Written by: Arthur Hayes
Compiled: Saoirse, Foresight News
Did you hear? Those big shots in the AI circle suddenly found their conscience and began to worry about the survival of humankind—because the near-silicon-based god they’ve built is about to be born. In fact, it has been not far from a silicon-based god for quite a while, but now, as it’s right at the final step, they suddenly started to reflect on the path to developing general artificial intelligence (AGI).
(Note: Silicon-God refers to that super general AI AGI that is supposedly about to be born—the term the author uses with sarcasm to question whether this grand narrative is just financing and lobbying rhetoric.)
That’s the story they tell to the outside world. But I, being naturally suspicious, checked the timeline. We’re already near the end of the third quarter, and Anthropic still hasn’t gone public. I can’t help wondering what their financials actually look like. The press releases repeatedly mention a rapidly rising annualized revenue curve—has it already started to slow? They claim that after stripping out all operating costs, the company is profitable. I can’t wait to read the S-1 filing they’re about to submit, to find out exactly how much it costs to deliver every Token to users. Also, how many customers are truly profitable for them? Is this profitable customer base expanding or shrinking? Unfortunately, I can’t get answers to these questions—because of “safety first.”
From my tone, readers can tell what I think: I believe Anthropic, OpenAI, and SpaceX claim they’re slowing general AI development under the banner of “safety first,” not out of concern for ordinary human well-being, but because of harsh economic reality—the market is unwilling to buy the AI products they sell at current prices in sufficient quantities. More specifically: demand for AI is very strong, but what people want is China pricing—prices that are only one percent of those in the United States.
When this “China pricing” shocks the American AI industry people who think highly of themselves, their first reaction is to shout: “But the products from China are of terrible quality.” And once the quality of Chinese models quickly catches up, they cry out: “China is distilling our models to make its own products.” The market doesn’t care why Chinese models are cheaper. The market only wants the cheapest intelligent services. So these AI people switch to lamenting: “We care about human safety, so we’re pausing R&D.” Sounds so noble... But later, they’ll still make demands: “We still need to beat China, so the government should step in—issue regulatory rules and keep funding this general AI race.”
These three leading U.S. AI labs use “safety first” as an excuse to slow down AGI R&D. This matters critically for financial markets and global fiat liquidity because their demand for computing power supports more than $1 trillion of investment-grade debt—and tens of billions more in low-credit-rated loans.
Show the procurement amounts OpenAI and Anthropic promised to the four largest U.S. cloud providers, and the portion of each provider’s unfulfilled revenue orders that this represents—so you can see how the two AI model companies bring huge long-term orders to the cloud vendors.
These AI labs, in total, have not generated any profits. Therefore, they need profitable technology firms like Nvidia, Broadcom, Google, and Microsoft to back them, providing off-balance-sheet guarantees for the debts associated with data center leasing and chip procurement. Follow-on procurement of chips and hardware depends on the AI labs continuously training cutting-edge models—the ones that are “almost, almost, almost about to become silicon-based gods”—and also handling inference requests for customers. But if “safety first” becomes the new core principle, spending to train new models may not go to zero, but it will definitely come down from its current high level. Companies will shift their focus toward improving efficiency in power-to-compute conversion, which means customers will spend less on computing power. In essence, “safety first” destroys the demand for computing power.
If AI capital expenditures were financed by operating cash flows, there wouldn’t be much to worry about. But the problem is that regardless of whether AI labs buy computing power or not, these trillions in debt still exist. Defaults won’t happen immediately, but once AI labs stop consuming computing power at the previously expected scale, the price of this debt will fall. The real core question is: who bought these debts, and did they buy with leverage? The answer is obvious—these speculators used leverage to buy this poor-quality debt. So who, ultimately, is the bag-holder?
Millions of American policyholders are, in fact, indirectly betting on the AI story. And “safety first,” without any buffer, will harm them. I originally didn’t fully understand this scheme, because Nick Nameth on Substack explained it in a very clear and thorough way. Next, I’ll write it in simple language for my readers in the crypto circle. The conclusion is: if you revalue AI-related debts at fair market value, a large part of the U.S. insurance industry is already insolvent. This leads to the core investment question in parts of the global economy dominated by reserve-based banks: the U.S. government has two choices—either, in the name of national security, act as the final buyer of computing power; or print money to rescue insurance companies mired in losses.
No matter which path we choose, Bitcoin holders and crypto investors are winners. If the government ignores market signals and insists on pouring money into developing this non–commercially profitable “silicon-based god,” then it will need to print money to fund such non-productive spending—inevitably fueling more financial speculation and pushing up the price of Bitcoin. If the government chooses to bail out the insurance industry, it will print money to take on bad AI debt, expanding the money supply and, in turn, pushing up the price of Bitcoin.
The rest of this article will break down how this mechanism works.
In the name of China
As long as you slap a national security label on it, the U.S. government can almost always find a reason for any action. Think about what the U.S. portrayed to the public as a threat after 9/11, then launched a global war on terror—how much destruction did that cause? This time, the imagined enemy being manufactured is people who hunker down to study math, steal top U.S. technology, and then sell products back to the United States at one percent of the price. To defeat China, you’d need to implement national socialism within the capitalist system.
AI elites successfully persuaded Trump and his aides to ignore two realities: the market has already proven that AI businesses aren’t profitable, and voters across party lines oppose building large numbers of new data centers and demand compensation for data theft. Since China can provide affordable AI products, the U.S. response is to put in even more money and build that “almost, almost, almost, almost about to be born” silicon-based god. (I’ll keep adding “almost,” because we’re only short on the belief that AGI is near, on data theft, and on taxpayer-funded support.)
Ultimately, everything in all domains—war, the economy, robotics—comes down to general AI. Therefore, other countries in the world must use AGI according to the will of the United States. Under this narrative, the U.S. has the most inclusive and fair culture in the world, and it must never allow this silicon-based god to be controlled by non–Judeo-Christian civilizations—such as China. (Rolls eyes, then rolls huge eyes.) I have my preferred place to live, and others do too. Even if I believe the moral culture I agree with is superior to other countries, I still wouldn’t be willing to devote all my financial resources, or even my life, to impose those values on the entire world. You can believe that American or Western culture is better, but don’t hand tens of trillions of dollars in taxpayer money to Elon, Sam, and Dario. (The heads of three leading U.S. frontier AI companies: xAI, OpenAI, Anthropic)
It’s said that the superiority of the U.S. system is that most of the time, hundreds of millions of informed citizens use free markets to determine prices for goods and services, with the government not interfering—letting market signals decide what to produce and in what quantities. But now, because of national security, based on an assumption—that by投入巨额资金 (injecting huge amounts of money) to feed data and predict the next Token, you can create AGI—the signals the market sends are judged to be wrong. Therefore, the U.S. government has to increase investment, build the next-generation frontier model, and use cultural advantages to outcompete China. That’s arrogance. Remember Icarus flying too close to the sun—what happened to him?
Alright, stop talking in big theories—talk about the bailout plan.
“Safety first” means a decline in the three U.S. AI labs’ demand for computing power. At that point, the government can step in and sign take-or-pay agreements to guarantee the AI labs stable profits—like the U.S. does with certain defense and mining companies’ contracts. The government would use this computing power to advance general AI R&D. Finally, the government can also lease out its self-developed models back to the AI labs: the labs would sell inference services to customers in the U.S. and allied countries at very high prices.
This scheme would allow the government to control frontier models for the Western world to use according to its own wishes. But the private AI lab model has a problem: these labs belong to globalized companies. Sometimes, to profit, they sell model access to anyone around the world, which can conflict with governments’ national security objectives. If China’s labs partially rely on distilling U.S. frontier models to make technical progress, and R&D is directly controlled by the government, China could fall behind by months or even years in the general AI race. Even if that scenario is true, it would still ultimately fail—just like when efforts were made to block advanced chip manufacturing equipment but still couldn’t stop China from producing cutting-edge chips. Information naturally wants to flow freely. In the internet era, information lockdown is simply impossible. Even before the internet was born, after the U.S. successfully developed the atomic bomb, it still couldn’t stop the Soviet Union from obtaining related intelligence. People who think AGI is an exception don’t understand how, under competition for national-level interests, human intelligence and adaptability matter.
Funding this silicon-based god requires issuing more debt. This plan is easy to sell because decision-makers in the monetary sphere—Treasury Secretary Bessent and Federal Reserve Chair Waller—both believe AI can boost productivity. They’re confident that if the U.S. fully embraces AI, it can rely on economic growth to escape the burden of massive debt—an assumption that isn’t entirely wrong. In June 2026, the U.S. nominal year-over-year GDP growth rate was 6.6%, while the effective federal funds rate was about 3.6%. Bessent keeps issuing short-term Treasury bills; by issuing this debt, the government can earn around 3% in return, while depositors bear the loss. If the budget deficit stays within 3% (a big premise), then the debt-to-GDP ratio will fall. Economic growth is mainly driven by AI data center construction, and what supports it is the computing power demand from AI labs. So from a financing perspective, if issuing short-term Treasuries can still earn a 3% return, it’s financially feasible for the government to act as the final buyer of computing capacity.
There’s no free lunch. The U.S. government runs deficits year after year, so it can only rely on borrowing to fund spending. If Waller cooperates, this is easy to do. But so far, the Federal Reserve he leads and the Treasury led by Bessent are not in step.
Since July 2023, the U.S. has raised interest rates for the first time in monetary policy: last week’s meeting, the Federal Reserve voted unanimously to raise the policy rate by 0.25%. The total money supply created by the Fed is no longer growing; as of August 14, the RMP short-term Treasury purchase program has been stopped.
If the government implements this plan but the Federal Reserve doesn’t lower the cost of capital and doesn’t expand its balance sheet, large-scale issuance of debt will push interest rates higher. Higher interest on mortgages, credit cards, and auto loans would only fuel voter anger toward AI-related policies. If Trump and Bessent can’t secure support from at least 7 FOMC members, the viability of this plan will be greatly reduced.
What’s described above is Fed policy from a traditional perspective, which makes it easy to take a bearish view on the market. But don’t forget there’s another powerful money-printing machine: commercial banks. Both Waller and Bessent argue that the banking industry should take over the responsibility for currency issuance. After the RMP purchases stopped on August 14, banks expanded their total assets and created more than a hundred billion dollars in new money—enabled by the relaxation of liquidity regulatory constraints behind the scenes.
Besides balance-sheet expansion, after this 0.25% rate hike, banks can still earn an additional $7.5 billion per year in interest on excess reserves deposited at the Fed. This money will be used for new lending and speculation in financial markets. Therefore, you can’t just look at the Fed’s rate hike and the stop to balance-sheet expansion—you also need to consider the impact on the commercial banking system. Combined, the overall effect is still stimulative. In other words, if the government wants to, the liquidity environment is enough to support additional borrowing and invest in building AI computing power capacity.
But if the government doesn’t take over the final procurement of computing power, these debts will face impairment. People who are leveraged on holding this kind of debt will fall into crisis. Now let’s dig deeper into this captive insurance scam.
Captive Insurance
I had never studied the insurance industry before. Large private equity firms use the assets of captive insurance companies to raise investment capital, and at first, it didn’t seem like a scam to me. But after digging into the operating mechanism, I found the ultimate victims.
In every round of credit bubbles, there’s a group of final buyers. Usually, the money is managed by trustees with impressive credentials who handle the funds of ordinary retail investors. The trustees invest with other people’s money and earn a double return: they collect management fees and then sell assets they hold to retail investors, inflating the prices of their own positions. And this time, the victims are policyholders who bought U.S. life insurance and annuity products. To understand this scheme, we first need to understand the lifeline tactics devised after the private equity golden age ended.
After the 2008 global financial crisis, the private sector deleveraged, and the Fed cut rates to near zero. The classic playbook of private equity: find mature companies with stable cash flows and almost no debt, lever them up, then extract cash by taking dividends, and finally leave the wrecked company behind in the private equity market. When enthusiasm in public markets rises again, take the rotten company public once more, completing the cycle. This logic works perfectly in a low-rate era. Ordinary people carry negative-equity mortgages, struggle to make monthly payments, and can’t afford to increase consumption. Private equity tycoons don’t want to expand production or provide better products—they just want to maintain existing cash flows, cut costs, and distribute cash dividends to their own investors.
U.S. total credit as a percentage of GDP. Orange is private market credit, blue is government-related credit. After the 2008 financial crisis, private credit kept falling while government credit kept rising, reflecting a shift in debt structure from the private sector to the government sector.
Private equity and venture capital (PE&VC) asset management scale, AUM—even through economic recessions—has continued expanding since 2000 and surpassed $1.5 trillion by 2025. The grey shadow is the recession periods marked by the NBER.
In the post-pandemic era, the cost of capital rose, and the law of diminishing marginal returns hit private equity returns hard. Because financing became cheap, to win a deal you have to offer higher valuations for cash-flow assets. Private equity returns then decline. To complete the next fundraising round, private equity tycoons began searching for a long-term pool of capital that doesn’t care about short-term redemptions—insurance companies entered the scene. Insurance companies sell life insurance and annuity policies. Policyholders pay premiums, and the insurers invest that money to earn returns, then pay out the policies decades later. This is exactly the kind of perpetual capital pool private equity dreams of: it can be poured into private equity funds packed with highly valued private companies and high-yield private credit.
So private equity tycoons acquire insurance companies, appoint themselves as investment managers, and package and sell low-quality assets to unsuspecting policyholders. This is captive insurance.
The most outrageous part of this deal is how the captive insurance company satisfies legal capital buffer requirements. Asset prices fluctuate, and regulators require insurers to hold capital buffers to ensure policy payouts. This gives rise to reinsurance companies that assume the payout risk of the original insurers. Normally, the original insurance and reinsurance are two independent institutions; reinsurers price the risk at fair market rates. But these rules don’t work for a private equity insurance scam. The core of this scam is to find a bag-holder, while private equity itself doesn’t need to put up large amounts of its own money. So the private equity-backed insurance company sets up a related captive reinsurance entity. The parent company only needs to inject a tiny amount of its own capital to obtain reinsurance coverage.
These are regulated entities that must periodically disclose information. If policyholders knew an insurance company was running such a system, would they still buy policies? To conceal the scam, the U.S. capital system has sided with private equity. Some state regulations—such as those in Vermont—conflict with national prudent regulatory standards. The original insurer and the related reinsurance entity can privately set up reinsurance risk assets, with the capital buffer size arbitrarily determined. Once state regulators approve it, the relevant documentation is sealed and kept confidential.
There are more details. Before continuing to dissect this bold scam, I’ll draw an analogy using an example from the crypto world. Remember Terra Luna? Luna collapsed because holders of USDT sold their stablecoins, breaking the dollar peg. If Do Kwon had had the resources of those New York private equity giants who were close to seventy years old, what would the outcome have been?
When the price of USDT falls, Luna acquires an insurance company called Alameda Insurance. Luna uses premium funds to buy USDT, trying to stabilize the peg. Alameda sells life insurance to people in California and holds billions in cash. Alameda can’t directly buy synthetic stablecoins, but it can purchase investment-grade corporate bonds. Luna gets a Moody’s analyst to rate its own company debt as investment-grade. To attract buyers like Alameda, Luna offers interest 5% higher than the yield on 10-year U.S. Treasuries. What a dazzling return! Alameda registers a reinsurance company called Three Daggers in Vermont. For every $100 of reinsurance assets, Alameda only pledges $1 of its own equity. Then Alameda tells regulators: the $10 billion of investment-grade debt Luna acquired using policyholder premiums is safe, and if anything goes wrong, Three Daggers will pay out. Everything looks fine. But as USDT keeps falling, Luna’s coin price collapses. A few weeks later, Luna can’t pay bond interest. Even if the Moody’s analyst hands over a Rolex watch to conceal the rating, you still can’t ignore a default event—you have to downgrade the debt to junk status.
Once credit ratings are downgraded, the whole structure collapses. A regulated insurance company, after a debt downgrade, must replenish its capital. But the reinsurance assets were fake from the start. Three Daggers never transferred tens or hundreds of millions of dollars in cash to the parent company to meet the capital requirement. When the scam is exposed, Alameda is insolvent on paper, and regulators can only clean up the mess.
Policyholders will suffer massive losses. In most U.S. states, the insurance coverage cap is only $250,000 to $300,000. If your policy should pay millions of dollars, the difference can’t be made up. Worse still, insurance protection funds are contributed after the fact by surviving insurers, completely different from the bank-insurance system. The FDIC (the U.S. Federal Deposit Insurance Corporation) requires banks to pay premiums in advance. This after-the-fact funding mechanism effectively encourages institutions to take extreme risks, because they don’t have to pay for the crisis upfront.
Back to traditional finance. Replace the debt of so-called fake projects with private credit to software companies impacted by AI, and with AI data center debt—whose value depends entirely on the three AI labs’ continuous procurement of computing power.
In the table, the column labeled “related reinsurance” is this kind of fictitious capital buffer. Private equity giants like Apollo, KKR, and Brookfield use it to package all of their AI-related investments. The true quality of the capital on self-procured insurance institutions’ balance sheets is impossible to verify, because they intentionally register in states and jurisdictions that can conceal real financial data. But based on publicly reported news, for nearly every major AI data center debt issuance, there is such a large private equity firm behind it. They are also the biggest private credit funds, investing heavily in SaaS companies. These private credit funds have already limited investor redemptions because these illiquid loans cannot be quickly sold off at a discount to raise cash.
Nick Nameth believes the scale of these fictitious captive reinsurance assets is as high as $1.54 trillion. We can’t determine the true amount, but he gave an example from Brookfield’s reinsurance assets: the book-recorded value is $148 million, but the entity providing reinsurance reported to regulators that its actual payout responsibility was 0.
Private credit and the AI debt market start showing cracks under the crushing weight of massive existing debt. Before the market fully realizes that private equity captive insurance companies are insolvent, we already see cracks forming. The fuse is when the credit rating for securitized investment-grade AI data center debt gets downgraded. Insurance companies originally bought the highest-rated tranches and enjoyed higher yields than U.S. Treasuries of similar maturities. Once leading labs can’t procure computing power at the expected scale, the resulting cash flows are insufficient to pay the data center debt. Credit rating agencies will eventually downgrade the debt ratings, triggering the parent company to top up capital—yet the related reinsurance can’t come up with cash. For investors like us who profit from money printing and monetary expansion, those multitrillion-dollar holes in insurers’ balance sheets will inevitably lead to bailouts. Baby boomer policyholders hold votes, and they will vote to push bailout measures. In 2008, when AIG became the last resort buyer, it absorbed a large amount of bad second-layer CDO debt and the government stepped in to rescue it. Don’t forget: after the bailout funds filled the AIG hole, the money went straight to Goldman Sachs, which in 2009 paid record bonuses. Ordinary people got foreclosure notices, while the elite held fat checks.
This scene will play out again. But Bessent and Waller aren’t stupid; they won’t repeat the same kind of pure free-market rhetoric from back then. Paulson and Bernanke (Hank Paulson and Ben Bernanke, the former Treasury Secretary and former Fed Chair) insisted on a bottom line: if investments fail, they should go bankrupt. They allowed Lehman to fail—that decision was wrong, and it made the public see how financial institutions preyed on ordinary people. This time, Bessent and Waller will never allow large insurance companies to go bankrupt and stage a similar financial disaster like (the Big Short). They will keep printing money to avoid this liquidation. After 2008, populist political forces rose. People won’t be as compliant as before. Back then, Obama—an officially progressive Democratic Party figure—won partly by riding the rhetoric about the financial crisis and punishing bankers, but once in office he still approved bailout plans and didn’t massively stop foreclosures. By 2028, AOC won’t be that easy to talk to. So Waller and Bessent must prevent a publicly visible credit catastrophe.
If Trump chooses not to be the final buyer of computing power, rating agencies will downgrade AI data center debt. The government printing of money will then be rolled out in batches and slowly, preventing the market from fully realizing that the insurance industry is insolvent.
History repeats itself
“Safety first” won’t immediately bring large-scale money printing. This is up to Trump to choose, but as a Bitcoin and crypto investor, no matter which path he takes, there will eventually be more money injection. This article will make you sure that after a small rise toward the end of August, the crypto market’s choppy period will quickly come to an end. The total supply of dollars will keep expanding, and the prices of Bitcoin and some altcoins will rise.
As the founder of the AI/crypto project Flop Network, this macro backdrop is very favorable for me. The U.S. government won’t allow free markets to halt data center construction; the original cost of computing power will fall, spot supply will become abundant, and that will drive the adoption of AI intelligent agents. In addition, a large amount of newly created dollars will push funds into crypto assets—during monetary expansion cycles, crypto assets perform best.