One Chinese AI Model Wiped Out $1 Trillion In A Single Day — And They're Just Getting Started

The stock market has become one big bet on AI, and something just happened that makes that bet look a lot riskier. >> They've got $50 billion a year. They're spending $1.4 trillion a year. What do they think they're going to do? How are they going to make up the money? >> 80% of every dollar the US stock market has gained in value over the last 3 years has come from artificial intelligence. The 10 biggest companies in the S&P 500, nearly all of them riding on the AI trade, now make up over 40% of the entire index. If you own an index fund, you don't really own the broader market anymore. You own the AI bet. And what is it that happened that made all of this that much riskier? Coinbase just cut its AI bill nearly in half by moving its engineers off of America's top AI models and on to cheaper open-source ones that are proving to be almost as good. That should be a good thing though, right? Well, you have to consider this. The arrival of open-source models to the US market is not simply the result of competition. These new models are actually the result of a geopolitical battle that's putting the stability of the US AI industry as a whole at risk. The question is, how could something that's good for consumers actually be bad for the industry? That's a critical question that we have to answer because it's going to have a major impact on you as both an investor and as someone that's just impacted by the health of the US economy as a whole. So, let's walk through what's really going on because there's something at play here that most people are missing. I've broken it into four parts so it's easy to see how the mechanism works, where the increased risk is coming from, and ultimately what you need to do to protect yourself. Okay, the first thing to note is that AI is better understood as an arms [music] race than as a traditional industry, and the US AI industry is right now under assault from China. Now, global competition is nothing new, obviously, and neither is competition from China, but what is new is that both China and the US understand the potential winner-take-all dynamic of AI. That's why the US has leveraged export controls on the chips that are necessary for AI training, and why China is now attempting to steal US AI technology and undermine the financial viability of our AI market as a whole. [music] China fully understands just how vulnerable the US AI industry is to investor confidence and the absolutely staggering, crushing debt obligations that are threatening to take down the entire current wave of AI investors, even without China. Here's the strategy that China's using. For years, America denied China access to the most advanced chips, because without advanced chips, there's just no way for them to build competitive frontier models. This meant that China could not go against the US in a head-to-head competition, but given the importance of AI, they couldn't just sit out of the race. That's not an option. So, certainly not if they want to be a global hegemon. Instead, they focused on winning the race via computational efficiency. To do that, they began leveraging our own models against us, potentially illegally, and working to undercut the one thing the entire US AI industry absolutely depends on, revenue growth. The way that they have built deeply efficient models from a cost perspective is something called distillation. The way that it works is that you build a simple AI model and train it on the outputs of a massive, well-trained frontier model like ChatGPT or Claude. You create tens of thousands of fake accounts, >> [music] >> send millions of queries to the frontier model, you capture its answers, and you use the distilled set of patterns to teach a smaller, cheaper model to behave the same way as the large, expensive model. Because you don't have to build a physical data center that's big enough and robust enough to read all of human knowledge and then distill the patterns from that gigantic corpus of data, but you still get the distilled patterns, you can build a highly competent model for a tiny fraction of the cost through distillation. Now, technically, distillation is a standard. It's a legal technique. Every major lab does it to its own models. The problem starts when a rival nation is doing it to a competitor's model that they're not supposed to have access to in the first place. Now, this is not me making this up. In February, Anthropic accused three Chinese labs, DeepSeek, Moonshot, and MiniMax, of doing exactly this, alleging more than 16 million queries were run through roughly 24,000 fake accounts, and in April, the White House directly accused China of running, and this is a quote, "deliberate [music] industrial-scale campaigns to steal American AI models." Then, on June 10th of this year, Anthropic sent a letter to the Senate Banking Committee accusing Alibaba of the largest such effort to date. According to Anthropic, China used these queries as a way to effectively clone Claude's intelligence. Alibaba denies it, of course, and so far, these are just allegations, but China's track record of stealing foreign IP is well-documented. [music] Now, in fairness, it really is a brilliant way to get around advanced chip export controls, >> [music] >> and it seems to be working because the cheaper models produced by China are proving to be good enough to attract some of America's largest companies >> [music] >> as customers are discovering how insanely expensive frontier models >> [music] >> really are at this stage in their development. Uber spent its entire 2026 budget for AI coding tools by April and ended up having to cap each engineer at about $1,500 a month. Meta sent around an internal memo warning of an exponential increase in AI spending. Amazon scrapped an internal leaderboard ranking employees by AI usage because people were gaming it and running the bill sky-high without commensurate improvements in the product. And a KPMG survey found only 26% of companies have a clear view of what they're actually spending on AI full stop. When something delivers real value but comes at this kind of extraordinary cost, you can expect people are going to start shopping around for cheaper alternatives just as China knew they would. Coinbase cut its AI bill nearly in half by routing its engineers to Chinese open-source models, one of them made by Moonshot, the very lab Anthropic named in their February complaint. A company called Lindy also moved off of Anthropic's Claude and onto another Chinese open-source model when its AI cost grew larger than its payroll. So, the reason is simple. The Chinese open-source models run about five times cheaper than the top US model while scoring within a few points on standard AI performance benchmarks. In a price-to-value ratio, the Chinese models are the clear winner. This presents an extreme danger for the US AI market, which is currently playing a game of chicken with the massive amount of debt it's had to accrue to build out the infrastructure required to create and scale frontier models. Given the size of the debt, the Chinese open-source models don't have to siphon much revenue away from the US industry to do [music] extreme damage. Not just to our AI industry, but to our economy as a whole because it's so tied to AI. The US is aware of the problem. China's obviously aware of the dynamic, and the House has already opened an investigation into the Chinese model makers, but the cost pressure is so real that we may not have the time for a diplomatic or legal solution to play out. There are signs that the market may have already noticed that revenues are not keeping up with costs in AI. SpaceX has already given back most of its post-IPO gains, falling about 32% from its peak within just 2 weeks of going public. On June 23rd, a broad tech sell-off erased close to $700 billion in a single morning. Oracle had its worst week since the dot-com crash of 2001 on concerns over how the AI build-out is being financed, more on that later, and the big AI stocks have started moving apart, likely due to investors beginning to separate the companies with strong enough technology and hopefully sufficient revenue growth from the ones that are just funding everything on debt and not showing the indication that they're going to grow rapidly enough. The thing everyone has to factor into how invested they get into the AI sector is that the pressure on AI revenue isn't likely to be a passing price war. [music] A huge part of it is a deliberate strategy in the US versus China Cold War and the hyper-consequential battle for AI supremacy. So, the question becomes how long can China continue to force prices down? They don't have the same pressures on them that we do. Given the top-down authoritarian control that G has, he can just point the Chinese economy wherever he wants to and he can redirect funds however he needs them to be redirected to allow Chinese companies to build open-source models with or without revenue and drive the revenue potential of US models down. And given that the distillation strategy is so much cheaper, he doesn't have to worry about the mega infrastructure build-out that the US currently has to manage. All right, the second thing you need to understand about the dangers facing the USA AI industry is that even without China, the revenue just isn't coming in fast enough. MIT recently studied how generative AI is actually performing inside of companies. They found that 95% of corporate AI projects that they looked into produced no measurable impact on profits, not even small returns. Now, it's not a big deal for a young industry with low startup costs, but it's a major problem for a technology with the most expensive infrastructure buildout possibly in human history, and it's all financed on debt. Debt already puts you on a ticking clock, but when you add hype to the mix, the clock speeds up. Whenever something creates euphoria in the market in the psychotic way that AI has, retail investors become completely irrational. They ape in without looking at the fundamentals, and they expect big returns fast, and they often buy in on leverage creating a second debt-based danger zone. One for the company who is building using debt, and one for the investor who's investing on companies using debt by using debt. In both cases, the debt just makes it harder for the participant to withstand the volatility that inherently exists in the markets, especially markets that are completely detached from fundamentals like AI. A normal technology company reinvests somewhere between 5 and 20% of its revenue back into building things, but with AI, that number has gotten completely insane. Oracle is now spending the equivalent of 57% of its revenue on the AI buildout. Microsoft is around 45%, and across the biggest players, AI infrastructure is eating up close to 94% of the cash that their core businesses are throwing off. There's no cushion left. Sequoia, one of the most successful venture firms in history, tried to estimate the size of this hole. Their belief is that the AI industry now needs to generate about $600 billion in brand new annual revenue just to justify what's already been spent. And the industry is nowhere near generating that kind of revenue. And that number isn't shrinking as the technology matures. The number is growing. The problem is that history tells us that kind of revenue is not going to come in fast enough. Not to save the current crop of investors from getting hammered. And given how systemically important AI has become to the US market, there could be a contagion effect as this initial round of investors gets hammered and everyone else gets hammered by proxy. I went into a ton of detail on this exact problem in this video here if you want more info. We'll get back to the show in a moment, but first, let's talk about the thing your business just can't survive without. I go live 3 days a week at 7:00 a.m. And every morning you guys, you incredible people out there show up. You're there, you're ready. And if my connection drops in the middle of that live stream, that moment is gone forever. You do not get a second take when it comes to live content. And I know a lot of you are in the same position. Whether you're running live events, processing transactions, or managing a remote team, your business depends on staying connected. Not sometimes, every [music] time. That's why I trust AT&T Business. They're built for business owners who can't afford downtime. AT&T Business is a reliable provider for small business owners. For small business month, we celebrate small businesses by helping them run better. >> [music] >> This means reliable uptime, easy switching, smart communications. Impact Theory is powered by AT&T Business, built to work. Get AT&T Business at business.att.com. We will be right back to the show, but first, let's talk about your standards. [music] Most people have two sets of standards, the ones they talk about and then the ones they actually live by when nobody's watching. When you're busy, moving fast, and hungry, those standards [music] get negotiated down. You grab whatever is closest and you tell yourself it doesn't count, [music] but it does. Your body is keeping score, trust me. That's why I keep PaleoValley beef sticks on hand whenever [music] I'm at risk not being able to get whole food because I refuse to let hunger be the thing that breaks my standard. [music] These are 100% grass-fed and grass-finished beef, zero sugar, no artificial preservatives, no soy, gluten, or corn, just beef and organic spices with 6 g of protein per stick. When you're running between meetings, traveling, or just don't have time to think about food, reach [music] for a PaleoValley beef stick. Right now, you can get 30 beef sticks for just $36. That's real nutrition for barely over a dollar per stick. Click the link below to get your 30 beef sticks for just $36. [music] And now, let's get back to the show. For now, suffice it to say that revolutionary new technologies with big infrastructure build-outs typically bankrupt the first wave of investors. It happened with the canals in the UK, the railways in the UK, and again the railways in the US, and with the internet. They all followed the same pattern. The technology went on to be as amazing as everyone thought it would be, but the revenue didn't come in fast enough to save those early investors. They all got wiped out. Hype just makes people get way too far out over their skis. They do it on leverage, and that's where the problem arises. And the extra bad news for AI is that at least in those previous cases that I just mentioned, the gap between the money spent in the infrastructure build out and the money earned by the new technology over time narrowed as the customer demand caught up. With AI, analysts at Man Group and elsewhere are going to great lengths to point out that it's doing the opposite. With AI, spending is accelerating while the price companies can charge for its usage is falling, partly because of the open-source models that China is pushing onto the market, just like we covered. So, even as more people use AI, it's entirely possible that the revenue won't climb fast enough for the industry to remain viable in its current composition. And the market history shows that the composition of an industry can change very suddenly. Just look at what happened during the dot-com bubble burst. Companies that seemed like can't-miss investments went bankrupt almost overnight and got replaced by the companies that would eventually win the internet race, but it took time. To complicate things even more, the revenue and costs of AI may actually be worse than the industry is trying to let on. A meaningful amount of revenue is actually circular payments, and according to Michael Burry, the true cost of the chips are being hidden behind an accounting trick. Let me explain. Nvidia agreed to invest up to a hundred billion dollars in OpenAI, but that money is largely just flowing right back into Nvidia whenever OpenAI buys the chips that it needs. Nvidia also owns a piece of a cloud company called CoreWeave. Nvidia has committed billions of dollars to buy CoreWeave's unused capacity, So, their own spending makes their own investment worth more, sort of. The investment firm GMO compared this whole arrangement to the circular financing of the dot-com bubble era companies that made them so fragile. And AI's fragility goes way beyond just circular payments. The four biggest US cloud companies are sitting on about $2.1 trillion in future revenue commitments. But by one analysis, roughly half of that is owed by Open AI and Anthropic alone, two companies that are deeply in the red. The whole industry starts looking very precarious when you put that all together with Michael Burry's warning that clocking the life cycle of a data center chip at 5 to 6 years instead of the more realistic 2 to 3 years, that could be hiding more than $175 billion in losses already sustained. Even the insiders are starting to get cautious. Microsoft has stepped back from its commitment to supply all of Open AI's computing power, letting other companies absorb some of that exposure instead. And honestly, it's a smart move because we're already seeing how a single Chinese open-source model can crack investor confidence. In January of 2025, China's Deep Seek model hit the US market like a meteorite. Overnight, investors started questioning the whole US is going to lead the world in AI narrative. And the result was a trillion dollars of US AI market value gone in a single day. It smacked Nvidia so hard, they set the record for the largest single-day loss by one company ever. That's what one low-cost Chinese model did. What's going to happen over time as they put more and more on the market, especially when the cost of carrying the debt is just rising? The Fed was expected to cut rates this year, but instead, due to rising inflation from Iran and AI itself, the Fed is now signaling future hikes. Plus, timing couldn't be worse. Japan Central Bank just raised its rate to the highest level in 31 years, which pulls back hard on global liquidity. The liquidity that high-risk markets like AI have been able to count on for years and are most likely going to need if things don't change. Higher rates make every dollar of infrastructure debt more expensive to carry, and they punish pay-off later companies like OpenAI and Anthropic the hardest. So, the runway for increased customer adoption to drive sufficient enough revenue to save the AI companies from their own debt is getting shorter by the day. Now, none of that is to say that AI demand isn't real or that AI won't change the entire world. It is real, and it almost certainly will change the entire world. But, that doesn't mean that the current investors won't get wiped out before that plays out. Okay. We've got the cheap Chinese models being trained, potentially illegally, by the US frontier models, and then deployed as a way to weaken the much-needed revenue stream required for the US AI companies to overcome the crushing debt that they've had to take on, combined with AI taking longer to deliver on its promises than anyone wants to admit, plus debt getting even more expensive, plus liquidity being pulled out of the market, plus investor confidence starting to look less and less steady, and we have a historical record of revenues taking way too long to save the hype-driven early investors. On top of all that, AI has become so systemically important that as you wipe out that first layer of investors, you might take the economy down with it and find yourself in a recession or a depression. Now, the third thing that we must metabolize on top of all of that to understand this moment in AI investing is how the AI companies themselves are responding to these pressures and what their responses are likely to mean for you. To state it plainly, the AI companies are acting like predictable and seeking the shelter of the US government and by proxy, you, the US taxpayer. Last November, AI floated the idea that the federal government should backstop its financing, that taxpayers should help guarantee the debt behind its buildout. Thankfully, the backlash was instantaneous and the company's CFO walked the comment back within hours saying she had, quote, muddied the point. But, from where I'm sitting, this was not a slip. This is a tried and true strategy that should really piss people off and it happens [music] all the time. This is like a preemptive bank bailout and it's a pattern of behavior that we see over and over. Back in March of 2025, Open AI had already sent a letter to the White House asking for tax credits, loans, and other vehicles the US government can direct towards companies building AI infrastructure. Now, I actually don't have a beef with the government deciding that an industry is so strategically important it's worth defending or helping to build, but it is a very slippery slope from generic incentives to protect an industry to creating regulatory moats and protecting individual companies. Competition is necessary if an industry is going to innovate and thrive in the long run and we're already running the risk, that given AI's systemic importance, that it will get protected to the point of codifying the winners and artificially lowering competition. So, we have to be paranoid. Now, to be fair, Open AI's leadership has publicly denied wanting a bailout. CEO Sam Altman said directly that OpenAI, and this is a quote, does not have or want government guarantees, and the taxpayers shouldn't have to rescue companies that make bad bets. And the White House's AI czar, David Sacks, has also said that there will be no federal bailout for AI. However, that same AI czar also pointed out what I've been saying here, that AI investment now accounts for half of America's economic growth, and that a reversal of that growth in AI would risk a full-on recession. And Altman has mused that when something gets big enough, quote, the government is the insurer of last resort. So, you will be forgiven if you remain extraordinarily paranoid about bailouts and regulatory capture. It is all too clear that these are not the statements of an industry that's confident that it can pay its own way. They're the early seeds being planted by industry insiders who are building the case that AI is just too important to be allowed to fail. And to be honest, I agree with them. But, the question becomes, how do you protect the industry while aggressively avoiding what one policy expert called a request for regulatory capture in its worst form, especially knowing that when you can't win on price, the next best thing is to make it harder for competitors to exist at all. And the easiest way to do that is create a narrative that the competition is dangerous. The government is going to have to find a way to thread that needle, because Anthropic is already going hard in the paint to convince the world that open-source models are dangerous. >> In terms of the scaling of open-source models, I think it's going down a very dangerous path, and if the path continues, I think we could get to a very dangerous place. >> Anthropic CEO Dario Amodei beats the we need regulation to protect against dangerous AI drum constantly. Not long ago, he published an essay just calling for binding government regulation of AI. Mandatory safety testing by outside parties, the way we test cars and airplanes, and to grant the government the power to block or reverse the release of an AI model it considers dangerous. Now, there's no doubt. There's a real argument here. AI is very powerful. But, there's also extreme danger that this is a one-way ticket to the kind of regulatory capture that ends up working against the working class because it makes everything worse and more expensive. AI is powerful. And its rollout needs to be thoughtful. But, the last people who should be making that decision are ignorant government officials who make millions of dollars from insider trading on their own decisions and the company executives that stand to benefit financially from icing out the competition by getting in bed with said politicians. A smart startup or an open-source project can't compete when the government makes it impossible to get started. Major analysts and even some of Anthropic's own allies have pointed out that the same rules that make AI safer also conveniently raise the barrier to entry. And the hardest thing to regulate is exactly the kind of cheap open-source competition that is flowing out of China and eating into US revenue. Tough regulation sold as safety will only serve to crush the kind of company that will actually comply with sensible light-touch regulation and serve to innovate and drive costs down while taking quality up. So, the thing that companies like Anthropic are pushing for aren't likely to help you with China and will stop US companies from competing with them. But, none of that's going to matter to a large and growing segment of the public who just hate AI with an increasing ferocity. They just want it shut down. Surveys show that only about a quarter of Americans hold a positive view of AI and nearly half who actively view it negatively. 71% of Americans say they don't want an AI data center built near them. That's higher than the percentage of people who oppose a nuclear plant going in next to them. People are watching their power bills climb and blaming the data centers. And the AI build-out is sending the price of everyday electronics to the moon. Tim Cook called the rapid cost increase a 100-year flood and Elon Musk said it was the biggest price jump he's ever seen in anything. So, as a society, we're in this super weird place where we have a technology that a huge number of people hate, that has already become so systemically important both from a national security standpoint and an economic standpoint, that everyone should fear what happens if it fails. And the makers of that technology are under so much financial strain from the infrastructure build-out costs that they have to position themselves as being in need of regulation for safety reasons, while also doing everything they can to ensure they get bailed out if there's economic trouble. So, the lingering question is, what do you actually do in the face of all of this? Well, you start by understanding the game that's actually being played. All markets and all strategies are a game of risk and reward. The importance of risk in the AI calculus is a huge part of what makes this moment in AI investing so interesting and investing in general, to be honest, given how tied everything has become to the AI bet. Now, what do I mean by that? This is the fourth part that we have to get. If you're going to develop a coherent strategy moving forward, you have to identify where the risk lives and watch how banks and the companies themselves are spreading that risk around. Let's start with the banks. The banks making those AI loans are not naive. They know exactly how shaky the industry is given the amount of debt that has already been required and how much more is likely to be required in the future. They know how much of the companies' revenue are being swallowed up by the infrastructure buildout and they understand perfectly well the circular revenues and shipped appreciation schedules that are being put forward. Understanding risk is their job. And now it's your job, too. So, let's zoom in on the risk and look at what history tells us they're likely to do. This is what I call the 2008 playbook. They're likely to take the risky debt, slice it up, dress it up to look safe, and then sell it off. In fact, they're already doing it. And once packaged, it's going to get sold into private credit funds, insurance companies, and pension funds. In fact, they're already doing it. I walked through the mechanics of how this will play out in this video here. So, if you want all the sorted details, be sure to check it out. For now, I'll just note this. Hyperscalers took on well over a hundred billion dollars in new AI data center debt in 2025 alone. That's many times more than what they borrowed just the year before. A law firm tracking this found that lenders are now pooling those loans and selling pieces of them to pension funds and asset managers the way I just outlined. They're doing this so they can spread the risk around. And the Federal Reserve reported that major life insurers already have close to a trillion dollars tied up in private debt and pension funds have also started tying up money in the debt funds fueling companies like Meta and Oracle that have huge exposure to AI. Now, why are the banks working to distribute the risk in this particular fashion? Because spreading the risk across vulnerable parts of the economy where regular retail investors live does two things at once. It moves the risk off of the banks and onto you, and it spreads that risk so deeply through the financial system that if AI stumbles, the government will feel forced to step in and use money printing to backstop the AI industry, which is exactly the kind of bailout insurance the industry needs to move forward with these huge debts against comparatively small revenues and do it with confidence. So, as you assess your path forward, the most important thing you can do is know what you actually own. AI risk is not going to show up with a warning label. It's going to hide in your pension, in your insurance, in the quote-unquote safe bond fund, and and as I said at the start, in the index fund you thought helped you own the broad market, but now is really just getting the majority of its returns from AI. Go look. Ask where your yield is actually coming from, and then diversify. [music] It was a failure to do that that left so many people vulnerable to the housing collapse in 2008. It's critical to remember that you can believe passionately and with total conviction that AI is going to drive the bulk of returns of the next decade or more because it's just that transformational. But when a game of chicken is being played between debt and revenue timing, history has a brutal warning for investors. Having the right thesis, but getting the timing wrong is the same as just being outright [music] wrong. You can bet on the right technology and still get hammered by the timing and the hidden debt ownership, especially when China is trying to attack the industry by being pro-consumer. That's the wild part. China is going to drive the cost down. So customers are definitely going to bite. They're even going to be advantaged. But that increases the likelihood that the US industry, the one part of the US economy that is still working, will slow down. And if it does, it's going to crash headlong into the tsunami of debt obligations that have already built up in the system, creating the kind of fragility that we last saw in 2008. As an investor, you guys have to play the long game. No one can tell you what the precise timing of all of this is going to be. The people who have gotten wiped out historically are the ones who needed their money back in a year or two, or the ones who placed concentrated bets and then got the timing wrong, or just did it all on debt and couldn't survive the volatility. So be humble. Diversify way beyond AI, even if you have huge conviction about AI. Don't underestimate China. And really don't underestimate people's desire to get cheap products. All of that creates fragility for the market. The system really is rigged, but it's still winnable. >> [music] >> So play defensively. All right, if you guys want to see me explore ideas like this in real time, be sure to hit that subscribe button and join me Monday, Wednesday, and Friday as I go live and talk about topics just like this. Hope to see you there. Till next time, my friends. Be legendary. Take care. Peace. If you like this conversation, check out this episode to learn more. The US government owes 39 trillion dollars. There's no plan to pay it back. Instead, they plan to steal it all from you. I mean that literally.