dynamics, export-control-driven algorithmic innovation, and the inverted-U of innovation that suggests the gains depend on the rivalry remaining evenly matched.

→ Read the full paper summaries on AI World

5. Two policy briefs on education and open source

A World Economic Forum Insight Report places the risk of AI in education not in teacher replacement but in the removal of productive struggle: one study it cites found students who used an AI tool scored worse than peers who never had it once it was withdrawn, and the report organises the conditions for durable learning gains into four interdependent levels covering data governance, institutional capacity, pedagogical practice, and inclusive access.

An OECD paper for the G7 presidency finds that open-weight models are now only months behind closed ones and represent four fifths of downloads on Hugging Face despite being two thirds of licensed models, with a 10% rise in global open-source AI work tracking a 0.5% rise in long-run GDP across the 33 countries studied, while noting that openness at the model layer does little for real autonomy when compute and chips remain concentrated.

→ Read the full analysis on AI World Linkedin page

6. Startup spotlights of the week

This week's selection showcases the application of AI to structural gaps in clinical documentation, enterprise knowledge accumulation, and industrial condition monitoring.

🇳🇱 OurMind: Netherlands-based, co-founded by Paul Koning, Fredrik Gustafsson, and Marco Ferraz. OurMind records clinical consultations as ambient audio, generates structured medical notes pushed directly to patient records via EHR integration, and handles surrounding administrative tasks including referral letters, sick notes, and inbox triage, currently live at over 300 general practices and 14 hospitals across the Netherlands. Raised €2.1 million in a round led by 4impact Capital.

🇬🇧 Zaro: London-based, co-founded by Michael Bajwa and Qian Zheng. Zaro provides a shared context layer that sits below individual AI tools, storing company data, decisions, workflows, and operational history in one persistent workspace so that agents and applications can access the same accumulated knowledge rather than starting from an empty context. Raised $5.1 million in a pre-seed round led by Cherry Ventures.

🇫🇮 Rotomate: Finland-based, founded 2024 by Mikko Kuusisto and Dr Jesse Miettinen. Rotomate ingests vibration measurements alongside maintenance records and operational history to produce explained alerts, root-cause analysis, and prioritised work-order suggestions, with outputs tailored by role across vibration analysts, plant managers, and maintenance planners. Raised €2.1 million in a pre-seed round led by Kvanted.

What’s next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s round-up directly in your inbox and stay ahead of key developments.

Have a great weekend!

Gaia Cavaglioni

On behalf of the AI World Team.

This week at a glance:

• The real obstacle to AI in European businesses is not the technology: Eurostat data across 157,000 firms finds that 70% of companies that considered AI but did not proceed cite a lack of relevant skills, a barrier distributed almost evenly across company sizes, pointing to an organisational integration problem rather than an access one.

• The cheap way to use AI is the expensive one: three converging studies find that productivity gains from AI concentrate around complementarity, while substitution produces thinner returns and invisible downstream penalties estimated at millions per year for large organisations.

• Three papers on memory, agents, and hardware: a Carnegie Mellon paper embeds sleep-like consolidation into language models, a systematic EU compliance mapping finds that high-risk agents with untraceable behavioural drift cannot currently be placed on the EU market, and a Hugging Face and CEPS audit finds that more than 40% of the top downloaded models disclose no recoverable signal of the chips that trained them.

• Startup spotlights: Bayshore, Gradient Labs, and Gigaton each closed rounds this week across legal and compliance automation, regulated financial operations, and AI-driven industrial process control.

1. The real obstacle to AI in European businesses is not the technology

Eurostat's 2025 survey of 157,000 European businesses finds that 20% used at least one AI technology, up from 8% in 2023, but the adoption gap between large firms (55%) and small ones (17%) has widened from 24 to 38 percentage points in two years. Among companies that considered AI but did not proceed, 70.3% cite a lack of relevant skills as the primary barrier, a figure that varies only marginally by company size: 70.9% among small firms, 65.1% among large ones.

The McKinsey Global Institute's parallel finding for the US market, that fewer than 40% of organisations investing in AI report tangible impact on results, and that the gap is explained by organisational rather than technological factors, suggests a consistent mechanism: the bottleneck is not access to tools but the capacity to integrate them into workflows in a way that produces measurable output. If that mechanism holds in the European context, the widening adoption gap between large and small firms likely reflects a deeper and compounding gap in organisational readiness.

→ Read the full analysis at AI World

2. The cheap way to use AI is the expensive one

This week's selection examines the conditions under which AI actually generates productivity, from the complementarity findings of randomised trials to the hidden costs of miscalibrated output and the structural incentives that currently favour automation over augmentation.

→ Read the full papers summary on AI World

3. Three papers on memory, agents, and hardware

A Carnegie Mellon and University of Maryland paper embeds a sleep-like consolidation mechanism into language models, where once active memory fills the model compresses recent context into its weights rather than discarding it, with gains appearing specifically on tasks requiring multi-step reasoning rather than simple recall. A systematic compliance mapping from Piccadilly Labs and the Association of AI Ethicists finds that the EU AI Act regulates what an agent does rather than how it is built, with obligations triggered by external actions across GDPR, the Cyber Resilience Act, the DSA, and sector-specific rules; the paper's sharpest conclusion is that a high-risk agent whose behaviour drifts over time in untraceable ways cannot currently be placed on the EU market at all.

A Hugging Face and CEPS audit of the top 4,000 most downloaded models finds that more than 40% disclose no recoverable public signal of the hardware that trained them, that NVIDIA dominates but Chinese developers increasingly mix in domestic accelerators, and that serving infrastructure matters more than is disclosed: one model scored anywhere from roughly 0% to 85% on the same benchmark depending solely on which provider served it.

→ Read the full analysis on AI World Linkedin page

4. Startup spotlights of the week

This week's selection showcases the application of AI to structural gaps in legal and compliance operations, regulated financial workflows, and energy-intensive industrial process control.

🇩🇪 Bayshore: Munich-based, co-founded by Paul F. Welter, Philipp Wiegand, and Erik Krauter. Bayshore encodes regulations and internal policies as deterministic, machine-readable logic built by specialist lawyers, enabling AI agents to resolve routine compliance cases automatically and route complex ones to human reviewers with full audit trails, EU-hosted and ISO 27001 certified. Raised $8 million in a seed round led by Earlybird Venture Capital.

🇬🇧 Gradient Labs: London-based, founded 2023 by Dimitri Masin, Neal Lathia, and Danai Antoniou. Gradient Labs has built an agent platform for regulated financial operations, handling card replacements, account freezes, fraud triage, complaint resolution, and lending workflows within audited constraints, targeting the complex tier of customer operations that most AI agents do not reach. Raised $26 million in a Series A extension led by Octopus Ventures and CommerzVentures, bringing total funding to approximately $42.8 million, June 2026.

🇬🇧 Gigaton: London-based, founded 2020 as Carbon Re, a joint spinout from UCL and the University of Cambridge. Gigaton's self-learning control platform replaces rather than overlays existing industrial control infrastructure, autonomously adjusting fuel mix, kiln speed, and oxygen levels based on simulation and predictive models, continuously retraining on live production data. Raised $26 million in a Series A led by Plural, bringing total funding to over $35 million, June 2026.

What's next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s round-up directly in your inbox and stay ahead of key developments.

Have a great weekend!

Gaia Cavaglioni

On behalf of the AI World Team.

This week at a glance:

• Cybench, a yardstick AI outgrew: Claude Mythos Preview has solved every task on the Cybench benchmark at first attempt, two years after GPT-4o managed 12.5%, on a benchmark Anthropic’s own system card had already declared no longer useful for tracking capability progression.

• Three policy briefs on AI exposure and European chips: an OECD capability index remaps occupational automation risk, a Bruegel analysis questions whether Europe's hardware advantage is structural or contingent, and a Microsoft Institute report finds Norway, Ireland, and France each outpacing the United States on AI adoption as a share of working-age adults.

• Legal AI past the productivity question: five studies confirm productivity gains of 38% to 140% in legal tasks, while also finding residual hallucination rates in validated vendor tools, measurable bias amplification in predictive systems, and a regulatory framework that addresses the technology one step downstream of where it is being deployed.

• Startup spotlights: NEX Health Intelligence, Certo, and Lucis each closed rounds this week across hospital infection prevention, consumer goods regulatory compliance, and preventive blood biomarker analysis.

1. Cybench, a yardstick AI outgrew

Cybench went from 12.5% to 100% in two years. Claude Mythos Preview solves every adversarial cybersecurity task at first attempt, tasks that took human expert teams nearly 25 hours, after the Opus 4.x line crossed the 60s, 70s, and 90s in successive releases. Anthropic had already declared the benchmark saturated and gated Mythos behind Project Glasswing. The eval outlived its usefulness just as the capability crossed into territory that matters.

→ Read the full analysis at AI World

2. Three policy briefs on exposure, chips, and who actually uses AI

The OECD’s new AI Capability Gap Index ranks billing clerks and typists as the most exposed occupations, not actuaries or mathematicians, because full occupational profiles include social judgement that current systems do not reach. Bruegel flags the same structural risk in European chip strategy: an industrial consortium built on national allocation rather than comparative advantage repeats the fragmentation problem.

A Microsoft Institute report adds the sharpest corrective: Norway, Ireland, and France each outpace the United States on AI adoption as a share of working-age adults, with the US ranking 21st globally at 31.9%.

→ Read the full analysis on AI World Linkedin page

3. Legal AI past the productivity question

This week's selection traces two converging lines: the empirical record on AI in legal practice, from randomised trials through vendor evaluation to distributional effects, and the governance gap between where the EU AI Act draws its high-risk boundary and where legal AI is actually being deployed.

→ Read the full paper summaries on AI World

4. Startup spotlights of the week

This week’s selection explores the use of AI in addressing structural gaps in hospital infection control, consumer goods regulatory compliance and preventive health monitoring.

🇬🇧 NEX Health Intelligence: UK-based, co-founded by Dr Ashleigh Myall and Dr Chang Ho Yoon. NEX has built an AI platform for infection prevention and control that operates across risk intelligence, outbreak detection, and automated case review surveillance, integrating with existing hospital infrastructure and compliant with NHS DTAC and related standards. Raised €1 million in a pre-seed round led by Brighteye Ventures, May 2026.

🇫🇷 Certo: Paris-based, co-founded by Bastien Deliège-Coste and Jean Duquenne. Certo's domain-specific AI agents cross-reference cosmetic and consumer goods products against a proprietary regulatory database covering the EU, US, China, South Korea, Japan, and Latin America, with five modules covering the full compliance process - from ingredient approval to market entry documentation. Raised $4 million in a seed round led by Daphni, with Entrepreneurs First, Motier Ventures, and Transpose Platform participating, May 2026.

🇫🇷 Lucis: Paris-based, founded 2025 by Maxime Berthelot and Baptiste Debever. Lucis analyses more than 110 blood biomarkers through a network of over 400 certified partner laboratories across France, the UK, Ireland, and Portugal, delivering AI-powered personalised health guidance reviewed by healthcare professionals, with longitudinal retesting compounding the precision of recommendations over time. Raised $20 million in a Series A led by Singular, with General Catalyst and Y Combinator participating, bringing total funding to $28 million.

What’s next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s round-up directly in your inbox and stay ahead of key developments.

Have a great weekend!

Gaia Cavaglioni

On behalf of the AI World Team

This week at a glance:

• Gemini 3.5 Flash reprices the budget tier: Google’s new Flash model beats 3.1 Pro on coding and agentic benchmarks but arrives at three times the price of its predecessor, closing the gap with frontier models while opening a new one for developers who built on Flash’s cost-effectiveness.

• Small models, big infrastructure question: new research confirms that sub-billion parameter models now match previous-generation large ones on key tasks, and a heterogeneous SLM-plus-LLM architecture is emerging as the practical design for agentic systems.

• Three policy briefs on AI’s social and labour costs: a JAMA Summit report finds that most FDA-cleared health AI tools were never evaluated for outcomes, while ILO data shows women in high-income countries face nearly three times the top-band automation exposure of men.

• Startup spotlights: AVIAN, LawX, and 9amHealth each closed rounds this week across industrial fire prevention, legal operations infrastructure, and virtual chronic disease care.

1. Gemini 3.5 Flash reprices the budget tier

Google has released Gemini 3.5 Flash, which ranks fourth on the AI World aggregated leaderboard, just three points behind Claude Opus 4.7. It outperforms Gemini 3.1 Pro on coding and agentic benchmarks, processing more than 280 tokens per second. While the claim that it is cheaper than the frontier models it now outperforms is accurate against GPT-5.5 and Claude Opus 4.7, the comparison that matters most to developers is against its own lineage. At $1.50 per million input tokens and $9.00 per million output tokens, Gemini 3.5 Flash is three times more expensive than Gemini 3 Flash and six times more expensive than Flash-Lite.

This continues a pattern visible across the major labs: the era of subsidised inference is coming to an end, and this time the repricing reaches the volume tier where cost predictability was the primary value proposition. Google has indicated that Gemini 3.5 Pro is undergoing internal testing and is expected to be released next month.

→ Read the full article at AI World

2. Small language models and the architecture of on-device AI

This week's selection examines the maturing architecture of on-device AI, from battery constraints and model depth trade-offs to routing logic between small and large models, and the practical limits that still separate benchmark performance from real deployment.

→ Read the full paper summaries on AI World

3. Three policy briefs on AI's social and labour costs

A JAMA Summit report finds that over 1,200 AI tools hold FDA clearance and AI scribes are now present in most US health systems, yet most were never evaluated for their effect on health outcomes and most are not required to be. A new ILO paper finds that one in four workers worldwide has a job with some exposure to AI, but only 3.3% are in the top automation band, with women in high-income countries outnumbered by men in that band.

A Lancet Global Health Series paper examining six African countries finds that digital tools are already reshaping HIV prevention reach, but structural limits, broadband coverage, intermittent power, and criminalisation risks define where such services can be used.

→ Read the full analysis on AI World Linkedin page

4. Startup spotlights of the week

This week’s selection showcases how AI can be applied to address structural gaps in industrial fire prevention, legal and notarial operations, and employer-sponsored chronic disease management.

🇨🇭 AVIAN: Zurich-based, founded by Drew Hanover and Thomas Laengle. AVIAN deploys continuous thermal camera monitoring across critical industrial assets such as bearings, conveyors, and electrical cabinets, with AI models trained per asset to detect heat drift before it reaches failure or ignition point, already operating profitably on a bootstrapped basis for two years before its first external raise. Raised $2.6 million in a pre-seed round led by Founderful, May 2026.

🇩🇪 LawX: Berlin-based, founded by Dr Norman Koschmieder. LawX consolidates case management, document processing, billing, and calendar coordination into a single platform built to German and European legal and notarial standards, targeting the administrative infrastructure surrounding legal work rather than research or drafting assistance. Raised €7.5 million in a seed round led by Motive Partners, with WENVEST Capital, xdeck, and SIVentures participating, May 2026.

🇺🇸 9amHealth: San Diego-based, founded 2021 by Frank Westermann, Anton Kittelberger, Paul Geevarghese, and Bernhard Schandl. 9amHealth delivers virtual specialty care for obesity, type 2 diabetes, hypertension, and high cholesterol through AI-assisted clinical workflows, at-home laboratory testing, and remote monitoring, contracted directly by employers and pharmacy benefit managers. Raised $26 million in a Series B led by Define Ventures, May 2026.

What's next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week's round-up directly in your inbox and stay ahead of key developments.

Have a great weekend!

Gaia Cavaglioni

On behalf of the AI World Team

This week at a glance:

• Hugging Face reaches one million datasets: the fastest-growing category is now robotics and reinforcement learning, reflecting the shift from language modelling to embodied AI.

• Where the grid stops being a benchmark: new research finds that AI agents trained in simulation degrade sharply once deployed on real electricity networks, exposing a gap between benchmark performance and operational reliability.

• Two policy briefs on EU AI governance and labour exposure: a Bruegel brief makes the case for a DSA independent agency, while EPC data locate Europe's AI adoption problem in depth, not speed.

• Startup spotlights: Pillar, NanoStruct, and Tolemy Bio each closed rounds this week across construction operations, food pathogen detection, and cell therapy process optimisation.

1. Hugging Face reaches one million datasets

The Hugging Face platform has surpassed one million datasets, marking a significant milestone in the growth of open data as a shared infrastructure rather than a supplementary resource. The fastest-growing categories are robotics and reinforcement learning, reflecting a shift from language modelling to embodied AI systems that learn through demonstrations, sensor streams and interaction with the physical world.

The most referenced datasets include generative AI prompt collections, the FineWeb open-data corpus, and Anthropic’s RLHF and red-teaming dialogue data.

→ Read the full article at AI World

2. This week's most notable AI research papers

This week’s selection examines AI applications in energy distribution networks, from renewable forecasting and leak detection to voltage control, and the gap between benchmark performance and real-world grid operations that governance frameworks have yet to close.

→ Read the full paper summaries on AI World

3. Two policy briefs on EU regulation and the labour market

A Bruegel policy brief argues that delegating DSA enforcement to an independent agency, modelled on AMLA and spun off from DG-CONNECT, would remove the lever that makes US trade pressure on the Commission effective.

Meanwhile, an EPC brief applying a two-axis exposure and complementarity framework to EU-SILC microdata across 26 member states reveals that the group with the most adverse risk profile is concentrated among women, mid-career workers and middle-income earners - the very people who fund the European welfare state.

→ Read the full analysis on AI World Linkedin page

4. Startup spotlights of the week

This week’s selection showcases how AI can be used to address productivity gaps in construction project management and food safety compliance testing, as well as optimise processes in cell therapy manufacturing.

🇮🇹 Pillar: Milan-based, founded 2025. Pillar builds an AI-powered operating system for construction contractors, connecting accounting software, bank feeds, and on-site channels to extract structured records and identify margin discrepancies in real time without requiring workflow changes. Raised €12 million in a seed round co-led by Earlybird Venture Capital and Base10 Partners, bringing total funding to €15.2 million.

🇩🇪 NanoStruct: Würzburg-based, founded 2021 as a spin-off of Julius-Maximilians-Universität Würzburg. NanoStruct has developed nanostructured optical sensor chips that detect Listeria and Salmonella in food production within hours rather than the two-to-three days required by standard culture methods, using machine learning to classify pathogens from spectral readouts. Raised €2.6 million in a seed round co-led by HTGF, Bayern Kapital, and AUXXO Female Catalyst Fund.

🇬🇧 Tolemy Bio: Cambridge-based, founded 2024. Tolemy Bio’s Orbit platform combines bench data ingestion, virtual cell models, and AI research agents to recommend process and media adjustments for cell biology teams, with two commercial partnerships live in cell and gene therapy manufacturing. Raised €1.4 million in pre-seed funding led by Norrsken Evolve.

What’s next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s round-up directly in your inbox and stay ahead of key developments.

Have a great weekend!

Gaia Cavaglioni

On behalf of the AI World Team.

This week at a glance:

• Anthropic rents Colossus 1 outright: Anthropic has signed for the full capacity of SpaceX’s Tennessee data centre, over 200,000 Nvidia GPUs and more than 300 MW of compute, with rate caps lifted for paying users from this month.

• ICLR 2026 mirrors NeurIPS: Chinese institutions account for around 40% of poster-paper affiliations and 30% of orals; US institutions hold the inverse split, with European, Singaporean, and South Korean contributions sitting close behind.

• EU AI job postings, decentralised but uneven: Germany leads in absolute volume with around 75,000 postings in 2025, but Portugal, Poland, and Luxembourg lead on share, and California alone matches Germany, France, and Italy combined.

• Military AI’s tempo problem: this week’s selection charts how AI-enabled drone strike rates in Ukraine have moved from 10–20% to 70–80%, alongside the governance lag the figure exposes.

• Three policy briefs on AI’s economic footprint: the ECB, OECD-WEF, and Cedefop each map a different layer of adoption, with distribution rather than diffusion emerging as the binding question.

• Startup spotlights: Featherless.ai, QuoIntelligence, and CodeWords each closed rounds this week across serverless open-model inference, sovereign threat intelligence, and SME automation.

1. The Anthropic compute deal that nobody saw coming

Anthropic has signed an agreement with SpaceX to use the full capacity of its Colossus 1 data centre in Tennessee: over 200,000 Nvidia GPUs and more than 300 MW of compute, taking effect within the month. For paying users the consequences land immediately, with five-hour rate caps lifted, peak-hour restrictions removed, and Claude Code rate limits doubled on paid plans.

The deal sits on top of a multi-gigawatt agreement with Amazon, a 5 GW deal with Google and Broadcom, and a $30 billion Azure partnership with Microsoft and Nvidia, which together answer Dario Amodei’s earlier acknowledgment of strain on infrastructure during peak hours. The contract carries an unusual clause giving SpaceX the right to reclaim compute if Claude engages in actions that harm humanity, a framing that sits awkwardly alongside Musk’s earlier public criticism of the company but is consistent with the safety thesis Anthropic’s research agenda actually requires.

→ Read the full article at AI World

2. ICLR 2026: where AI research is actually published

ICLR 2026 reproduces the geographic pattern visible at NeurIPS 2025. Chinese institutions account for roughly 40% of author affiliations on poster papers and 30% on orals, the latter a status only 4% of papers receive; US institutions sit at the inverse split, with two of three Outstanding Paper Awards this year going to American institutions and the third to a European collaboration between RPTU Kaiserslautern-Landau, ETH Zürich, and the Max Planck Institute.

Outside the US-China axis, contributions from Singapore and South Korea are competitive with the EU-27 taken as a whole, and the underweighting of Chinese research in Western coverage continues to track the underrepresentation of Chinese model releases such as Kimi K2.6 and DeepSeek V4 against GPT-5.5 and Claude Opus 4.7. The ICLR 2026 analysis uses direct affiliations cited in the papers themselves rather than OpenReview registrations, a tighter methodology than that used for the NeurIPS dataset.

→ Explore the interactive chart on AI World

3. 27 PACES: WHAT AI JOB POSTINGS TELL US ABOUT THE EU

Drawing on Lightcast's categorised online job postings, the data shows Germany leading the EU in absolute AI job postings every year from 2018 to 2025, with France and Poland second and third by 2025. France moved from around 9,000 postings in 2018 to over 60,000 in 2025; Poland went from around 550 to roughly 56,000 over the same period.

On AI postings as a share of total postings, the ranking inverts: Portugal, Poland, and Luxembourg lead, while Germany sits behind despite its volume. The aggregate gap with the US remains structural - California alone posted as many AI roles in 2025 as Germany, France, and Italy combined, and the US share of AI postings overall sits at 2.5% against roughly 1% in many EU countries.

→ Read the full analysis on AI World

4. This week’s most notable AI research papers

This week’s research examines the speed at which AI is reshaping warfare, from a 70–80% drone strike rate in Ukraine to a doubling of global military AI spending, and the institutional and ethical frameworks struggling to keep pace.

→ Read the full paper summaries on AI World

5. Three policy briefs on AI’s economic and labour footprint

The ECB has found that the use of AI by euro area workers has risen from 26% to 40% in a year, with firm productivity increasing by 4% and no adverse effect on employment. However, euro area payments for US-based IP have increased fivefold in a decade, and almost 30% of European unicorns founded between 2008 and 2021 have relocated abroad. An OECD-WEF white paper surveying 167 foresight practitioners across 55 countries finds AI used by two-thirds, but only 53% of public-sector respondents feel skilled in it against 93% in the private sector.

A Cedefop analysis of 46,213 workers across 29 countries found that digitalisation does increase training demand, but the benefits mainly accrue to male, urban, highly educated workers in non-routine roles, complicating the EU’s “human-centred digital transition” framing.

→ Read the full analysis on AI World Linkedin page

6. Startup spotlights of the week

This week’s selection showcases the application of AI to structural gaps in open-model inference economics, EU cyber threat intelligence, and SME automation.

🇺🇸 Featherless.ai: founded in 2023 by the team behind RWKV, the attention-free architecture that became the first foundation-model project hosted by the Linux Foundation. Featherless is a serverless inference layer giving developers access to over 30,000 open language, vision, and audio models through one API, with subscription-tier pricing rather than per-token billing. Raised $20 million in a Series A co-led by AMD Ventures and Airbus Ventures, with BMW i Ventures, Kickstart Ventures, Panache Ventures, and Wavemaker Ventures participating, on top of a $5 million seed in 2025.

🇩🇪 QuoIntelligence: Frankfurt-based, founded 2020 as a spin-off from QuoScient’s intelligence team. QuoIntelligence pairs an AI platform, Mercury, with European analysts producing localised threat-intelligence briefs for the roughly 160,000 EU entities newly in scope under NIS2 and the 22,000 financial firms covered by DORA, with a conversational analyst, KARLA, making the briefs queryable from board to SOC. All intelligence data is held inside the EU. Raised €7.3 million in a Series A led by Elevator Ventures and co-led by BMH, with eCAPITAL and Mercurius Private Equity participating, bringing total capital to roughly €12.3 million.

🇬🇧 CodeWords: London-based, founded 2023. CodeWords’ agent, Cody, studies a business’s operations and autonomously builds and executes automations across more than 1,000 integrations including Notion, HubSpot, Slack, WhatsApp, and Gmail, writing the underlying Python and maintaining the workflows on schedules and triggers without manual intervention. The product targets the 38-point AI adoption gap between large EU enterprises (55%) and small ones (17%) flagged in Eurostat 2025 data. Raised $9 million in a seed round led by Visionaries, with Firstminute Capital, Sequoia, Illusian, and angels participating.

What’s next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s round-up directly in your inbox and stay ahead of key developments.

Have a great weekend!

Gaia Cavaglioni

On behalf of the AI World Team

This week at a glance:

• Opus 4.7, second best by design: Anthropic’s new flagship is an incremental step on public benchmarks, and the company is open about why, its cybersecurity capabilities have been deliberately reduced to test safeguards before a Mythos-class release.

• Three briefs, one frame: ILO, Carnegie California, and the JRC each draw a cleaner line between what AI changes, what democracies inherit, and what regulation has historically scaled rather than slowed.

• China’s embodied AI stack: three rounds in week 17 mapped the brain, the factory, and the home of embodied AI, with $455 million to Tars marking the largest embodied AI round in Chinese history.

• AI in architecture: five papers this week show the gap between idea and workflow has collapsed from 62 years to 2.5, while the gap between prototype and practice has not.

• Startup spotlights: four companies applying AI to construction compliance, building renovation, AEC workflows, and renewable site selection across Japan, Germany, and France.

Opus 4.7, second best by design

Anthropic has launched Claude Opus 4.7 with modest gains over 4.6, a six-thousandths lead on the AI World aggregated Leaderboard, two points on LMArena’s text category, and 0.6 on LiveBench. The company is unusually explicit about the reason. Opus 4.7 is a testbed for the automated safeguards Anthropic hopes to apply to Mythos, a more capable model whose autonomous zero-day discovery capabilities have kept it restricted to roughly 50 vendors under Project Glasswing. Cybersecurity capabilities were deliberately reduced during training. The result is a generally available flagship that openly admits it is not the frontier, and a dual-track strategy that separates the commercial product from the actual state of the art.

→ Read the full analysis on AI World

Three policy briefs reframing AI’s impact

The ILO clarifies that AI exposure measures what technology could technically do to tasks today, not whether firms will adopt it, whether adoption will be profitable, or whether jobs will survive the workflow redesign that follows. Carnegie California maps four intersection points between AI and democracy, elections, citizen deliberation, government services, and social cohesion, against a backdrop where autocracies outnumber democracies for the first time in twenty years. The JRC’s Science for Policy Brief traces the EU’s decade of AI policy through three phases, exploration, transition, and maturation, and argues that regulation historically scaled aviation, pharma, and finance rather than slowing them. Together, the three briefs draw a cleaner line between what AI changes and what it inherits.

→ Read the full analysis on AI World Linkedin page

AI investments of the week

Three Chinese rounds defined last week’s embodied AI activity, together mapping the full stack from reasoning to deployment. 🇨🇳 Tars closed a $455 million Pre-A, the largest embodied AI round in Chinese history, built around the first large-scale Vision-Language-Tactile-Action dataset. 🇨🇳 X Square Robot raised $276 million in a Series B led by Xiaomi Technology with ByteDance, Meituan, and Alibaba, targeting humanoid robots for household use. 🇨🇳 GigaAI raised for its 4D world models spanning autonomous driving and embodied robotics. With China’s workforce in decline since 2011, robotics is increasingly a labour policy, not a technology bet.

→ Explore the interactive chart on AI World

Design architecture and AI research papers

Five papers trace AI’s uneven penetration of architecture and design: the gap between a technique appearing in the literature and its first architectural application has collapsed from 62 years to 2.5, but workflow integration, expert productivity, and stylistic diversity are all moving in less encouraging directions.

→ Read the full paper summaries on AI World

Startup spotlights of the week

This week’s selection highlights the application of AI to structural gaps in construction compliance, building decarbonisation, AEC workflow automation, and renewable energy permitting.

• 🇯🇵 Mign: Japan-based, founded 2022. A vertical agentic AI platform for real estate and construction, covering compliance review, legal compilation, and architectural visualisation. Raised ¥230 million (approximately $1.5 million) in a pre-Series A in April 2026, led by Angel Bridge.
  

• 🇩🇪 Meinhaus.digital: Germany-based. An AI platform combining software with architect and energy consultant advice, helping property owners plan renovations under the updated EPBD’s national renovation plan requirements. Raised €1 million in a pre-seed round in April 2026 from L-Bank Baden-Württemberg.

• 🇩🇪 Conxai: Germany-based, founded 2020. A no-code agentic AI platform for the AEC sector, pulling together site photographs, sensor feeds, and project documents for tender analysis and site monitoring. Raised €5 million in a Series A in April 2026, led by BayBG Venture Capital and Capricorn Partners, bringing total funding to approximately €7.7 million.

• 🇫🇷 Plume: Franco-American, co-founded 2024. Aggregates 150+ geospatial datasets and deploys AI agents on regulatory documents, letting renewable developers query territorial constraints through a natural language interface, addressing RED III permitting delays. Raised €3.9 million in April 2026, led by AENU.

What’s next

For more updates, research and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s round-up directly in your inbox and stay ahead of key developments.

Have a great week ahead!

Gabriel Rossi

On behalf of the AI World Team

Public and expert optimism on AI’s societal impact

A 2024 Pew Research Centre survey of 5,410 US adults and 1,013 AI experts, cited in the Stanford HAI 2026 AI Index, finds a consistent gap between public and expert optimism about AI’s societal impact. Expert optimism exceeds public optimism by roughly 30 percentage points in medical care, K–12 education, jobs, and the economy, and by approximately 28 points in arts and entertainment. In areas tied to trust and social cohesion, elections and the news media, both groups converge around low expectations, suggesting that shared scepticism, not disagreement alone, defines the current baseline. The data point to a structural communication gap, one that without structured dialogue will continue to develop in parallel, driven by separate reference points.

Source: Stanford HAI, AI Index Report 2026, Figure 9.2.1. Survey data: Pew Research Centre (2025).

Generative search and the restructuring of information access

The way information is retrieved is changing structurally. Where traditional engines return ranked lists that invite comparison, generative AI produces a single synthesised answer with no visible editorial trail.

Research finds these systems diverge substantially in which sources they surface and how they define authority (Kirsten et al., 2025), with audits showing consistent bias towards US commercial media over academic and non-Western outlets (Li & Sinnamon, 2024). AI summaries also tend to strip uncertainty language from their sources, presenting secure answers where research remain cautious (Huang et al., 2025). The downstream effects are measurable. Khosravi & Yoganarasimhan (2026) find AI Overviews reduce daily traffic to source pages by roughly 15%, creating a new story dictated by inherent bias and a new SEO framework.

Startup spotlights

🇸🇪 Validio (Sweden, founded 2019) monitors enterprise data pipelines in real time, using automated anomaly detection to catch quality failures before they reach production AI systems and generating auditable data lineage records relevant to regulatory reporting. In March 2026, the company raised a $30 million Series A led by Plural, bringing total funding to $47 million, to support expansion into the US, UK, and Northern Europe. EU AI Act Article 10, enforceable from August 2026, requires that training and testing datasets be representative, error-free, and covered by documented governance practices.

🇩🇪 Penemue (Germany, founded 2023) integrates with an organisation’s social media channels and internal systems via API, scanning content in real time across 89 languages to detect hate speech, threats, and harmful material, including coded language and emojis, using NLP and behavioural analytics. The platform allows users to remove content or escalate it for legal proceedings, and collaborates with police authorities and public prosecutors across Germany and Europe. In April 2026, the company raised over €1.7 million from investors including TION Health, Beyond Tomorrow, and Berlin Angel Fund.

🇫🇷 Omniscient (France, founded 2024) deploys a network of specialist AI agents, each assigned to a defined domain such as regulation, supply chain, or media narratives, to scan press, social media, the web, and internal corporate data, synthesising the results into a real-time executive briefing accessible in multiple languages. The platform is designed to give boards and C-suite teams a unified view of regulatory and reputational risk without relying on manual reporting cycles. In April 2026, the company raised €3.5 million in pre-seed funding led by Seedcamp, with participation from Plug and Play, Bpifrance, and others.

🇪🇸 EchoWi (Spain, founded 2025) addresses the measurement gap opened by generative AI search, running natural-language queries on repeatable schedules across eight AI engines to track how each model selects and ranks sources over time. The platform monitors visibility, share of voice, sentiment, and cited domains, comparing results against competitors in parallel. The company raised a €750,000 seed round in March 2026, led by Successful Fund I FCRE, to scale its infrastructure.

This week at a glance:

• OpenAI winds down Sora: After roughly two years, OpenAI is shutting down its text-to-video platform, citing compute reallocation priorities, a signal about the economics of generative video more broadly.

• EU funding gap widens at later stages: Data from Crunchbase shows the EU-to-US ratio of companies at Series D stands at roughly 1:18, suggesting a structural ceiling on European AI ambition.

• AI in psychiatric monitoring: Three papers this week chart the distance between experimental results and clinically validated, production-ready tools for mental health.

• Startup spotlights: Vienna-based Nyra Health, Paris-based Callyope, and Helsinki-based Bliss each closed rounds this week across neurological rehabilitation, psychiatric monitoring, and culturally adaptive therapy.

1. OpenAI Closes Sora - A Verdict on generative video economics

OpenAI has shut down Sora roughly two years after its debut. The official framing: a refocus on world simulation and robotics research, is plausible but incomplete. Video generation is computationally expensive; every rendered second consumes GPU resources that, reallocated to coding and reasoning workloads, generate substantially higher returns. Bill Peebles, Head of Sora, had already flagged generation limits last year due to chip constraints.

→ Read the full article at AI World

2. The European AI funding ceiling

New data illustrates the structural divergence between European and American AI funding ecosystems. At seed stage, the EU records approximately 856 classified companies against 2,605 in the US, a ratio of roughly 1:3. That gap compounds at each subsequent round. By Series D, only 8 EU companies appear in the data against 142 in the US, and EU representation becomes negligible from Series E onwards.

The pattern is consistent with a funding ecosystem that performs comparably to the US at the earliest stages but loses capacity to support high-magnitude growth bets as companies scale.

→ Explore the interactive funding chart on AI World

3. This week’s most notable AI research papers

This week’s selection examines four papers assessing the use of AI in psychiatric and mental health contexts, a field where promising signals and methodological limitations continue to coexist.

→ Read the full paper summaries on AI World

4. Startup Spotlights of the Week

This week’s selection highlights the application of AI to structural gaps in neurological rehabilitation, psychiatric monitoring, and culturally adaptive mental health care.

🇦🇹 Nyra Health: Vienna-based, founded 2021. Nyra’s myReha platform delivers AI-adaptive therapy for stroke, dementia, and brain injury patients across clinic and home settings, adjusting in real time to performance, fatigue, and progress. Deployed across 100+ rehabilitation clinics in the DACH region and reimbursed by 28 statutory health funds under Germany’s DiGA framework. Raised €20 million in a Series A in February 2026, led by Armira Growth.

🇫🇷 Callyope: Paris-based, founded 2023. Callyope built a speech foundation model trained on psychiatric data to monitor schizophrenia patients between the four-to-six week gaps in outpatient consultations. The model analyses hundreds of speech biomarkers via a patient smartphone app; clinicians receive continuous symptom scores and AI-drafted documents. Seven clinical studies are under way across 1,000+ patients. Raised €2.2 million in seed funding co-led by 360 Capital and Bpifrance Digital Venture.

🇫🇮 Bliss: Helsinki-based, originally founded in Albania. Bliss pairs users with licensed therapists using AI-driven cultural and linguistic profiling, and offers a digital companion modelled on a specific clinician’s approach to extend care between sessions. Early B2B traction at Raiffeisen Bank; 50+ therapists onboarded across 10 countries. Raised €232,400 in an angel round led by Keiretsu Forum, Finest Love VC, and Plug and Play.

What’s next

For more updates, research, and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s roundup directly in your inbox and stay ahead of key developments.

Have a great week ahead!

- Gabriel Rossi, on behalf of the AI World Team

This week at a glance:

• GLM-5 emerges as open-source leader: our latest data confirms GLM-5 as the top-performing open-source model on the AIW Model Leaderboard, marking a significant milestone in the development of high-performance open architecture.

• Germany’s AI ecosystem (2023–2025): an overview of how the German ecosystem has developed, highlighting shifts in research output, corporate adoption and investment resilience over the past three years.

• Research papers: our selection of five notable publications examining the latest technical milestones and evolving governance frameworks that are shaping the global AI field.

• Startup spotlights: a look at three pioneering companies that are using AI to overcome challenges in areas ranging from the circular economy to medical sector regulation.

1. GLM-5 is the top performer Open Source model in the AIW Model Leaderboard

Z.ai’s latest release, GLM-5, has set a new standard for open-weight performance, securing first place among open-source LLMs in the AI World Models Ranking.

With a score of 50 on the Artificial Analysis Intelligence Index, GLM-5 is the first open-source model to cross this critical threshold, narrowing the capability gap to just three points behind the proprietary leader, Claude Opus 4.6.

Powered by a 740B Mixture-of-Experts architecture, GLM-5 demonstrates outstanding performance in long-term reasoning and autonomous task execution.

Yet its transition to open weights at this scale introduces a strategic trade-off between infrastructure and API. Despite competitive per-token pricing, the 1 TB+ memory footprint required for local deployment may eclipse cost savings for many organisations.

→ Read the full analysis of GLM-5’s performance and deployment costs on AI World

2. Germany’s AI ecosystem: an overview from 2023 to 2025

From 2023 to 2025, Germany’s AI ecosystem was characterised by a distributed research footprint that increasingly aligned with concentrated financial support. Innovation remained geographically widespread, driven by a prestigious network of institutions including the Technical University of Munich, KIT, and Heidelberg country.

By 2025, Munich and Berlin had become the country’s leading cities for scientific output, with Stuttgart, Cologne and Frankfurt am Main also making significant contributions. Investment patterns increasingly reflected this research density, with capital flowing towards regions with high concentrations of innovation.

The German ecosystem is currently characterised by a narrowing gap between where knowledge is generated and where it is financed. Unlike other markets, which are defined by regional disparities, Germany is moving towards a self-sustaining cycle in which research excellence, distributed across the country, reliably precedes future capital allocation.

→ Read the full analysis on AI World

3. This week’s most notable AI research papers

This week’s research looks at the impact of AI on human capital through educational empowerment, historical integrity and medical morality.

Curious for more? Explore this week’s 5 most notable AI research papers to stay up to date on the latest insights and developments in the field.

4. Startup spotlights of the week

This week’s startup selection showcases the application of AI to systemic challenges in sustainable mobility, information integrity and the life sciences.

🇱🇹 Tingit has developed an AI-powered marketplace with the aim of streamlining the fragmented repair sector. The platform instantly identifies and estimates damage, connecting consumers with specialist craftsmen via proprietary algorithms. The Vilnius-based startup closed a €1.5 million seed funding round to expand across the EU.

🇦🇹 Flinn offers an AI-native platform that enables medical device and pharmaceutical manufacturers to automate labour-intensive regulatory workflows. The Vienna-based company recently secured $20 million in funding led by HV Capital to drive international growth and accelerate product development.

🇬🇧 Klaris provides an AI-driven solution aimed at alleviating the regulatory bottlenecks of the EU Medical Device Regulation (MDR). The platform carries out accurate gap analyses to ensure documentation is aligned with evolving standards. The start-up raised $1 million in pre-seed funding to expand its engineering capabilities and strengthen its presence in the European market.

What’s next

For more updates, research, and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s roundup directly in your inbox and stay ahead of key developments.

Have a great week ahead!

Gaia Cavaglioni
On behalf of the AI World Team

This week at a glance:

• Claude Opus 4.6, leader of the leaderboards: an analysis of the latest benchmark performance and what it means for the current lead in proprietary model competition.

• Mapping India’s AI ecosystem - a snapshot from 2023 to 2025: a multi-year data review tracking the evolution of regional research output and the shifting of strategic investment capital over a 24-month horizon.

• Weekly funding: Data & Cloud dominated the investment landscape this week, with companies securing more than $268 billion overall.

• Research papers: our selection of five notable publications exploring the latest technical and governance breakthroughs in the field.

• Startup spotlights: a look at three high-impact companies leveraging AI to address systemic challenges in healthcare, disinformation defence and the logistics of the circular economy for electric vehicles.

1. Claude Opus 4.6, leader of the leaderboards

Anthropic has officially released Claude Opus 4.6, a model that has quickly become the best performer on several industry-standard evaluation platforms, securing first place on the LMSYS Arena leaderboard. Its high scores in reasoning and multi-step workflows on objective benchmarks such as LiveBench suggest significant resilience against test-set contamination.

In addition to standard metrics, the model incorporates structural enhancements tailored to complex enterprise and research environments. These include expanded contextual capacity (a 1 million-token context window, currently in beta), adaptive thinking, and dynamic effort controls. Independent data from Artificial Analysis underscores the model’s superiority in agentic task metrics, especially in multi-domain problem solving.

As a result of these technical advancements, Claude Opus 4.6 now leads our AIW Model Leaderboard.

→ Explore the interactive chart for a detailed view

2. Mapping India's AI ecosystem: a snapshot from 2023 to 2025

From 2023 to 2025, India’s AI ecosystem was characterised by a growing synergy between academic output and financial support. Research activity maintained a robust national presence, driven by a diverse network of institutions, including the SRM Institute of Science and Technology, Saveentha University and the Vellore Institute of Technology.

By 2025, Madras and Delhi were leading the country in scientific output, with rising contributions also coming from Bengaluru, Chandigarh and Poona. Investment patterns increasingly mirrored this research density in 2025, with capital flowing towards regions with high concentrations of innovation.

The Indian ecosystem is currently characterised by a narrowing gap between where knowledge is generated and where it is financed. Unlike other markets with regional disparities, India is moving towards a self-sustaining cycle in which distributed research excellence reliably precedes future capital allocation across the subcontinent.

→ Read the full analysis on aiworld.eu

3. AI investments of the week - Week #5

Last week, global AI financing reached a significant $ 268 billion, maintaining the momentum of capital deployment across the broader AI ecosystem. High-profile funding rounds were led by🇺🇸 OpenAI $40 billion, 🇦🇺 Firmus Technologies $10 billion, 🇸🇪 IFS $3.42 billion, 🇨🇳 Inspur Cloud $2.88 billion among others.

Most active sectors in AI investment this week:

1. Data & Cloud: Scale (US), Crusoe AI Data Center (US), Databricks (US), Inspur Cloud (CN)

2. Education: OpenAI (US), Thinking Machines Lab (US)

3. Finance: xAI (US), Anthropic (US), Polymarket (US)

→ Explore the interactive chart for a detailed view of these trends

4. This week’s most notable AI research papers

This week’s research focuses on the impact of AI on human capital, the mitigation of bias through architectural reasoning, and the integration of AI as a strategic collaborator in complex decision-making processes.

Curious for more? Explore this week’s 5 most notable AI research papers to stay up to date on the latest insights and developments in the field.

4. Startup spotlights of the week

This week’s startup selection showcases the application of AI to challenges in sustainable mobility, information integrity and life sciences.

🇱🇺 R3robotics provides an AI-driven dismantling platform aiming at automating the recovery of vital components from electric vehicles once they reach the end of their lifecycle. The Luxembourg-based company recently secured €20 million in Series A equity and grants to expand its infrastructure.

🇬🇧 Refute offers an AI platform designed to identify and neutralise disinformation, providing response strategies to protect state and corporate institutions. The company secured £5 million in seed funding led by Amadeus Capital Partners to establish its technology as a core pillar of the modern information security ecosystem.

🇪🇸 Biorce delivers an AI-native platform that aims to optimise the structurally inefficient clinical trial process by automating the creation of trial protocols and identifying regulatory risks before they delay drug development. The company recently closed a $52 million Series A funding round, led by DST Global Partners, to accelerate its global expansion and product development.

What’s next

For more updates, research, and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s roundup directly in your inbox and stay ahead of key developments.

Have a great week ahead!

Gaia Cavaglioni
On behalf of the AI World Team

This week at a glance:

• Mapping China’s AI landscape 2023-2025: a multi-year analysis tracking the evolution of China’s research contributions and shifting investment priorities over a 24-month period.

• Weekly funding: Data & Cloud dominated the investment landscape this week, with companies securing more than $7.2 billion overall.

• Research papers: A selection of five notable publications exploring the latest technical and governance breakthroughs in the field.

• Startup spotlights: three companies that are using AI to solve critical challenges in renewable energy, digital defence and legal reasoning.

1. Mapping China's AI Landscape (2023–2025)

From 2023 to 2025, there was a discrepancy in China’s AI ecosystem between geographically dispersed research excellence and concentrated capital flows. Research activity emerged from a diverse national footprint, with Beijing and Shanghai ranking among the top five cities for scientific output in 2025.

Investment patterns, however, followed a more selective trajectory. In 2025, funding remained primarily concentrated in well-established eastern and southern hubs, with Beijing, Shanghai and Shenzhen dominating the financial landscape.

While Beijing maintains a self-sustaining cycle of innovation and commercialisation, the Chinese ecosystem’s defining challenge is the disparity between regional centres, where high research output has yet to attract proportional capital.

→ Read the full analysis on aiworld.eu

2. AI investments of the week - Week #4

Last week, global AI financing reached a significant $7.2 billion, maintaining the momentum of capital deployment across the broader AI ecosystem. High-profile funding rounds were led by 🇺🇸 Databricks $1.8 billion, 🇸🇦 HUMAIN $1.2 billion, 🇬🇧 Fluidstack $450 million, 🇪🇨 Jelou.ai $10 million among others.

Most active sectors in AI investment this week:

1. Data & Cloud: Databricks (US), Fluidstack (GB), Upscale AI (US), Inferact (US), Zanskar (US), Cambio (US), Cosmos (US), Helin (NL), Legato AI (IL), Cashmere (US), Thunder Compute (US), Lucend (NL), Offerswap (FI)

2. Education: StepFun (CN), Playad.ai (US), Sparkli (CH), Taalentfy (ES), Boshi Artificial Intelligence (CN), Leracom AI (MX), LeMong (KR), Lila Sciences (US), ALTEK AI (NO), Sufra Technology (QA), Digital Interview AI (US)

3. Industry & Mobility: HUMAIN (SA), humans& (US), Baseten (US), Ricursive Intelligence (US), Haier New Energy (CN), Uni-Ubi (CN), Unbox Robotics (IN), Juwei Technology (CN), Feijiecos (CN), PowerEasy Technology (CN), Ruihe Electronics (CN), AIsphere (CN), Elyos AI (GB), GeneralMind (DE), Nexxa.ai (US), Sapience AI (US), Seven Things Group (CN), Ringg (IN), Aniai (US), Cimba.ai (US), Nucleus 4D (US), Sanmi Technology (CN), Voxelo.ai (GB), Embodied AI (CH), Quantum Thor (VN)

→ Explore the interactive chart for a detailed view of these trends

3. This week’s most notable AI research papers

This week’s research looks at the effect of bot behaviour on social interaction, the detection of subtle hate speech, and how autonomous reasoning layers evolve in LLMs.

Curious for more? Explore this week’s 5 most notable AI research papers to stay up to date on the latest insights and developments in the field.

4. Startup spotlights of the week

This week’s startup selection showcases how AI is being applied to critical infrastructure and professional service challenges in the fields of renewable energy, security and real estate law.

🇪🇸 TetraxAI provides an intelligent platform designed to streamline the due diligence process for clean energy assets by combining legal expertise with machine learning. The company recently secured €15 million in funding to improve its risk analysis capabilities and expand its market reach.

🇬🇧 Pallma offers a specialised security and control layer that is tailored to protect AI agents by providing real-time oversight of internal decision trees and intent. The London-based start-up secured $1.6 million in pre-seed funding to develop its AI-native protection platform and expand its technical team.

🇬🇧/🇺🇸 Orbital delivers an AI-native platform designed specifically for real estate law, merging spatial visualisation with property data to automate complex title reviews and boundary analysis. The company has raised $60 million in Series B funding to accelerate its expansion into the US market and establish new regional hubs.

What’s next

For more updates, research, and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s roundup directly in your inbox and stay ahead of key developments.

Have a great week ahead!

Gaia Cavaglioni
On behalf of the AI World Team

This week at a glance:

• Who powers open-source AI research? Our analysis of Hugging Face Daily Papers reveals the growing global diversity of the open-source community and identifies the key institutions driving frontier research.

• Mapping the US AI ecosystem (2023 - 2025): a longitudinal review of the US AI landscape tracking the evolution of research output and capital allocation over the past two years.

• Weekly funding: Cybersecurity emerged as the leader of the week, with companies securing more than $5.7 billion overall.

• Research papers: A selection of five notable publications to keep an eye on this week

• Startup spotlights: three companies applying AI to biological material analysis, the simulation of complex biological systems, and the structural requirements of the global energy transition.

1. Who powers open-source AI research? A look at Hugging Face daily papers

Academic AI research is increasingly built on open-source, and Hugging Face Daily Papers provide insight into the organisations shaping this ecosystem.
Our analysis of 6,524 papers published in 2025 reveals that researchers cited over 9,000 models and 3,200 datasets, with no single organisation dominating.

Notably, Big Tech is not at the centre: Microsoft is absent, while Google only ranks 20th. Impact is instead fragmented across a broad mix of international nonprofits, companies, academic labs, and individual contributors.

It is interesting to note that models are cited far more often than datasets (around four times as often); yet, evaluation frameworks like MTEB dominate dataset usage, accounting for 35.9% of all dataset references, highlighting how shared benchmarks shape research practices.

Overall, the picture that emerges is of a resilient, decentralised research ecosystem where leadership is distributed globally and driven by non-profits, academia and large technology companies in equal measure.

→ Explore the interactive chart for a detailed view

2. Mapping the US AI ecosystem (2023 - 2025)

Between 2023 and 2025, deep technical breakthroughs shaped the US AI landscape, attracting major investment. The East Coast led the way in terms of research output, with Boston and Washington, D.C. ranking among the top five cities in this respect in 2025.

Although research activity was spread across Atlantic and Pacific hubs, investment patterns were far more concentrated. In 2025, the West Coast remained the dominant funding centre, reflecting a clear preference for regions where research can be rapidly translated into market-ready products.

While other regions struggle to bridge the gap between research outputs and funding, the defining challenge for the US ecosystem is the outsized influence of the West Coast, which is increasingly setting the direction for AI development nationwide.

→ Read the full analysis on aiworld.eu

3. AI investments of the week - Week #3

Last week, AI companies raised over $5.7 billion globally, signalling ongoing investment in a variety of AI applications.

Leading global funding rounds included 🇺🇸 Skild AI $1.4 billion, 🇩🇪 Parloa $350 million, 🇸🇬 Atome $345 million, 🇦🇺 Omniscient Neurotechnology (o8t) $13.37 million, 🇳🇬 Terra Industries $11.75 million among others.

Most active sectors in AI investment this week:

1. Cybersecurity: Playlist (US), Armadin Security (US), Terra Industries (NG), CloudSEK (SG), Fencer (US), Valent Projects (GB), Phinge (US)

2. Data & Cloud: Etched.ai (US), ClickHouse (US), GovDash (US), Seasats (US), Fractile (GB), Klearly (US), Elyos AI (GB), SkyFi® (US), FinOpsly (US), Coxwave (KR), AlphaBitCore (US), Greenphard Energy (JP), EggNest.ai (US), Cumulus Labs (US), Gravity GTM (US), StrateSea Technology (US)

3. Industry & Mobility: Skild AI (US), Harmattan AI (FR), X Square (CN), Mytra (US), osapiens (DE), Higgsfield (US), Hydrosat (US), WitnessAI (US), Manifold AI (CN), Hexalog (IN), RARA Factory (IT), AVES Reality (DE), Cimba.ai (US), Luna Systems (IE), Wanyigui Technology (CN), Spot Ship (GB), Meet Caria (US), Overwatch Imaging (US), CleanHaus (US), OrbiSharp Robotics (US)

→ Explore the interactive chart for a detailed view of these trends

4. This week’s most notable AI research papers

This week’s research highlights the systemic impacts of automation on labor, the security of information integrity, vulnerabilities in agent evaluation, and the longitudinal analysis of global climate discourse.

Curious for more? Explore this week’s 5 most notable AI research papers to stay up to date on the latest insights and developments in the field.

5. Startup spotlights of the week

This week’s startup selection showcases the application of AI to critical infrastructure challenges in renewable energy financing, non-invasive biological imaging and generative drug discovery.

🇩🇪 Cloover: provides a unified AI-driven operating system that integrates workflow management and financing for the distributed energy economy, thus facilitating inclusive financing for green technology adoption. The company recently secured $22 million in Series A equity funding to accelerate its platform’s deployment.

🇺🇸🇮🇱 Converge Bio: develops an LLM stack of biological foundation models fine-tuned to understand molecular and multi-omics data. The platform can design novel antibodies, predict developability issues, and simulate patient-level responses. The startup recently secured a $25 million Series A round to enhance its model suite.

🇩🇪 Orbem: combines accelerated MRI hardware with deep neural networks to examine biological materials at production-line speeds without invasive probing. Its Genus platform enables real-time fertility detection in poultry and is also developing applications for non-invasive human healthcare diagnostics. The company recently raised €55.5 million in a Series B funding round to fuel its global expansion and R&D.

What’s next

For more updates, research, and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s roundup directly in your inbox and stay ahead of key developments.

Have a great week ahead!

Gaia Cavaglioni
On behalf of the AI World Team

This week at a glance:

• The EU AI ecosystem 2023–2025: mapping the widening gap between world-class research output and late-stage capital deployment.

• Weekly funding: Data & Cloud infrastructure takes the lead, with AI-driven enterprises securing $1.1 billion in new capital.

• Notable research: a curated review of five publications defining the current trajectory of AI transparency and efficiency.

• Startup spotlights: three companies using AI to solve complex challenges in supply chain transparency and ESG compliance, biodiversity intelligence, and preventive clinical screening.

1. The EU AI ecosystem 2023–2025: mapping the gap between research and capital

Analysis of the European Union’s AI landscape from 2023 to 2025 reveals robust expansion of the region’s ecosystem. However, the 2025 landscape shows that, while AI knowledge production remains geographically diverse, it is also becoming increasingly polarised, with frequent misalignment between research output and capital deployment

Data indicates that innovation is not confined to usual economic heavyweights. Southern Europe maintains a strong presence, with Madrid, Rome and Milan ranking among the top five cities for research output in 2025, along with Paris and Munich. Nevertheless, capital remains concentrated in a few flagship hubs, primarily in France, Germany and the Nordic countries.

This divergence highlights a structural imbalance, thus giving rise to the risk of talent migration from research-intensive regions on one hand, and an opportunity for underfunded hubs to attract investment and accelerate innovation on the other.

→ Read the full analysis on aiworld.eu

2. AI investments of the week - Week #2

Last week, global AI funding reached $1.1 billion, signalling a sustained appetite among investors for both frontier models and specialised industrial applications. Some of the largest funding rounds included 🇨🇳 Moonshot AI $500 million, 🇺🇸 Hadrian $131.31 million, 🇩🇪 Mercura $2.1 million.

Most active sectors in AI investment this week:

1. Data & Cloud: GovDash (US), Mechademy Incorporated (US)

2. Finance: Lyte AI (US), SignalRank (US), Unusual AI (US), Rynaty AI (MU), Zendly Holdings Inc (US)

3. Industry & Mobility: Moonshot AI (CN), Hadrian (US), Mochi Intelligence (CN), Rewaa (SA), Rembrand (US), Linker Vision, Looki (CA), Memsilicon (CN), Ninetech (CN), Moushen Intelligent (CN), Ruifusheng (CN), Robolution (KR), Xiaoou AI (CN)

→ Explore the interactive chart for a detailed view of sector trends

3. This week’s most notable AI research papers

This week’s research shifts its focus towards the macroeconomic and societal architecture of an AI-driven future. The latest publications examine the structural reconfiguration of labor markets, how to preserve fiscal stability, and the complex psychological factors, including significant gender disparities, that affect the adoption of AI.

Want to know more? Check out this week's five most notable AI research papers to keep up to date with the latest technical and strategic developments in the field.

4. Startup spotlights of the week

This week’s startup selection illustrates how three companies are using AI to translate fragmented data into high-stakes operational intelligence across ESG compliance, preventive healthcare, and biodiversity.

🇩🇪 osapiens: addresses the growing regulatory burden around ESG and supply-chain transparency, using AI to automate data collection and compliance across complex global value chains, unifying ESG reporting with operational efficiency. The company recently raised a $100 million Series C to scale internationally.

🇨🇭 Ahead Health: by combining full-body MRI and advanced blood analysis to surface early health risks before symptoms appear, the company applies AI to preventive medicine. Ahead Health secured a $6 million seed round to grow its team and open new clinics in other European cities.

🇬🇧 Verna: provides a platform that navigates the complexities of nature recovery by harmonising fragmented ecological data through the use of AI and geospatial analytics. Verna has secured $4 million in funding to scale its data infrastructure.

What’s next

For more updates, research, and insights across the AI ecosystem, visit aiworld.eu. Subscribe to receive next week’s roundup directly in your inbox and stay ahead of key developments.

Have a great week ahead!

Gaia Cavaglioni
On behalf of the AI World Team