If you’re thinking about AI at the micro level, in terms of how you can actually use it right now, you know that most analogies are useless.

But most of the people talking about it from a market and investing perspective aren’t thinking about what it’s like to use it. They’re trying to think about how it behaves as part of the economy. And for that, it’s natural to reach for analogies.

The more you think about markets, the more it seems like everything is a rehashed version of something else. Hence the famous line: the four most dangerous words in investing are “this time it’s different.”

But of course, it is different. Don’t take my word for it. Go to ChatGPT or Claude and spend five minutes asking it to produce something that would have previously required several hours of work from a highly paid professional. A legal doc, coding work, art, music, poetry. Or something ridiculous, like what I posted below.

A reasonable person can’t attempt this and walk away uncertain that this time is different.

Okay. So back to analogies.

It’s worth saying that almost all of them are invoked to make a bear case. The bull case is “this is the most powerful technology that has ever existed, and it will have a larger and faster effect on the market than any other technology,” so comparing it to any other technology is naturally bearish.

My view is that the analogies are useful, and that they are all actually bull cases if you think about them. In basically every one, the thing AI is being compared to is a massive, world-changing technology that produced trillions of dollars of economic value, and AI matches or surpasses it in size, importance, and speed of adoption.

The five I want to look at are the internet, electricity, oil, money, and labor.

The internet

The idea here is that the buildout of compute is like the buildout of internet. Both are speculative bubbles fueled by the idea that there’s a beautiful, expensive future coming soon, and if we build it we’ll get handsomely paid for it. But the buildout is running too far ahead of the actual usefulness of the products, and we’ll wind up with a bubble bursting.

I don’t think this idea makes a ton of sense for a few reasons. The firm-level economics are very different. The internet buildout (by companies like Cisco and Worldcom) was debt-financed. There weren’t end customers. The compute buildout is financed mostly from the cashflow of hyperscalers. And they are quickly using the compute and reselling it to their existing billions of customers at huge margins.

It’s also different in that “access to the internet” or “increased internet speed” have no intrinsic value, but computer intelligence does. Certainly we see this in the extremes. The first person with access to the internet has nothing of value. The first person with access to ChatGPT could use it to make a lot of money. Imagine ChatGPT didn’t exist. Someone wants to hire you for an engineering position. You just say “give me a difficult, specific coding task that you think would take me one month to do”, and then you do that coding task in 1 hour.

A stronger argument is about the incremental value of each unit of ‘internet access/speed’ and each unit of ‘computer intelligence’ once we’ve already hit a certain level of saturation. Once enough people had internet access, and it was fast enough, the economics of building more internet to speed it up slightly for existing customers didn’t work. Maybe it’s the same with AI? ChatGPT rolls out an update and you now have access to a model that’s 5% better and 20% cheaper. Are you going to be measurably more effective? Possibly not. A reasonable-sounding argument here is that the intelligence is so good and so cheap that making it a little better doesn’t change anything. Therefore people stop buying more, data centers don’t get filled, server racks don’t get turned on, revenues evaporate like they did at the internet bust.

I just don’t think we see a lot of evidence for that.

An individual might not notice the incremental improvements of AI, but enterprises do. They notice when something that used to cost them $10M can now be done for $8M. People at the frontiers notice these things too, when the new version can solve a problem that the old version couldn’t.

AI is an enterprise product. Enterprise users are the ones who have endless demand for it. As a consumer, you probably run out of usage every now and then on your ChatGPT or Claude, and you probably think that’s not that big a deal. Maybe you tap into API usage now and then if it’s important.

It’s not that way for a business. They have not run out of ideas for how to use their tokens. They will keep buying more next year, because they have a positive return on tokens.

Electricity

The AI bear analogy to electricity is that electricity is an input to every business on earth and nobody got rich selling it. The profit went to the application layer.

That’s not quite right. Lots of people did get rich selling it, and still do; nowadays, their profits are constrained by regulation.

An important distinction between electricity and AI is that electricity is not valuable on its own. When it became available, people didn’t start piping it into their homes for its own sake. Companies had to invent products that used electricity and then sell those products to consumers and enterprises. Electricity itself was never a killer product; electric motors and lightbulbs were.

AI is not like this. From the moment ChatGPT appeared it was something people could buy and immediately use. Unlike electricity, AI is both the app layer and the input that facilitates the app layer.

Oil

If we’re comparing AI to electricity, it’s worth thinking about the other great energy technology of the last two centuries.

Oil is the better analogy for the production side. Like compute, oil is very expensive to get out of the ground. It takes rare technical expertise, a huge balance sheet, long-dated partnerships, accurate forecasting, and a tolerance for ten-year time horizons. It’s valuable because everyone wants it as an input. Its app layer is cars, trucks, ships, and planes.

Are any of those better businesses than Exxon?

Nvidia, TSMC, and ASML are closer to Exxon than they are to electric utilities. The “compute is a commodity” bear case neglects how expensive and difficult it is to build compute.

Money

Tokens are sort of like money. They’re fungible, they flow through the economy quickly, and a company’s decision to buy more of them looks like a capital allocation decision.

One claim of AI skeptics is that there isn’t yet sufficient evidence of return on token spend for end users. Chamath’s version is that some businesses are handing enormous sums to Anthropic and OpenAI because they feel they have to, and one of these days a company is going to miss an earnings report because it bought a lot of tokens and got nothing back.

Yes. That will happen.

It’s also possible to borrow a lot of money from a bank and invest it badly, and to hire fifty engineers and ship a bad product. When a company says it can’t get a positive return on its capital, we don’t conclude that money isn’t useful. We conclude the company is bad at using money. The same is true with tokens.

Trace a token from generation to consumption. It throws off profit at the production layer (ASML, TSMC, Nvidia, the memory makers), then at the refinement layer (Amazon, Microsoft, Alphabet), then at the model layer (Anthropic, OpenAI), then possibly at a router or a reseller. The bear case is that the Fortune 500 company at the end of that chain isn’t getting a positive return on the last dollar.

So what? It’s not an indictment of oil that drivers don’t report positive ROI on their gas spend. Some goods are consumed rather than propagated, and some returns aren’t measured.

The only reason we’d care about weak end-user returns is as a leading indicator of falling demand. But demand is not falling.

It’s fine if most token buyers underperform. Most users of capital underperform the cost of capital. The pool is positive because a few uses are so productive that they carry everyone else.

Labor

Labor is the tightest analogy, and it’s useful for thinking about the ROI argument.

When companies buy tokens, they’re doing labor replacement. Tokens do work that would otherwise be done by people. What’s powerful about them is how they differ from conventional labor: you can buy a dollar of it at a time, at any hour, from anywhere, pre-trained and capable of thousands of superhuman feats, with no obligations and no notice period. You can turn it off mid-task.

Because tokens are more like labor than like capital, we should expect their returns to be measured the way labor returns are measured - which is to say badly, and mostly not at all.

Sometimes you hire a salesperson and you can clearly see how much incremental revenue they bring in vs. a replacement.

Most labor isn’t like that. Nobody at any company you’ve worked for has produced a dollar figure for the return on a recruiting coordinator. Some hires are worth billions. Some cost billions.

The pool needs a positive return; each hire does not.

Tokens make the cost side legible but don’t make the return side any more legible.

So when someone says there’s no proof of ROI on token spend, a reasonable response is that there was never proof of ROI on labor spend either, and companies have been buying labor for ten thousand years.

The Elon Premium has been durable in the public markets. Tesla still has an enormous Elon Premium. If Elon were gone and Tesla were valued at 3x the average multiple of the next 10 largest automakers, it would be worth $167B. Its MC is currently more than $1.2T. That implies the Elon Premium at Tesla is more than $1T.

Now Elon is shifting his attention, recruiting, and compute to SpaceX. Presumably SpaceX will become the primary beneficiary of the Elon Premium.

SpaceX’s assets and capabilities are many times more valuable than Tesla’s. Starlink is already at basically the same profit run rate as Tesla ($4B), while growing faster with a much stronger moat.

SpaceX’s other assets include:

• a launch company that does a majority of the world’s launches (51%), controls a larger majority of the world’s satellites (66%), and is responsible for an even larger majority of payload sent into orbit (80%)

• hundreds of thousands of GPUs and the ability to build and get data centers online faster than any other organization

• $26B in annual compute commitments from Anthropic and Google

• A frontier or near-frontier LLM

• Twitter/X

The most reputable public business valuations (Morningstar and Damodoran) put SpaceX’s business at $780B and $1.3T. Let’s call it $1T of fair value.

If the Elon Premium is a static $1T, and it is fully replicated or transferred to SpaceX, then SpaceX is at $2T, already 14% above its IPO price of $1.75T. (In practice, I would expect Tesla’s Elon Premium to drop, which might also make it easier for SpaceX to acquire Tesla.)

If Elon’s premium is a multiple of business value rather than a static amount, then SpaceX will be worth significantly more than its IPO price.

The current Elon Premium at Tesla is about 7.2x business value. Here’s the value of SpaceX and share price increase at different Elon Premiums:

In other words, if you think that the Elon Premium is a multiple of the value of his business/assets, and if you think SpaceX has a better business and assets than Tesla, then SpaceX looks like it’s fairly priced with lots of room to grow.

If you don’t think the Elon Premium is transferrable, or you don’t think SpaceX is a significantly better business/asset portfolio than Tesla, then SpaceX looks overpriced.

Median income is a small part of the picture.

The gap between white and black households expands as income increases.

A household that’s in the 99th percentile of wealth among all American households is 42x more likely to be white than black, normalized for population.

Data is harder to come by for the super wealthy - the top 0.1% or the top 0.01%.

Just looking at the Bloomberg Billionaires Index, it seems that the richest 175 white Americans have more wealth than all black Americans combined.

If you hear a Congressperson talk about why they voted for or against a bill, you’ll usually hear some version of “I’m doing what’s best for my district”. And the person interviewing them will just move along, because that’s a totally normal thing we expect a Congressperson to say.

And yet the fact that Congresspeople make decisions based on “what’s best for the district”, and we all accept this and even agree that that’s what they’re supposed to do, is at the very core of our government’s dysfunction.

Somehow we’ve forgotten that the job of a Congressperson is not to do what’s best for their district. It’s to do what’s best for their country. A seat in the House of Representatives or the Senate is a job in the federal government.

We don’t make this mistake in other branches of the federal government.

The judiciary doesn’t operate this way. Judges in the Eighth Circuit Court, based in St. Louis, don’t interpret federal law in order to benefit St. Louis. They interpret laws for the United States. No one in the judiciary branch is at all confused about this.

Nor are they confused in the executive branch. Nobody thinks that if the Attorney General happens to be from Virginia, she should enforce laws in a way that benefits Virginia. Or that the Secretary of Defense from Florida should try to bring a new military base to the Orlando suburbs so he can create some jobs in his hometown.

Yet in the legislative branch, we’ve somehow all come to believe that the job of each Member of the House of Representatives is to do what’s best for the people in the place they happen to come from. Rather than make laws in the interest of America, they proudly announce that they serve the roughly 0.23% of Americans who cast ballots in their district.

It should shock us that we allow them to get away with this. Congresspeople go out and brag that they’re doing what’s best for their district, and what they’re really telling us is that they don’t care about serving the federal interest - which is their explicit job - but rather they care about serving the narrow slice of Americans who control whether they get reelected or not. It’s like if a President got into office and started telling interviewers, “I’m just trying to do what’s best for the people of Pennsylvania, Michigan, Wisconsin, Georgia, and the one to four other states that look like they might be within the margin of error when I run for reelection.”

If the job of a Congressperson is to do what’s best for their district, then there is no legislator whose job is to do what’s in the interest of the country. That can’t be right.

This situation neatly explains one of the bizarre political realities of the last several decades. Americans overwhelmingly disapprove of Congress - which typically has an approval rating between 12% and 20% - while at the same time they overwhelmingly reelect their Congresspeople. In the 2024 elections, 97% of Congressional incumbents who ran for reelection won.

If every Congressperson’s job is to do what’s best for their district, then nobody is responsible for the output of Congress. That’s why we expect so little of them and why they live up to our expectations. Each individual member can do a great job of bringing (or appearing to bring) benefits to their home district while the institution as a whole produces nothing of value for the country.

One reason we’ve allowed Congress to fall apart is confusion over the word “representative.”

There are two kinds of representatives - what you might call “client representatives” and “perspective representatives”.

A client representative is what we tend to immediately think of when we hear the word. They do the kind of representation a lawyer or agent would do for their client. The job of a client representative is to put the needs of their client first, no matter what.

A perspective representative represents the viewpoint, rather than the interests, of a particular group. In any deliberative body you can think of - a company’s board of directors, a church’s event planning committee, an NFL team’s coaching staff - each person is a perspective representative. The job of the offensive line coach, for example, is not to fight for the interests of the offensive line above everyone else. If he did that, he’d be terrible at his job and pretty quickly fired. His role within the team’s decision-making apparatus is to share insights that come from the offensive line in order to improve outcomes for the whole team.

The mistake we make, which I believe is the central enabler of Congressional dysfunction, is thinking that Congresspeople should be client representatives, when actually they should be perspective representatives.

It’s important that Congress has representatives from New York City, rural Georgia, and the Colorado suburbs not because we need people to represent those districts the way lawyers represent their clients, but because of the perspectives and knowledge that come from different parts of the country. If we want to make good agricultural policy, it’s probably a good idea to learn from parts of the country that do agriculture. Having a legislative body whose members include a cross-section of the country makes it more likely that the legislature will understand the nation’s problems and develop effective solutions.

It turns out there’s a simple mechanism for determining whether a representative should be a client representative or a perspective representative.

If decisions are made by a third party that does not include the representatives, then it’s a competitive decision-making process that requires client representation.

If decisions are made by a group of the representatives themselves, then it’s a cooperative decision-making process in which client representation is actively destructive, and perspective representation is the way to go.

Congress is in the second situation.

Let’s look at the kind of decision-making apparatus that requires client representation - which is decision-making handled by a third party after competition among client representatives.

The legal system is a great example. Lawyers represent their clients. It’s not a lawyer’s job to find the best, fairest, most sustainable solution. Actually, if a lawyer did that, it would be catastrophic - the whole case would be thrown out. The lawyer’s job is to advocate for their client’s interest, to the exclusion of everything else. The lawyer on the other side does the same. Then a decision is made by a third party - judges or juries - who weigh the case made by each client representative and determines the outcome.

A similar mechanism is at work in a market economy. The job of each company, and every person at each company, is to do what’s best for their company. The company is their client. The system would break down if people didn’t do this. A market economy works because representatives of different companies pursue their own interests, and third parties - consumers - weigh the options and make decisions.

Client representation is not only helpful but necessary in competitive scenarios. If GM and Toyota cooperated rather than competing, they could agree to raise their prices and reduce investments in future products, increasing profitability at the expense of consumers. Removing the competitive element between opposing legal counsels would similarly reduce the quality of outcomes. That’s why governments pass laws to ensure that competitive decision-making bodies remain competitive.

Now let’s consider cooperative, first-party decision-making - the kind that requires perspective representation.

Perspective representation is what you want when you’re running a company. When Apple is having executive team meetings and wants to decide how to allocate resources, they bring in representatives from different functional areas and business lines to make better decisions. They need perspectives from leaders in charge of supply chain and software engineering and the App store. They need these perspectives to get a fuller picture and make better decisions.

What you really want to avoid is that the decision-makers in the room view themselves as client representatives. If leaders are competing with each other over credit and resources, the whole company suffers. (The way you avoid this, of course, is through a combination of cultural expectations and aligning incentives toward the success of the firm rather than individuals.)

Similarly, imagine a seven-member school board in a Midwestern city of 100,000 people. Most of us understand that we don’t want everybody on the board to have the same background. It would be bad if the board were composed of seven wealthy businesspeople from the same part of the city who all send their kids to private school. It would also be bad if all seven were 11th-grade social studies teachers. The ideal composition would include a diversity of perspectives.

The school board members aren’t there to represent their geographic or ideological faction like a lawyer represents a client. It’s obvious to all of us that if that happened, the whole school system would suffer.

Congress is not like our legal system, or like a market economy. It is a cooperative decision-making body. A basket of elected officials, each representing a particular geographic segment of the country, is put into a chamber and tasked with creating laws on behalf of the federal government. There isn’t a mechanism by which competition among Congresspeople helps the country. It only makes things worse.

So we should not tolerate - much less praise - Congresspeople who say they’re doing something for their district. Next time you hear that, you should know that that Congressperson is unwittingly bragging about undermining the American government.

The competitors try to build their own version of the product to escape the tax. They pour billions into it. They recruit teams, sign manufacturing contracts, launch with fanfare.

But they can’t compete. The innovator is too far ahead. Too much production capacity, too good a supply chain, too strong a product. They fail, write off billions, move on.

It’s 2007, and the product is the iPhone. Or it’s 2022, and the product is the NVIDIA GPU.

Among the companies in the Mag 6 (excluding Tesla because they’re not in the same universe as the others in terms of business performance), Apple and NVIDIA are the only ones that make their money from hardware.

Right now Meta makes 99% of its money and Google makes 82% of its money by gathering eyeballs and purchase intent and selling advertising. Microsoft makes its money from its cloud business and enterprise software. Amazon makes money from cloud services and a retail marketplace. None of these companies need new product releases for new business. They just have the business.

Apple and NVIDIA are different. They’re hits businesses. Every year or two, they put out a new physical product. Their success as a company is mostly based on how many units of that physical product they produce and sell. When those products are announced, and when they come out, there’s a frenzy to understand the specs. For phones, it’s camera quality, battery life, weight, materials. For chips, it’s energy efficiency, token output per second, cost per token.

Both companies have historically produced premium products. The absolute top tier. The most expensive, with the highest margins, and almost universally seen as the best. Producing those products at the level of scale and quality they’ve achieved is extraordinarily difficult. They have to have the best design and engineering teams in the world, and the best supply chain management. Both companies purchase a huge share of the world’s scarce capacity for the inputs they need. In many cases their suppliers are the same (often Taiwanese) companies.

Amazon launched the Fire Phone in 2014. It was dead within the year. They sold roughly 35,000 units before slashing the price from $199 to 99 cents on contract and eventually writing off the entire project.

Microsoft acquired Nokia’s phone business for $7.2 billion. The Windows Phone peaked at 3.6% global market share, then collapsed. Microsoft wrote off $7.6 billion and discontinued the platform entirely.

The Facebook Phone (the HTC First, running Facebook Home) launched in April 2013 and was discontinued within months. AT&T dropped the price from $99 to 99 cents within a month of launch.

Only Google managed to have any real success. Their Pixel phone line is still going. It has about 3% market share in the US and roughly 2% globally after more than a decade of effort. Google’s real win in phones was not in hardware but in software. They bought Android, the OS that runs basically all non-Apple phones, and they decided to pay Apple $20 billion a year to be the default search engine on iPhones. Even Google, the most technically capable company on the planet, could not (or did not) build a phone that seriously threatened the iPhone. They wound up paying rent instead.

What happened to Apple because of the iPhone? At the end of 2008, Apple was the 35th largest company in the world, with a market cap of $76 billion. By the end of 2012, Apple was the largest company in the world - more than twice the size of Microsoft and more than 20% larger than the second-largest company - with a market cap of $500 billion. Apple has increased in value by roughly 8x since then, and iPhone sales still represent the majority of their revenue.

NVIDIA’s trajectory so far has been even more explosive. At the end of 2022, just before ChatGPT launched, NVIDIA’s market cap was roughly $360 billion. By the end of 2025, it was more than $4 trillion. That’s more than an 11x move in three years - faster than Apple’s 7x move from 2008 to 2012.

We should consider the possibility that the same dynamic plays out again. NVIDIA’s competitors - the same ones who went up against Apple and the iPhone - might be unable to keep up with NVIDIA’s accumulating advantages: its relationships with suppliers, its expertise and focus on chip design, its product cadence, and the hundreds of billions of dollars of chips it now produces at enormous scale. NVIDIA might remain the dominant player in AI chips for the next decade. It might, like Apple, grow much faster than the market and still be the largest company in the world at the end of 2035.

NVIDIA comparison you hear most often is not Apple but Cisco. The argument goes like this: people have overhyped AI just like they overhyped the internet in the ‘90s. Any stock with AI in the name is soaring, just like any company with .com in the name soared in the ‘90s. Cisco went up nearly 40x in five years from January 1995 to March 2000. It briefly became the most valuable company in the world at $569 billion, trading at 220 times earnings. Then the bubble burst. Cisco fell 88%. It took twenty-five years for the stock to reach a new all-time high - it finally broke $80 again in December 2025.

This comparison is useful because of how different things are this time.

In the late ‘90s, telecom companies built hundreds of billions of dollars’ worth of fiber-optic cable funded by debt, with no attached end customers and no clear path to revenue. Cisco’s stock price was built on the assumption that internet traffic would keep growing forever. That assumption turned out to be right! The internet did keep growing. But the infrastructure companies had overbuilt so far ahead of demand that the revenue took years to catch up to the capital expenditure. The financial structure collapsed even though the technological thesis was correct.

NVIDIA, on the other hand, has already sold out its 2026 inventory. They can’t build enough to meet demand. Their customers are the world’s most profitable companies, with fortress balance sheets, funding their purchases from operating cash flows. Microsoft, Google, Amazon, and Meta are not speculating that AI might produce revenue someday. They’re buying chips because those chips are already producing massive, measurable returns on their core businesses.

The broader market environment is different as well. There have been small, rolling bubbles throughout the last few years as markets try to make sense of what kind of future AI will bring. But the bubbles keep popping. Speculative names like C3.ai and BigBear.ai saw huge run-ups in early 2023 and then got crushed back down by 50–70% when investors realized the revenue wasn’t there. The stocks that have actually held their gains are companies with real earnings. NVIDIA has gone up roughly 11x in three years and is the largest company in the world, but it still trades at a P/E ratio that’s lower than Costco’s, even while NVIDIA grows its revenue 60% per year and Costco grows its revenue 8% per year.

The GPU-to-phone comparison is useful because it’s so many of the same players. In the 2010s, Amazon, Google, Microsoft, and Meta all tried and failed to build phones. Now they’re all trying to build chips. The question is whether the outcome will be the same.

People often talk about the “iPhone tax” and the “NVIDIA tax.” But they’re quite different things.

The iPhone tax was a toll. Apple took a 30% cut of every transaction made through the App Store. As the world moved toward mobile, companies had no choice but to use Apple’s platform because that’s where customers were. Companies paid the iPhone tax because they had no alternative.

The NVIDIA “tax” is different. Companies aren’t forced to pay NVIDIA. They buy NVIDIA chips because the more chips they buy, the more money they make. In order to accelerate their core businesses - advertising and cloud for Google and Amazon, cloud and enterprise software for Microsoft, advertising for Meta - these companies made the rational decision to buy NVIDIA chips. NVIDIA makes them richer. What people call the “NVIDIA tax” is just a way of saying that NVIDIA commands very high margins on the products they sell. Those margins exist because the product is extraordinarily valuable to the buyer, not because the buyer is trapped.

This distinction matters because it shapes the competitive dynamics. With the iPhone, the moat was deep - companies had to be on the platform because that’s where the users were, and switching costs were enormous. With NVIDIA, companies buy chips because it’s ROI-positive. That’s a less durable moat. But “we don’t want to pay the NVIDIA tax” doesn’t work as a justification for building your own chips, because it’s not actually a tax - it’s an investment that pays off every time. What the hyperscalers would really love is if they could purchase as many NVIDIA chips as they want while none of their competitors could. In reality, they fight over the scarce allocation.

The fact that the relationship is transactional rather than captive means the hyperscalers have both the motivation and the ability to try building alternatives. And they are trying. Amazon, Meta, Microsoft, and Google are all developing custom silicon. None of them contribute nearly the level of resources to it that NVIDIA does, and none of them own as much of the stack - they’re all reliant on design partners like Broadcom. But they all have something, and they all view it as strategically important.

Again, Google is the most successful. They started building chips more than ten years ago. They’re the only AI model developer that has trained a state-of-the-art model on non-NVIDIA chips. They’ve made a deal to sell up to one million TPU chips to Anthropic and have had talks about selling to Meta. They are probably the second-best chip maker in the world. But even Google’s chip team has talked about potentially getting up to 10% market share - which is about three times the Pixel’s share in the US phone market, but still not a threat to NVIDIA’s position at the top.

There are a few reasons the hyperscalers want to make their own chips, but none of them seem like an existential threat to NVIDIA’s dominance.

First, they simply need more chips. NVIDIA can only sell so many. They would buy more NVIDIA chips if they could, but they can’t. Making their own chips is not as efficient for general-purpose workloads as buying the best NVIDIA chips on the market (though this may no longer be true for Google’s TPUs, which trained Gemini), but it’s better than not having as many chips as they want.

Second, by designing their own chips, they can build them for more narrow use cases. NVIDIA’s GPUs are the best all-around chips, and they’re the best for training new models, but they’re overpowered for some jobs. For purely inference workloads - using a model rather than building it - and for jobs specific to their business, they can get better bang for their buck with custom silicon. Meta, for instance, has some chips built just for Instagram recommendations. It’s more efficient for them to develop narrow, company-specific chips for narrow, company-specific workloads.

Third, they don’t want to give NVIDIA too much pricing power. Without the credible threat of using alternative chips, NVIDIA could squeeze more and more margin out of them.

Fourth, they view NVIDIA as a potential competitor and don’t want them to get too powerful. NVIDIA is already the largest company in the world and not lacking in ambition. If NVIDIA goes unchecked, what’s to stop them from challenging the hyperscalers in their own businesses?

Of course, there is a legitimate concern that as the hyperscalers make more of their chips in-house - even if they never sell them externally - it will erode NVIDIA’s revenue and margin over time. This is worth watching. But so far, the hyperscalers’ custom silicon has been additive to NVIDIA purchases, not a substitute. They’re buying everything NVIDIA can produce and building their own chips on top of that.

The internet buildout and the growth of smartphones are useful comparisons for NVIDIA and AI because they took place in reasonably similar market environments, and in many cases with the same companies. But in terms of technological and societal impact, the better comparison is electrification.

The primary economic impact of both electrification and AI is that they augment, automate, and replace labor. The internet and smartphones, by contrast, are primarily communications-based technologies - heavily tilted toward the consumer, although both enabled new business processes. Electricity and AI are a level deeper. They’re more foundational. They will create new communication systems, new consumer products, new ways to have fun. But primarily, they enhance and replace human labor. That’s what makes the demand dynamics so different from phones.

Both electricity and AI produce effectively unlimited demand. You can never have enough energy or enough intelligence. If you have more of it, there are always more things you can do with it. Every new application that electricity enabled - lighting, then motors, then radio, then refrigeration, then television, then computing - created demand that dwarfed what came before. The grid kept building and building.

Both electricity and AI can be applied to essentially any task to make it significantly more effective or more efficient. Today this is most evident in any kind of work that involves analyzing or producing text, code, video, and audio. But there is another leg of innovation already underway that will extend AI into the physical world. We’re seeing it first with autonomous driving. Waymo delivered over 14 million fully autonomous rides in 2025, tripling its volume from the prior year, and went from 10,000 rides per week in May 2023 to more than 450,000 rides per week by December 2025 - a 45x increase in two and a half years.

The electricity comparison is especially instructive because of how long it took to realize efficiency gains that appeared obvious in hindsight. Economist Paul David documented a thirty-to-forty-year lag between the invention of the electric motor in the 1880s and the appearance of measurable productivity gains in the 1920s. The reason for the lag is that the first generation of adopters used the new technology to do the old thing slightly better, rather than rethinking their entire operation.

David called this “group drive.” Factories in the 1890s and 1900s replaced their central steam engine with a central electric motor but kept the same layout - the same system of belts and pulleys driving all the machines from a single shaft. They got modest fuel savings and nothing more. It wasn’t until a new generation of managers adopted “unit drive” - putting an individual electric motor in each machine - that factories could be redesigned from scratch around the flow of materials and labor rather than the constraints of power delivery. Single-story, open-plan, well-lit, flexible. The assembly line itself was only possible with unit drive. You can’t power a quarter-mile production line with belts from a central steam engine.

Most companies today are in group-drive mode with AI. They’ve bolted a chatbot onto their customer service page or they’re using Copilot to autocomplete code. These are improvements, but they’re modest. The unit-drive moment for AI - where organizations are redesigned from the ground up around AI capabilities - hasn’t happened yet at most firms.

But to be clear, AI is already producing efficiency gains much faster than electricity did, and faster than the internet or smartphones did. US GDP grew at an annualized rate of 4.4% in Q3 2025, even with elevated interest rates - the strongest quarter since the post-COVID rebound in late 2021, and excluding that anomaly, the best growth since 2014. This in a period when hiring has slowed considerably: Big Tech companies cut new-graduate hiring by 25% between 2023 and 2024, and a Chief Executive survey found that nearly one in three companies have already reduced headcount due to AI. The economy is producing more with fewer new workers.

It makes sense that AI would be faster than previous technologies. Each wave has been faster than the last. Electrification took thirty to forty years to transform manufacturing. The internet took about fifteen years to transform media and commerce. Smartphones took about seven years to become the dominant computing platform. And AI has a structural advantage that none of the prior technologies had: the buildout of the internet and smartphones has made it trivially easy to get access to incredibly powerful AI. If you have a phone, you can access AI that can produce tens of thousands of dollars worth of knowledge work at basically zero marginal cost. No previous general-purpose technology was this easy to deploy.

Still, the productivity gains are painfully slow. Faster than from any other technological breakthrough, but still too slow. The abilities of AI are accelerating at a speed well beyond our collective knowledge of how to use it. Part of this is about the user interface - model providers and the apps built on top of them are still figuring out how to make AI easy to use. Part of it is about imagination. We just don’t know all the things we can do yet. And part of it is about the pace of change within businesses - their reluctance to cannibalize their existing products and their hesitancy to overhaul their processes. Every enterprise sales cycle takes time. Every reorganization takes time. The technology is ready before the institutions are.

This is the bull case for chip demand being sustained for a very long time. We’re somewhere in late Phase 1 of the electrification analogy. The technology works. Early adopters are spending aggressively. But the big transformation - the unit-drive moment, where industries are reorganized from the ground up - hasn’t happened yet. When it does, the demand we’ve seen so far will look like a warmup.

There is one more similarity between electricity and AI that is worth naming plainly. Both are obviously labor-replacing technologies, and both have produced - and will continue to produce - social and political backlash. When electricity automated factories in the early twentieth century, it generated enormous wealth and enormous dislocation. Entire categories of manual work disappeared. The political responses included the labor movement, the New Deal, and eventually a massive expansion of the welfare state.

AI will generate a similar set of responses. It might be the most likely constraint on the pace of AI adoption. That, more than any competitor or any business cycle, might be the biggest risk to the thesis that NVIDIA and AI chips will enjoy decades of extraordinary demand growth.

The history of electrification is instructive here too. Samuel Insull built an electricity empire that, at its peak in 1929, served four million customers across thirty-two states and was valued at over $3 billion - roughly $55 billion in today’s dollars. His strategy was to make electricity so cheap and so abundant that everyone uses it constantly for everything. He called it the “gospel of consumption.” He got exclusive franchises, built massive generating capacity ahead of demand, and cut rates relentlessly to drive adoption. His stock prices more than doubled in the summer of 1929.

Then the Depression hit. His holding-company pyramid collapsed. Investors lost $700 million. He was indicted for fraud, fled to Greece, was extradited, stood trial, and was acquitted. The operating companies survived - Commonwealth Edison still exists today as part of Exelon. The technology was real. The demand was real. The empire was real. But the financial structure and the political environment couldn’t sustain the pace of growth

.

NVIDIA is not Insull. Its balance sheet is clean, its customers are solvent, and its margins are earned rather than engineered through leverage. But the lesson from electrification is not that the technology fails. Electricity won. It just took longer than the first generation of investors expected, and the path from here to there was not a straight line. The biggest risk is rarely that the technology doesn’t work. It’s that everything around it - the politics, the financial structures, the social institutions - can’t keep up.

NVIDIA’s chips are the most consequential hardware product since the iPhone, and possibly since the electric motor. The demand is real. The technology is transformative. The competitive moat is wide. The question of whether the world can absorb rapid AI adoption is what makes NVIDIA hard to model.

For a few reasons, Congress has been systematically giving away its powers for decades. The biggest one is that the primary goal of most Congresspeople is to get reelected, and legislating makes it harder to get reelected, because you might put yourself on record for doing something that turns out to be unpopular. Better to do nothing at all.

One of the powers that Congresspeople still have, but don’t really want to do, is impeach and remove federal officeholders. Only Congress can do this.

Americans found out in 2024 that Supreme Court Justice Clarence Thomas has received at least $4M in compensation from a billionaire and others during his time as a justice. Payments came in the form of luxury vacations, trips on private planes, large loans that were forgiven without being paid, the purchase of a house where Thomas’s mother lived, and full tuition paid for Thomas’s adopted son at an expensive private school.

Supreme Court Justice Samuel Alito received large gifts from a different billionaire benefactor, whose investment fund went before the court at least ten times. The gifts included a flight on a private jet to a luxury fishing trip in Alaska.

Neither Thomas nor Alito recused themselves from the cases involving the billionaire benefactors.

This is rather straightforward corruption of the kind that in a society with a strong anti-corruption culture would lead to the disgrace and resignations of the offending judges. In a country with a weaker anti-corruption culture, it would lead to their impeachment and removal from office. In a country that no longer has the capacity to remove corrupt federal officials, the justices would remain in office. The last scenario is the one we’re in.

Actually, AOC introduced articles of impeachment against Justice Thomas and Alito, but in a Republican-controlled House, they did not get a vote. They will never get a vote as long as the Republicans have a majority in the House.

I was too broad earlier when I suggested America doesn’t have the capacity to remove corrupt federal officials. In fact, we can very easily remove corrupt Democratic officeholders. What we can’t do is remove corrupt Republican officeholders. That’s because in the Trump era, the Republican party is pro-corruption.

When a Republican is found to be corrupt, Republicans do nothing about it. When a Democrat is corrupt, Democratic prosecutors charge him, Democratic lawmakers remove him from office, and then the corrupt Democrat leaves the Democratic party and allies with Trump, who publicly praises the corrupt Democrat.

This happened in 2023 with Senator Bob Menendez, a Democratic Senator who was charged with bribery by Biden’s DOJ and resigned from office after a majority of Democratic Senators (but not a majority of Republican Senators) called for him to resign from office. It happened again in 2024 with NYC Mayor Eric Adams, who was indicted for bribery by Biden’s DOJ. Adams has since had the charges dropped against him by Trump’s DOJ and is now running for NYC Mayor as an independent.

This is the context in which 10 Democratic Congresspeople proposed the Judicial Ethics Enforcement Act, which would create an Inspector General within the Judiciary Branch.

Inspectors General are watchdogs within federal agencies whose job is to find corruption - and yes, waste, fraud, and abuse - within the agencies they monitor. Not surprisingly, the Trump Administration fired 17 Inspectors General shortly after taking office.

This proposed Inspector General for the Judiciary Branch would be authorized to investigate any federal court and issue subpoenas. They would also be appointed and subject to removal at any time by the Chief Justice of the Supreme Court.

This bill won’t pass, and it probably wouldn’t accomplish much if it did. It’s like asking your teenage child to self-report every time they skip class.

That’s where we are right now. The non-corrupt parts of our government are so desperate for oversight that they’re begging other branches to create an avenue for self-reporting, and even that’s not going to happen.

There’s a very good reason to institute that ban. It’s illegal for anybody to make stock trades based on important, non-public information they have about companies or markets. A common example is a person who works at a company and knows they’re about to announce some big thing that will cause the stock to go up or down, and they buy or sell based on their inside knowledge of that.

Congresspeople have a LOT of access to material, non-public information that also happens to be classified. That means that they have frequent opportunities to do insider trading, and it’s really difficult to have a mechanism for figuring out when they do it.

This is not a new issue. There’s been plenty of suspicion for more than a decade that some Congresspeople are buying and selling stocks based on inside information. The issue gained heat in 2011/2012, and Congress nearly unanimously passed the STOCK Act, which Obama signed into law. The STOCK Act was ineffective for a number of reasons:

• It didn’t ban trading. It instead affirmed that Congresspeople and other government officials were not allowed to trade on inside information

• It required Congresspeople to publicly file their trades, but it gave them a $200 fine if they didn’t do so, which is not a sufficient deterrent

• Congresspeople’s trading filings initially went into a searchable database, but an amendment removed that requirement and instead organized filings into unsearchable PDFs

There’s been increased energy to curb Congressional stock trading since 2020. The face of Congressional stock trading has mostly been Nancy Pelosi.

Pelosi herself doesn’t seem to be an active trader, but her husband, Paul Pelosi, is a very wealthy professional investor who has been actively trading, mostly via call options, and has made a lot of money since the pandemic.

It’s certainly plausible that Paul Pelosi could make great investments without using inside information from his wife, who’s been in Congress for 38 years and has led the Democratic caucus in the House from 2003 to 2023. It is not plausible that during all that time, Paul never gleaned any inside information from his wife. Spouses tend to talk about their jobs, and there would be hundreds of days during her time when Congresswoman Pelosi’s job involved market-moving information.

Every Congressperson and their family has access to inside information. It’s not practical to try to investigate every Congressperson. Given that their freedom to trade stocks and buy call options is not a particularly important freedom, it seems obvious that they should not be allowed to make trades. Of course, they should still be allowed to put their money into a trust and grow their wealth through stocks and bonds.

The idea of banning stock trading for Congresspeople is back in the news because Senator Josh Hawley has proposed a new bill trollishly named the PELOSI Act, which would make it illegal for Congresspeople to buy, sell, or hold any stocks, commodities, derivatives (e.g., options) or security-like instruments. Basically all they can own are diversified mutual funds and ETFs and treasury bonds. If a Congressperson violates the Act, then all gains from their assets that are in violation would be given to the treasury.

This bill seems pretty good!

Two things that might be problematic are a) they don’t define “diversified” in this document, and it’s definitely possible to use inside information to gain an advantage in buying and selling mutual funds and ETFs, and b) it’s not clear if this includes crypto.

It’s not clear whether this bill will make it through Congress. In the past, these kinds of bills haven’t made it to a full vote. The best hope for this one is that Trump loves the idea of being rude to Nancy Pelosi, so even though he typically wants nothing to do with Congress, he might find it worth his while. He has already said he would sign the bill.

Another noteworthy component here is that unlike some previous bills, this does not apply to the President or anyone else in the executive branch.

Since we all agree we shouldn’t allow Congresspeople to do insider trading, it seems like, for all of the same reasons and more, we shouldn’t allow the President to buy, sell, and hold individual securities - especially memecoins that have no actual value and are irrefutably being used as bribery vehicles.

Make it harder to prosecute or sue police officers

The EO directs the DOJ to provide indemnification and legal resources for officers “who unjustly incur expenses and liabilities for actions taken during the performance of their official duties.” One resource is “the use of private-sector pro bono assistance.” A month ago, Trump threatened to take away the security clearance and disadvantage the clients of a bunch of elite law firms who had represented Trump’s opponents previously. They caved against the pressure and agreed to give him almost $1B in pro bono legal assistance. Presumably these are the lawyers who will be forced to defend police officers who kill civilians.

Summary: If you sue an officer, the government will pay the officer’s legal bills and supply top-tier counsel, whom Trump previously extorted.

End federal consent decrees

Since 1994, DOJ has attempted to rein in abusive departments through court-enforced consent decrees, which are basically performance improvement plans that incentivize the departments to reform themselves. The EO orders DOJ to “modify, rescind, or move to conclude” existing decrees that “unduly impede” policing.

Summary: Abusive departments are now off the hook. Future abuses presumably will be ignored by the DOJ.

Threaten judges and local officials who enforce state laws over federal ones

The EO tells DOJ to “prioritize prosecution” of officials who “willfully and unlawfully direct the obstruction of criminal law.” This looks like an intimidation attempt aimed at jurisdictions that don’t aggressively deport all undocumented immigrants. Under Supreme Court precedent, local officials can’t be forced to carry out deportation raids. This is a threat to expand prosecutions against those who don’t participate in federal deportation efforts.

Summary: Local law enforcement and judges will feel pressured to assist with deportations, and may be prosecuted if they don’t do it.

Threaten judges and local officials over DEI

The Attorney General is directed to prosecute officials who “engage in discrimination … under the guise of diversity, equity, and inclusion” in a way that “endanger[s] citizens”. This could be nothing, or it could be that the DOJ will charge law enforcement officers who once touted diversity if something goes wrong.

Summary: If you say anything positive diversity and then any bad thing happens where a black person or woman is involved, you might be prosecuted.

Give more military supplies to local police

U.S. policing is already heavily militarized. Every American police officer carries a gun. Most of them keep long guns in the trunk. Some of them have tanks and grenade launchers. In the UK, if a police officer knocks on your door he might say “Police!” or he might say “Armed Police!”, because only 4% of UK police officers are armed. The US has 4x the homicide rate and 138x the firearm murder rate of the UK. I wonder if giving more military supplies to American police officers is the answer.

Summary: More military equipment for local police officers.

Decide how military assets & personnel can “prevent crime”

Charitably, this is about the DOJ wanting to bring some high quality forensic techniques from the military to local police. But the more likely application of this section is that Trump wants the option to deploy the military domestically. He tried in 2020 and Defense Secretary Mark Esper said no. Trump fired Esper days after losing the election, and then Trump led an insurrection against the US government using civilians. One would imagine he’d rather use the military.

Summary: Trump would like to use the military against Americans.

You could say the goal was government efficiency, and Elon often did say that, and suggested that it would be wild for anyone to be opposed to what he was doing in the government - because how could anyone be opposed to government efficiency? Of course no one would oppose government efficiency. They just didn’t believe that Elon was doing it.

If you scratch the surface, then you quickly find that what Elon really meant by “efficiency” was “reducing government spending”. He started off by announcing that he would cut $2 trillion from the federal budget. Then $2 trillion became $1 trillion. And now Elon claims, almost certainly falsely, that he’s booked $150B in future savings, and he’s calling it a day.

Why people let Elon do this

This whole thing started with Twitter. When Elon bought Twitter, he pretty quickly fired about 80% of the staff. Super-wealthy CEOs LOVED this. They thought it was an example of incredible leadership. They were inspired; they wished they could do it too.

In the story Musk and these CEOs told, he was able to fire 80% of the staff and make Twitter even more efficient. But that’s not really what happened. Twitter’s revenue tanked. The company had been profitable before Musk bought it, and then it stopped being profitable after Musk fired everybody. If you’re on Twitter, you know that the product got a lot worse too. A platform that previously allowed equal distribution to everyone is now pay-to-play. If you don’t pay $100/year, your posts get throttled.

In business, you can say that “efficiency” is basically the same thing as profitability. By that definition, Elon failed to bring efficiency to Twitter. Profit margin is not a way you can measure the government. The government is not a company, and it’s definitely not a company like Twitter. Twitter does one simple thing. The government does thousands of complicated things.

Elon loves destroying “woke” agencies

Elon’s only vision for government efficiency was cutting government spending. Yet he did not choose go after the big drivers of government spending. The first thing he did was shut down, or attempt to shut down, a few agencies - in particular, the US Agency for International Development (USAID), the Department of Education, and the Consumer Financial Protection Bureau (CFPB). Elon didn’t choose these because they were particularly large or wasteful, or because DOGE had an idea of how government should alter its aid-based diplomacy, education, and consumer protection to make them more efficient. He just thought there were a lot of “woke” people in those agencies, and that those missions were kind of “woke” things for government to do, so he decided to go ahead and, in his words, feed them “into the wood chipper”.

Another agency he gutted was the IRS. The IRS is not popular, for obvious reasons. Nobody likes paying taxes. It's a hassle, and it should be easier.

We all hear a lot about “audits”, and there’s a popular idea that they’re some horrible, common thing that we’ll have to go through some day. It’s kind of like the way we all used to think about quicksand. The reality is that the people who get audited are mostly people who are cheating on their taxes, or very rich people who have complicated, unusual finances (and might be cheating on their taxes).

If the goal is to make the government more efficient, and the way you’re defining efficiency is about how much money the government takes in vs. how much it spends, then it’s a bizarre choice to fire a bunch of people at the one agency that collects revenue for the federal government.

Biden had put in policies to hire a lot more IRS agents because it turns out that if you have more IRS agents, then you increase the amount of tax revenue you generate. Super rich people find ways to get around paying taxes, so when you hire more IRS agents, what they do is figure out how rich people are cheating on their taxes and then collect more tax revenue. The IRS doesn’t spend a lot of time auditing poor people with super simple tax structures, because the money is all with the rich people, and they’re also the ones who hire armies of lawyers and accountants to help them develop sophisticated ways to avoid payment.

I wonder why the world’s richest man, who has the most convoluted, self-dealing business empire of all time, would want to defund and monkey around with the IRS?

Elon loves firing people

Federal salaries are about 7% of the federal budget, so firing people is not a particularly useful way to cut government spending. But it’s something Elon really loves doing, so firings and compelled resignations have been DOGE’s primary outputs.

There are actual inefficiencies within the federal government Civil service genuinely do make it very difficult to fire low performers. These protections also make it difficult to protect high performers from being replaced by cronies. Reforming civil service protections would be a useful way to create more government efficiency. On the other hand, it would require time and care, and Elon preferred to just do things fast and then leave.

Instead of working on a system that would allow government agencies to reward high performers and get rid of low performers, DOGE got rid of government employees in the most comically inefficient ways possible:

1. They offered EVERYBODY in the government a severance package that included 8 months’ pay.

2. They fired probationary employees, because they didn’t have the same civil service protections as more tenured employees

Offering everybody a severance package creates adverse selection - high performers take the severance package and low performers don’t. Imagine you’ve been working in the government for 20 years and you’ve made it to a senior position. One day, the government says “hey, we don’t want you anymore, you can take 8 months of pay and leave”. That probably feels like a slap in the face. If you’re a high-performing senior employee, you probably know some organizations in the private sector who would offer you a job tomorrow at a much higher salary than what you get with the government. Eight months of severance might seem pretty enticing.

On the other hand, if you’re a low performer, you probably won’t be able to turn around and get a better job next week. The mere fact of receiving this generous offer for a severance package might remind you that you can’t get fired from the government. Might as well stick around, slack off as much as you want, and keep collecting paychecks!

How could anybody think that offering everybody in the organization a severance package is a good way to run an organization? When you have good employees doing essential work, it’s a good idea not to offer them a severance package. They might take it!

In the case of the fired probationary employees, “probationary does not mean that the employees did something wrong. It just means that they are new to the job.

Again, you have adverse selection effects at work.

If there’s a person who’s been at the company for 30 years and is doing something that becomes less and less valuable every year due to changes in technology, circumstances, agency structure, etc., then firing that person might be efficient. If you hired somebody three months ago, you probably hired them to do something you actually need!

Not all probationary employees are brand new to the organization. Some of them are people who have done great work in the federal government and are now on probationary contracts because they just got hired into a new position. Is that really who you want to fire?

This slapdash mass firing/compelled resignation strategy has caused essential government employees to leave their jobs or be unable to do them. We know that DOGE fired 314 engineers and technicians who maintain our nuclear warheads, and that they canceled funding for Ebola prevention and then lied about having done so. It’s frightening to think about how many life-saving services have been quietly eliminated.

It’s also not clear that any of this is legal.

Everybody who’s been watching this knows that Elon is personally responsible for these strategies. He’s the one who’s gutted the agencies. He clearly personally wrote emails asking people to resign.

He also rather obviously doesn’t have the authority to do any of this. The federal government has been pretending in court that Elon is not the DOGE administrator, but Elon, Trump, and everybody in Congress have been very public about the fact that it’s Elon doing all of this - even without any official authority to do so.

Many of these firings will be challenged in court, at enormous cost in time, money, and, well, efficiency.

It also makes our government look like incompetent. Can you imagine being a federal employee who does a great job for a decade, gets promoted to a senior position, and then 6 days later you get an email from Elon Musk telling you you’ve been fired for your poor performance?

Progressive taxation means that richer people pay a higher proportion of their income in taxes than poorer people.

Regressive taxation means that poorer people pay a higher proportion of their income in taxes than richer people.

The status quo

Right now, income taxes are progressive (in large part), and every other tax is regressive. Individual income taxes represent about 51% of federal government revenue. Corporate income tax represents around 10%, payroll taxes represent about 35%, and excise taxes represent about 2.5%.

Regressive taxes

Excise taxes are taxes on the consumption of goods and services. This category includes all sale taxes and additional taxes on certain goods, including gasoline, tires, alcohol, and cigarettes.

Individuals all pay the same percentage tax on their purchases regardless of income, but the aggregate effect of these taxes is regressive because poorer people spend a much larger proportion of their income on taxable goods.

Mostly that’s because rich people have enough money that they can spend a lot of it to buy investment assets, which are not taxed at purchase. Poorer people spend a much higher portion of their money on food, clothes, and gas, so they wind up paying a much higher portion of their income on excise taxes.

This is how a flat tax rate actually has the regressive effect of costing poorer people relatively more of their income than it costs richer people.

Payroll taxes are even more regressive. Everyone who gets a paycheck for their work in America (including non-citizens) pays payroll taxes. About 73% of payroll taxes go toward funding Social Security. The Social Security portion is paid at 6.2% of employees’ salary until the employee earns over $176,100, at which point they have hit their Social Security cap and they stop paying Social Security tax. Social Security tax is an extreme kind of regressive tax where the richer you are, the lower the effective tax rate you pay.

Individual income tax is the only part of the tax code that has even a nod toward progressive taxation. The way it works is that there are a few different tax brackets. When someone makes enough money to move into a higher bracket, then every marginal dollar they make beyond the threshold of that bracket gets charged at the new rate, while all their previous income gets charged at the rates of lower brackets.

Notice that the current tax rates stop increasing after $609,351. These tax brackets are progressive. They increase as the taxpayer’s income increases. Democrats would like them to continue to progress. They argue that there should also be higher tax rates for incomes that go over $1M, $2M, $5M, etc.

The current tax brackets are the result of decades of political fighting in which Democrats argue for more progressive taxation that scales into the incomes of the super wealthy, while Republicans want less progressive taxation that stops progressing at a lower income level. Actually, many Republicans - including Trump over the last several months - would prefer that income tax be completely flat (which would have a regressive effect) or that it be removed altogether and replaced with excise taxes (which would be extremely regressive).

Loopholes in the individual income tax

The individual income tax rates look like they are progressive, even if they’re not as progressive as Democrats would like them to be. But these brackets are misleading. There are two big loopholes that wealthy people use that undercut the tax progression.

The first is capital gains tax. Taxes paid on capital gains - which are increases in the value of investment assets - are charged at a lower rate than regular income. If you hold an investment for more than a year, you’re charged a 15% tax on capital gains between $48,351 and $533,400, and 20% on capital gains over about $533,400. Poorer people don’t make much money from capital gains, but that’s how the wealthy make the vast majority of their money. A person who makes $10M in a year from their investments will pay a lower tax rate on that income than the marginal rate for a salaried worker making $48,000 a year.

The second is a more complicated one, summed up by the phrase “Buy, Borrow, Die”, and it’s the strategy used by mega-billionaires to pay almost nothing in taxes. Here’s how it works.

Let’s say you’re Mark Zuckerberg. You have about $180B in Meta stock. If you sold your stock, you’d have to pay a capital gains tax of about 20% on your profits. What Zuckerberg and other billionaires do instead is they take out loans against their equity stakes in these companies. Zuckerberg goes to a bank and says he wants to open a line of credit for a personal loan using his Meta stock as collateral. The bank is more than happy to do it; the collateral is enormous and reliable, and they want to have a good relationship with Zuckerberg because they want his company to use their bank for its banking needs.

The bank gives Zuckerberg a permanent line of credit at extremely favorable terms. He gets a lower personal loan rate than anyone else can get - lower than inflation.

Zuckerberg can take out billions of dollars from the bank anytime he wants, and use it to buy islands and mansions, send his kids to private school, hire an army of workers to manage his household, and so on. He’s living off of the billions of dollars he’s borrowing from the bank and he never has to pay taxes on it because it’s not income.

Zuckerberg will do this for the rest of his life. Then when he dies, his heirs will inherit his (trillion-dollar?) equity stake. They can choose to borrow against it for their whole lives and never pay taxes on it, the way their father did. Or they can sell it immediately and not have to pay taxes on it, thanks to something called stepped-up basis. What this means is that all previous capital gains accrued by the owner of the asset are wiped away upon inheritance, so the heir can sell the asset at its current value without having to pay a cent in taxes.

This whole process allows America’s wealthiest people to claim that they have low incomes even while they’re increasing their net worth by billions of dollars a year and spending hundreds of millions a year on lavish lifestyles. In 2011, Jeff Bezos was worth $18B, but reported so little income to the IRS that he was able to claim the Child Tax Credit reserved for couples making under $130,000.

When you look at a chart that shows how progressive our country’s individual income tax system is, remember that the ultra-wealthy live outside of that system and live off of tax-free loans while claiming to have paltry incomes.

• Obstruction of a federal proceeding (18 U.S.C. § 1505)

• Concealing a person from arrest (18 U.S.C. § 1071)

We haven’t heard Dugan’s full story yet. Here’s what happened according to an affidavit from an FBI agent on the scene:

ICE wanted to arrest and deport a 30-year-old Mexican citizen named Eduardo Flores-Ruiz, who was subject to a 2013 deportation order. Flores-Ruiz was in court that morning for a misdemeanor domestic-battery hearing based on allegations from his two male roommates.

At least six plainclothes officers from ICE, FBI, DEA, and CBP arrived at the courthouse and waited outside Courtroom 615 to arrest Flores-Ruiz. Judge Dugan stepped from behind the bench to confront the officers in the hall. She asked if they had a judicial warrant (a warrant signed by a judge that allows officers to enter and search private property). They said they did not; they only had an administrative warrant, which is a warrant issued internally by ICE and does not allow officers to enter or search private property.

Dugan asked some of the officers (not all of the officers identified themselves to her) to go speak with the Chief Judge of the courthouse to determine whether it was appropriate for them to make an arrest in the courthouse hallway. Some of the officers were escorted by another judge to see the Chief Judge.

Back in the courtroom, Dugan instructed Flores-Ruiz and her attorney to leave the courtroom via the jury room. Flores-Ruiz and the attorney did so. The exit from the jury room led back to the same public hallway where at least two agents were waiting. Two agents immediately identified their target. One of them followed him into an elevator without identifying himself. The agents followed Flores-Ruiz outside of the courthouse, and then identified themselves. Flores-Ruiz ran away and was quickly caught.

Arrests in courtrooms and courthouses

It’s entrenched in American law that people don’t get arrested for unrelated crimes or immigration offenses while they’re in a courtroom. It’s hard to take courts seriously as guilt-determining venues if a federal officer can just come in and abduct you while you’re standing trial. During Trump’s first term, he modified deportation policy to allow immigration officers to make arrests in lobbies and hallways of courthouses. Biden overturned that policy, and now Trump has reinstated his original policy.

Several states have passed laws making it illegal for immigration officers to make arrests in public spaces in courthouses without a judicial warrant. Wisconsin does not appear to have such a law. The Supreme Court has not yet weighed in on this issue.

The case for allowing immigration officers to make arrests in courthouses is that it’s easy for them to pinpoint the identity and location of their targets, and courthouses are secure buildings where it’s generally safe to make an arrest.

The case against allowing immigration officers to make arrests in courthouses is that it dramatically reduces the likelihood that undocumented immigrants or relatives of undocumented immigrants would report crimes or serve as witnesses during trial.

Why the case looks weak

The FBI’s case looks weak because there’s very little evidence of material harm. They claim that Dugan concealed a person from arrest and obstructed their attempt to arrest someone. But it appears that all she did was instruct him to leave the courtroom through a particular door that led to a hallway where agents were waiting to arrest him. He was no more “concealed” in the jury room than he had been in the courtroom, and Dugan’s intervention did not delay the government’s procedure; in fact, it sped things up.

The FBI politicizes the case

The FBI Director Kash Patel posted a photo on Twitter of Dugan in handcuffs. That’s explicitly against DOJ protocol:

“DOJ personnel should not voluntarily disclose a photograph of a defendant unless it serves a law enforcement function or unless the photograph is already part of the public record in the case.”

It’s clear that the FBI views this arrest as an opportunity to intimidate judges and other state officials into cooperating more fully with immigration officers.

Insider trading is illegal because it puts everybody who doesn’t have inside information at a disadvantage, and it incentivizes people who run the company to make money by causing large swings in the company’s valuation (and trading stock based on those swings) rather than by increasing the overall value of the company.

If you want to make a LOT of money doing insider trading, you want to buy options. Options are leveraged bets on the movement of a stock over a given period (e.g., one day, one week, one month) that allow you to earn bigger returns or suffer bigger losses than simply buying a stock.

Options are the right tool for the job

Suppose you have some inside information and that you think is going to make a stock or index pop and you’re right: the stock goes up 5%.

• Buy shares in the stock → +5% return

• Buy a zero-day (0-DTE) call option with a strike price 1% above the current price → +1,400% return

Of course, if the stock drops 5% on the day, then your call option goes to zero and you lose it all.

The same logic works on the downside. If you know there’s going to be a down market, you can buy zero-day put options with a strike price 1% below the current price and profit hugely from the drop.

What makes insider trading hard

1. You have to find inside information

2. You have to avoid prosecution when your trades look suspiciously good

If someone could create insider information and protect traders from prosecution, they’d basically have a perpetual motion cash machine for themselves and their friends.

Trump’s tariff machine

Trump has solved insider trading with tariff policy. He’s discovered that on any day, he can cause the market to drop by announcing new tariffs, or cause the market to rally by announcing a pause on tariffs.

On April 2, his announcement of global tariffs caused the S&P 500 to drop 9% on the week.
On April 9, he announced a 90-day tariff pause. The S&P 500 shot up 9.5% in a day.

Trump controls the DOJ and SEC, which are the two agencies that prosecute insider trading. Per the Supreme Court, Trump is immune to criminal prosecution for any “official” act as President.

There’s no proof that Trump tipped off any associates or allies before he decided to announce his tariff pause. However, we do know that:

• 85 million equity option contracts were traded on the day of the tariff pause - an all-time high

• Oval Office footage showed Trump joking with billionaires about how many billions they made on the day of the tariff pause

• Historically, Trump is not a guy who likes to leave money on the table or not commit crimes when they are available to him

And Bessent, too

It’s not just Trump. On April 22, Treasury Secretary Scott Bessent held a closed-door summit with some of the world’s largest institutional investors, including JPMorgan, Fidelity, and Point72. Bessent told the attendees that the tariff standoff was “unsustainable” and would de-escalate soon. After the lunch, option buying spiked again, and the market jumped 2.6%.

Presidents used to be very cautious when talking about markets because they didn’t want their words to cause ripples. Trump has now discovered a simple algorithm that makes him the most powerful market actor the world has ever seen. He threatens tariffs, markets go down. He eases off on tariffs, markets go up. Nothing is stopping him from telling his friends an hour ahead of time and giving them an opportunity to make 10x their money.

Welcome to the era of Presidential Insider Trading.

World Liberty Financial is a do-nothing crypto vehicle owned by Trump and his family. Seventy-five percent of all dollars spent on World Liberty Financial go directly to Trump’s pocket, with no restrictions on his spending! Justin Sun plunged $75M into $WLFI to end his fraud investigation, and he’s never looked better!

Favored by luminaries like Jeff Bezos and Tim Cook, a donation to the Trump Inaugural Fund is pure class. An estimated 30% goes to Trump, mostly via spending at his properties and consulting fees for his family and friends.

Note: This item is seasonal and is currently UNAVAILABLE. Check back in November 2028.

TrumpCoin is enjoyed by degens everywhere. This is one of only two crypto pumps officially sanctioned by the President of the United States. A 2% fee goes to Trump on every transaction. Trump owns 80% of the supply, which means at the time of writing this, Trump has $3B in wealth purely from this coin. Let’s make that number go up!

Show your loyalty without breaking the bank. Owned by thousands of Trump voters who don’t realize they’re paying $19.99 a month or don’t know how to opt out of the payment plan. Seventy-five percent of proceeds go to Trump’s legal bills.

For those who want to give directly to Trump but don’t want to worry about losing your crypto key, there’s $DJT. Used by billionaire Jeff Yass to keep TikTok alive. Fifty-seven percent of all proceeds go to Trump with no restrictions on his spending. However, it’s difficult for Trump to sell shares without making a stir, so $DJT’s not quite so liquid for DJT.

First he sent the information to a group chat that included the editor-in-chief of a magazine called The Atlantic, who was apparently added to the chat by mistake. Then he sent the information to a group chat that he created that included his wife and brother.

The attack plan that he shared included the type of aircraft and the exact flight schedule for a targeted strike against Houthi militants in Yemen.

Putting classified national defense information information onto your personal device is a crime punishable by up to 5 years in prison. Sharing classified national defense information with unauthorized people is a crime punishable by up to 10 years in prison.

The US military has historically taken the security of classified information very seriously, so it charges these cases aggressively. In 2023, a 21-year-old airman was found to have posted a number of classified documents on his Discord to show off for his friends. He was sentenced to 15 years in prison.

Any uniformed officer who had committed the offenses that Hegseth committed would be charged under the Espionage Act.

High-level national security officials like the Secretary of Defense are provided with government-issued phones that are loaded with secure government-run communications programs and cannot install unauthorized applications from the internet. Hegseth, along with the Vice President, Secretary of State, National Security Advisor, and other officials who were part of the first group chat, chose to communicate classified information via their non-secure personal phones, rather than their government phones. This is a crime, and for good reason.

It is far more likely than not that Hegseth’s personal phone has been hacked by hostile governments or actors. If you’re not sure about that, consider the fact that if you use the internet, you’re probably aware that you have been hacked. Imagine how many times you’ve been hacked without your knowledge. You probably don’t have access to state secrets. Imagine how much more incentive there is for hackers, including both freelancers and those who are employed by hostile governments, to hack Pete Hegseth than there is to hack you.

By sharing imminent attack plans, Hegseth risked American lives and interests. Houthi militants are capable of shooting down aircraft. Giving them the exact flight schedule and aircraft type that will be used in an upcoming attack makes their job much easier. Houthis could have alerted the target and gotten him to safety. They could have put civilians into the target area and recorded as American missiles killed civilians. They could have used their knowledge of Hegseth’s personal security breach to blackmail him.

There have been no consequences to Pete Hegseth. He has not been charged with a crime. Neither the President nor Republicans in Congress have begun proceedings to remove him from office, and they are not expected to do so.

Three days later, Abrego Garcia, who still had not been accused of a crime, was deported to El Salvador and put in a prison called CECOT. CECOT is a death camp. It’s a prison so horrible that the country’s justice minister has said that the only way to get out is in a coffin.

Abrego Garcia’s wife brought the case to a court. In court, the Trump Administration admitted that they had made an “administrative error”, and that Abrego Garcia should not have been deported, as a previous court order had given him immunity to deportation due to threats to his life in El Salvador.

A federal district court ordered the government to bring him back. The Administration refused. They appealed the decision. It went all the way to the Supreme Court, who ruled 9-0 that the government had to facilitate Abrego Garcia’s return to the US.

Even after that ruling, the Trump Administration refused to do it.

What the Administration is now telling the courts is that they can’t bring Abrego Garcia back. They’re saying that he’s outside of US jurisdiction, so there’s nothing they can do - a position that the Supreme Court says “refutes itself”.

Everybody knows that’s a lie. Trump could go on Twitter or talk into any microphone in the world and say “Send Kilmar Abrego Garcia home”, and El Salvador would do it.

El Salvador is not a hostile state. El Salvador is Trump’s client. The country is run by a 43 year old president named Nayib Bukele, who calls himself “the world’s coolest dictator”. Trump’s government is paying Bukele at least $6M for him to put Abrego Garcia, and anyone else we deport, into the death camp known as CECOT - a prison that nobody gets out of.

Trump doesn’t want Bukele to send people back. That’s what he’s paying him for. He’s paying him to hold a blacksite that’s outside of US jurisdiction, where Trump can send people without accusing them of crimes and keep them locked up forever.

On April 14, when Trump met with Bukele in the Oval Office, Trump said that he wants to send US citizens to El Salvador. Caught on a hot mic, Trump was heard telling Bukele that he needs him to build about 5 more prisons so he can send “home-growns” there.

Catan is a game about trading. It would teach Trump more about international trade policy than any of his previous experiences in political and business life have, because Trump has lived his life in a very unusual way that’s dependent on strategies that are not available in Catan or international trade policy.

Here are the three lessons Trump would learn from Catan:

1. Trade is usually positive-sum

2. Trading relationships are long-term

3. People will stop trading with you if you’re a bad partner

Trade is how wealth is created

The fact that trade is positive-sum is the most fundamental truth in capitalism and the world economy. It is the source of basically all global wealth.

Trump does not understand that trade is positive-sum. He keeps telling us this. It’s his whole stated reason for having tariffs:

“A deficit is a loss. We’re going to have surpluses, or we’re, at worst, going to be breaking even.” (April 7, 2025)

“We have trade deficits. We lose with everybody.” (October 1, 2018)

Presumably the reason Trump doesn’t understand that trade is a positive-sum, wealth-building activity is that it’s not how he got rich. Trump got rich by inheriting hundreds of millions of dollars from his father.

And then he became really rich after he became President by doing an unusual kind of trade called bribery. Trump became a billionaire by getting people to pay him money via shell entities that he owns ($TRUMP coin, World Liberty Financial, Trump Media & Technology Group) in exchange for giving them government favors, like an agreement not to prosecute them for fraud.

In Catan, you’re not allowed to inherit resources from your dad or use the threat of legal prosecution to receive bribes from other players. Instead, you have to trade one resource for another. If Trump were forced to play Catan, he would see that sometimes he has a lot more sheep than he needs, but not enough brick. If he traded a sheep for a brick, he would be better off. And if a trading partner agreed, presumably they’d be better off too.

When we trade dollars for bananas, we are doing it because it’s good for us. It also happens to be good for Guatemala, Ecuador, and Costa Rica. Trade is positive-sum. Catan teaches that, but Trump’s life hasn’t.

Trade relationships are long-term. Reputation matters.

Part of Trump’s schtick is that he’s a good “dealmaker”. What he means by that is that when the deal is done, the other guy says “That was painful, I never wanna do a deal with that guy again”.

Let’s assume that’s true, and let’s set aside the possibility that the people who Trump makes deals with don’t want to deal with him again because Trump lied and cheated. Let’s pretend that Trump doesn’t lie or cheat, but always exerts maximum leverage and extracts maximum pain from his counterparty. Trump really does “win” the deal.

The thing about deals is that they’re not always one-off. Often, dealmakers make multiple deals with the same person. Or they make deals with people who are networked with each other. If you make an exploitative deal with one guy, that guy’s not going to want to make a deal with you again, and neither will anyone else who knows about your reputation.

Catan teaches you that Trump’s brand of dealmaking is a losing strategy.

You can imagine Trump saying to Eric “I’ll give you a sheep but you need to give me two brick”. And then Eric asks Ivanka, and Ivanka says “I’ll do one sheep for one brick”. Eric makes the deal with Ivanka.

A one-for-one deal isn’t a “win” In Trump’s mind. He needs a two-for-one. He needs to “win the deal”. Trump’s not gonna get a lot of deals with that strategy. Maybe when Eric’s really desperate, he’ll do the two-for-one. But that means Trump’s only getting a deal every now and then. Meanwhile, Ivanka’s doing trades every turn. Trump’s getting a “better deal”, but that’s not the metric that counts. What counts is the overall growth in resources and capacity. In Catan, being a good dealmaker isn’t about making one painful, exploitative deal; it’s about being a good deal partner so that people will keep trading with you.

The same is true in international trade. Maybe Trump can strong-arm Vietnam into a deal that’s painful for Vietnam. He’ll call it a big win. But over time, Vietnam will trade less and less with America because they’re getting a better deal with China. Or Canada. Or Germany. All the other countries will learn from our deal with Vietnam that we’re not a good deal partner.

Unlike in the rest of Trump’s life, there’s not another sucker who’s gonna show up. There are only so many countries. When we show ourselves to be a bad, unreliable, dishonest trade partner, they’ll trade with us less and less. The countries that make lots of fair deals will get to do a lot of trade, which is how you build wealth. We won’t.

I just hope someone finds a way to sit Trump down for a few sessions of Catan, because it really could be the difference between 3% GDP growth and a global recession.

You’ve already done the research, marshaled the data that lead you to an informed opinion, and strung it together to construct a data-backed map of the world, which you’re now helpfully sharing with your audience. Why would you not include the sources that got you there? Why show them only the end product, when instead you could invite them to take the whole journey themselves so they will come to the same conclusion as you using the same, ground-truth data sources? 

This is what the audience wants. Studies show that as the number of sources hyperlinked in an online piece of writing increases, the average length of reading time increases. 

It’s not just that the audience is more engaged. Pieces with frequent hyperlinks are more likely to be shared and readers are more likely to retain what they learned in the piece. Pieces with frequent hyperlinks are judged as more accurate and, because readers click on an average of 18% of all the hyperlinks included in writing in major publications like the New York Times and the Wall Street Journal, readers of heavily-hyperlinked pieces are actually more informed. 

I’m lying, of course.

None of the links in the text above support my arguments. They’re all pictures of cats. 

Like everything on the internet, hyperlinking is a double-edged sword. Yes, it’s great that you can point to the data that support your argument. That can be useful. But most hyperlinks don’t get clicked on. And when they do get clicked on, the reader is unlikely to read the full source, evaluate its validity, and track back to the original piece to determine whether it does in fact support the argument that the author claims it supports. 

Note that I didn’t include hyperlinks in any of the claims I made above. I didn’t bother to look for any studies that confirm or deny what I’m saying. There may be some, although I doubt it. And yet you probably suspect that my statements from the previous paragraph are true because you thought about them and evaluated whether they were true. If I had included hyperlinks, you would probably have taken a shortcut and subconsciously decided that you didn’t need to evaluate my words because the existence of the hyperlink is itself a statement that says “this thing is true, I got it from a good source, here, I’ll show you”. That works even if you don’t click on the hyperlink. 

Again, I’m not citing a source for that claim. There are certainly plenty of findings in psychology that do support this claim (e.g., priming, confirmation bias, anchoring, cognitive load theory, mere exposure, authority bias). If you want, you can look for them and come to your own opinion. If I had given you a source, you probably wouldn’t click on it anyway. And if you did, you’d give it credibility that it doesn’t deserve. 

“I didn’t leave the Democratic Party. The party left me.”

• Ronald Reagan, Elon Musk (in meme form)

Conservatives really like this idea. They’ve always been reasonable, moderate people, but now the liberal party has gone too far. 

There’s some truth to it!

In 1857, reasonable, moderate people believed that slavery was distasteful, but an important part of the economy. You could believe that and have a home in the liberal party in America. But 10 years later, you couldn’t. 

The same was true if you opposed suffrage in 1915 or civil rights for black Americans in 1959.

The liberal party in America does leave people. It renews itself. It champions ideas that 10 years ago seemed unthinkably progressive, and that in 20 years everybody will cause everyone to say “Yes, of course I believe that. Always have.” 

The Democratic Party did leave Elon Musk, and happily. It has standards. 

Liberalism is about fighting for people who have less. That’s what it’s always been. 

Eventually, most rich people find themselves on the other end of that trade. Liberalism has gone too far, in their minds, when they realize that whatever social benefit they got from being seen as liberal is outweighed by the tax cuts or deregulation that conservatives are offering. 

It’s not always about wealth. Sometimes a person lets their mask slip in public and liberals say “hey, he’s not one of us”. Liberals have a strong immune response. They don’t want to keep an impostor around.

“The Democratic Party left me” is meant as an indictment of liberals. I see it the other way. I like being part of a movement that has standards. I like that people in my own party will criticize me if they think I’m doing something wrong. I like that there are things a person can do that make them persona non grata.

In 2017, stories about a popular Democratic Senator began to surface. He’d been accused years earlier of unwanted kissing when he was performing a USO show in Afghanistan. There were photographs of him miming suggestively in the direction of a woman sleeping nearby. For these transgressions, the majority of Democratic Senators called for his resignation. Al Franken resigned from office because he no longer had the support of his party. A strong immune response indeed.

Look at the other side of the ledger. There’s a Republican politician who bragged on tape about grabbing women’s genitals without their consent. A court found him liable of sexual abuse. While his wife was at home taking care of their infant son, he had an affair with a porn star, then paid her $150,000 to keep quiet about it. He ran a fake university and was charged $25M for the fraud he committed. That man’s Donald Trump, and he’s the frontrunner for the Republican nomination for President. 

Call me crazy, but I like the party that might leave me one day. Not the one that could never, no matter what I do, summon the moral courage to demand better

But not for the reason you might hope. 

We like to think that catching a politician in a swirl of hypocrisy will expose them and hurt them politically. But it doesn’t. Never. There’s no politician in modern American history who has suffered any kind of setback because he was caught being hypocritical.

Hypocrisy doesn’t work as a sword. But it’s the best shield in the world. 

You can get away with anything in politics as long as you accuse the other side of having done the same thing. 

Let me give you an example. 

In February 2016, Supreme Court Justice Antonin Scalia died, leaving a vacancy to be filled by the Democratic President Barack Obama. 

Appointing Supreme Court Justices is one of the most important powers a President has. Nobody had ever contested that power. 

So it was a surprise when Mitch McConnell, the Republican Senate Majority Leader, announced that for the first time in the country’s history the Senate was simply not going to allow a confirmation vote to fill the vacancy on the Supreme Court, and they’d instead “let the voters decide” in the Presidential election 9 months away.

McConnell’s justification is a frightening example of the hypocrisy shield. 

McConnell said that he would block the confirmation of a Supreme Court Justice because in June 1992, 24 years earlier, then-Senator Joe Biden had given a speech on a Thursday afternoon, presumably to an audience of 50 C-SPAN watchers, floating the idea that if there were a vacancy in the next few months (which there wasn’t) during the heat of the Presidential election, the nomination and confirmation of a Supreme Court Justice should be delayed until after an election. 

McConnell seized on the mere suggestion a quarter-century earlier by a politician who happened to belong to the opposing party to justify his unprecedented decision in 2016. To top it off, McConnell said he was following “the Biden Rule”, a phrase he’d just invented. 

In September 2020, there was another Supreme Court vacancy. McConnell was still the Senate Majority Leader, but now there was a Republican in office. And what a surprise – this time McConnell decided he didn’t care about the Biden rule, and he went ahead and confirmed Trump’s conservative nominee to the Supreme Court right away. 

Hypocrisy is a shield, not a sword. Accusing someone of atrocities lets you get away with atrocities, but being accused of hypocrisy, even accurately, can never hurt you in American politics.

McConnell pulled off the heist simply by alleging that the opposition would do the same thing if they were in his position.

Yet when McConnell’s actual hypocrisy came to light in 2020, he faced no consequences.

Hypocrisy doesn’t only work this way in politics. 

Imagine you’re in a car with your friend, and you’re texting while you drive. Your friend says “you shouldn’t text and drive”. Your friend is of course correct. Texting and driving is dangerous for everyone on the road. 

But you hold the trump card if you can say “but you were texting and driving earlier today”. Suddenly, you can get away with it. Not because it’s right. Nobody thinks it’s right. But because the other side is guilty of it too, so they don’t have standing to oppose you. 

Politics takes it to the next level. Because instead of you arguing with your friend, it’s you and all of your friends arguing with a group of people you hate. 

So you don’t have to prove that someone else on the other side did the exact same thing in the exact same circumstance. You just have to allege that someone in the opposition has done something similar to what you now want to do, or would do something similar if given the chance. 

In American politics, you no longer have to say that what you’re doing is “right” in an objective sense. You just have to allege that the other side would do it too, so it’s only “fair” that you do it right now, even if the thing is objectively wrong. 

When the hypocrisy shield is the reason for political action, it makes it impossible to determine what conduct we as a society actually want. Do elected officials, or voters, believe that it’s right to confirm a Supreme Court nominee during an election year or not? That issue has not been discussed.  Instead, the decision was made by whether the party in power chose to use the hypocrisy shield or not. 

And one more thing. 

Politicians know about this. They know they can get away with anything if they say the other side did it first, or would do it. So watch out for that. If an American politician is going on about something awful the other side did, you should think about whether that politician is priming you to accept the awful behavior they’re about to do.

One of the benefits of building a company in an emerging market is that sometimes, you get to design something new without the burdens of legacy systems and regulatory debris. 

We invested in Caafisom because we believe the company will play an important role in building a high-quality, patient-centric healthcare system in Somaliland and East Africa. 

Caafisom is live in about 40 hospitals and has served about 8,000 patients in the Somali region. Through its free mobile app, Caafisom helps patients, doctors, and hospitals streamline care delivery by managing patient records, appointment scheduling, and medication information in a simple, transparent way.

Medical capacity in Somaliland is still quite low – both in terms of equipment and trained personnel. As Caafisom helps drive efficiency within Somaliland, the company also makes it easier for patients to travel overseas to receive medical care that’s not yet available in-country. 

Tens of thousands of patients are already making those medical trips every year, and most people are doing them without much help. Caafisom makes it 10x easier for patients to travel abroad for essential care by coordinating with overseas hospitals on patients’ behalf and arranging visas, transportation, and accommodation for patients and their families.

Caafisom is our first investment in the Somali region. The first of many. We’re thrilled to be supporting a company that’s already making lives better in East Africa.