
Arthur Hayes: AI's Debt, Insurance's Time Bomb, Bitcoin's Fuel
Author: Arthur Hayes
Compiled by: Saoirse, Foresight News
Do you hear that? The heavyweights in the AI circle have suddenly developed a conscience and started worrying about the survival of humanity — because the product they’ve been building, an almost silicon-based god, is on the verge of being born. Actually, it has been close to becoming a silicon-based god for a while now, but with this final push seemingly within reach, they’ve suddenly started reflecting on the development path of Artificial General Intelligence (AGI).
(Note: "Silicon-Based God" refers to the super-AGI on the verge of birth, as spoken of in Silicon Valley. The author uses this term sarcastically, questioning whether this grand narrative is merely a pitch for fundraising and lobbying.)
That’s the narrative they’ve projected to the outside world. But the naturally skeptical me checked the calendar: it’s nearly the end of Q3, and Anthropic still hasn’t gone public. I couldn’t help but wonder what their financials actually look like. Is the high-growth annualized revenue curve, repeatedly mentioned in press releases, slowing down? They claim that after deducting all operating costs, the company is profitable. I’m eager to dig into their soon-to-be-filed S-1 prospectus to figure out the exact cost per token served to users. Moreover, how many customers are actually profit-generating, and is this profitable client base expanding or shrinking? Unfortunately, I won’t get answers to any of these questions. The reason? Safety first.
As my tone makes clear, I believe that Anthropic, OpenAI, and SpaceX are using “safety first” as an excuse to slow down AGI research not out of concern for ordinary human welfare, but driven by a harsh economic reality: the market isn’t willing to buy their AI products in sufficient quantities at current prices. To be specific, while demand for AI is strong, everyone wants China pricing — which is only one percent of U.S. prices.
When this “China pricing” hits the arrogant U.S. AI practitioners, their first reaction is to shout: “But those Chinese products are low quality.” And once the quality of Chinese models rapidly catches up, they wail: “China distills our models to make their own products.” The market doesn’t care why Chinese models are cheap; it only wants the cheapest intelligent service. So these AI practitioners switch their tune to lament: “We care about human safety, so we’ve paused R&D.” Sounds noble… But this exaggerated “simulation injury” performance, reminiscent of the World Cup, eventually shifts to a demand: “But we still need to beat China, so the government must step in, impose regulatory rules, and continue funding this AGI race.”
The fact that these three leading U.S. AI labs are using “safety first” as an excuse to slow AGI R&D is crucial to financial markets and global fiat liquidity because their demand for compute supports over $1 trillion in investment-grade debt, as well as hundreds of billions in sub-investment-grade loans.
Displaying the committed purchase amounts from OpenAI and Anthropic to the top four U.S. cloud providers as a percentage of each provider’s backlog reveals the massive long-term orders the two leading AI model companies bring to cloud infrastructure.
Collectively, these AI labs are generating zero profit. As a result, they rely on profitable tech giants like NVIDIA, Broadcom, Google, and Microsoft as backstops to provide off-balance-sheet guarantees for debts related to data center leases and chip purchases. Subsequent procurement of chips and hardware depends on the AI labs continuously training the most cutting-edge, “almost, almost, almost ready to become a silicon-based god” foundation models, while simultaneously processing inference requests for clients. But if “safety first” becomes the new core mandate, while the cost of training new models won’t hit zero, it will certainly retreat from its current peak; companies will shift focus to improving the efficiency of converting electricity into intelligence, meaning clients will spend less on compute. Essentially, “safety first” is destroying compute demand.
If AI capital expenditures were funded solely by operating cash flow, there would be little to worry about. The problem is that regardless of whether the AI labs keep buying compute, this multi-trillion-dollar debt remains. Defaults won’t happen overnight, but once the AI labs stop consuming compute at previously anticipated volumes, the price of this debt will drop. The core question is: Who bought this debt, and did they leverage up? The answer is obviously yes: speculators leveraged up to buy this subpar debt. So who ends up holding the bag?
Millions of U.S. insurance policyholders have actually been indirectly betting on the AI story. And the “safety first” pivot will hit them with zero buffer. I didn’t fully grasp this scam until Nick Nameth broke it down thoroughly on Substack. Next, I’ll explain it in simple terms for my crypto community readers. The conclusion is stark: If AI-related debt were revalued at fair market value, a significant portion of the U.S. insurance industry would already be insolvent. This brings us to the central investment thesis in our fractional-reserve, bank-dominated global economy: The U.S. government faces a binary choice — either act as the ultimate buyer of compute under the guise of national security, or print money to bail out deeply unprofitable insurance companies.
No matter which path is chosen, Bitcoin holders and crypto investors win. If the government ignores market signals and insists on funding the development of this commercially unviable “silicon-based god,” it will need to print money to finance this non-productive expenditure, inevitably breeding more financial speculation and driving up Bitcoin’s price. If the government chooses to bail out the insurance industry, it will print money to absorb toxic AI debt, expand the money supply, and subsequently push Bitcoin higher.
The remainder of this article will deconstruct this mechanism.
In China’s Name
By invoking national security, the U.S. government can justify virtually anything. Think about the devastation caused after 9/11 when the U.S. amplified threats and launched a global war on terror. This time, the constructed bogeyman is the Chinese researcher buried in mathematics, stealing top-tier U.S. technology, and selling it back to America at one percent of the price. To defeat China, the U.S. must implement state socialism within a capitalist framework.
AI moguls successfully convinced Trump and his aides to ignore two realities: the market has proven AI isn’t profitable, and bipartisan voters oppose building massive new data centers while demanding compensation for data theft. Since China can offer affordable AI products, the American countermeasure is to pour even more capital into forging that “almost, almost, almost, almost-ready-to-be-born silicon-based god.” (I’ll keep adding “almost” because we’re only missing faith, data theft, and taxpayer money to get to AGI.)
War, economics, robotics, and every other domain are ultimately subordinate to AGI. Therefore, the rest of the world must use AGI according to U.S. wishes. Under this narrative, America possesses the most inclusive and equitable culture globally, so this silicon-based god absolutely cannot fall into the hands of non-Judeo-Christian civilizations like China. (Roll eyes, roll bigger eyes.) I have my preferred place to live, and so do others. Even if I believe the moral culture I endorse is superior to another country’s, I refuse to bankrupt myself or risk my life to force these values upon the whole world. You can believe U.S. or Western culture is superior, but don’t hand hundreds of billions in taxpayer money to Elon, Sam, and Dario. (Founders of three leading U.S. frontier AI labs: xAI, OpenAI, Anthropic)
The supposed superiority of the American system is that, most of the time, hundreds of millions of informed citizens let free markets determine prices for goods and services without government interference, allowing market signals to dictate what and how much is produced. But now, under the banner of national security, based on the assumption that massive capital deployment to predict the next token will yield AGI, those very market signals are deemed wrong. Thus, the U.S. government must double down, build next-generation frontier models, and leverage cultural superiority to suppress China. It’s pure hubris. Remember Icarus flying too close to the sun?
Enough of the lofty rhetoric. Let’s talk about the bailout plan.
“Safety first” means reduced compute demand from the three major U.S. AI labs. At that point, the government could step in with take-or-pay contracts to guarantee stable profits for the labs, similar to U.S. defense and mining sector agreements. The government would use this compute to advance AGI R&D. Finally, it could lease its proprietary models back to the AI labs, which would then sell inference services to domestic and allied clients at heavily premium pricing.
This setup keeps frontier models in government hands for Western use on its terms. The private lab model has a flaw: these labs are global enterprises that sometimes sell model access to anyone for profit, conflicting with national security objectives. If Chinese labs gain an edge by partially distilling U.S. frontier models, direct government control over R&D would delay China’s AGI progress by months or even years. Even if this concept worked, it would ultimately fail, much like previous attempts to restrict advanced semiconductor manufacturing equipment that failed to stop China from developing cutting-edge chips. Information inherently seeks freedom. In the internet age, total information blockade is impossible. Long before the internet, the U.S. successfully developed atomic bombs yet still couldn’t prevent the Soviet Union from acquiring related intel. Anyone who thinks AGI is an exception fails to understand human ingenuity and adaptability amid geopolitical interests.
Funding this silicon-based god requires issuing more debt. This is easily sold because monetary policymakers — Treasury Secretary Bessent and Fed Chair Walsh — both believe AI boosts productivity. They are convinced that embracing AI head-on will allow the U.S. to shed its massive debt burden through economic growth, an assessment that isn’t entirely unfounded. U.S. nominal YoY GDP growth hit 6.6% in June 2026, with an effective federal funds rate around 3.6%. With Bessent continuously issuing short-term Treasuries, the government earns a 3% spread on this debt issuance while depositors bear the loss. If the fiscal deficit stays under 3% (a big caveat), the debt-to-GDP ratio will decline. Economic growth is primarily driven by AI data center construction, underpinned by AI lab compute demand. From a financing perspective, if the government can earn a 3% return on short-term bonds while acting as the ultimate compute buyer, the plan is financially viable.
There’s no such thing as a free lunch. The U.S. government runs perennial deficits and relies on borrowing to spend. This works smoothly if Walsh cooperates. So far, however, the Federal Reserve under his leadership and the Treasury led by Bessent are not marching to the same drum.
The first U.S. monetary tightening since July 2023 occurred last week, with the FOMC unanimously voting to raise the policy rate by 0.25%. The total amount of money created by the Fed is no longer expanding. Starting August 14, the RMP short-term Treasury purchase program has ceased.
If the government pushes this plan but the Fed refuses to lower funding costs or expand its balance sheet, large-scale debt issuance will spike interest rates. Rising mortgage, credit card, and auto loan rates will only inflame voter anger toward AI-related policies. If Trump and Bessent fail to secure support from at least seven FOMC members, the plan’s viability plummets.
The traditional view of Fed policy shown above easily lends itself to a bearish market outlook. But remember, there’s another powerful money-printing machine: commercial banks. Both Walsh and Bessent argue the banking sector should shoulder the burden of monetary creation. Since the RMP purchase halt on August 14, banks have created hundreds of billions in new money by expanding total assets, facilitated by relaxed liquidity regulatory constraints.
Beyond balance sheet expansion, the 0.25% rate hike also means banks will earn an extra $7.5 billion annually in interest on excess reserves held at the Fed. These funds will flow into new lending and financial market speculation. Therefore, one cannot just look at Fed hikes and halted QE; the impact of the commercial banking system must be factored in. Combined, the net effect remains stimulative. Meaning, if the government wants, the liquidity environment is more than capable of supporting increased borrowing for AI compute deployment.
But if the government doesn’t step in as the ultimate compute buyer, the debt will face impairment losses. Those holding this debt on leverage will face a crisis. Next, we dive deep into the captive insurance scam.
Captive Insurance
I had never researched the insurance industry before. Using captive insurance company assets for private equity fundraising initially didn’t strike me as a scam. But after digging into the mechanics, I found the ultimate victim.
Every credit bubble has a group of final bagholders. Typically, retail money is managed by polished trustees. Trustees invest other people’s money for a dual payoff: charging management fees and dumping their own holdings onto retail investors to inflate the price of their positions. This time, the victims are policyholders purchasing U.S. life insurance and annuity products. To understand this scam, we must first look at the lifeline private equity titans devised after the golden age of PE came to an end.
Post-2008 Global Financial Crisis, the private sector deleveraged and the Fed cut rates to near zero. The classic PE playbook: find mature companies with stable cash flows and minimal debt, lever them up, strip dividends for cash-out, and leave the hollowed-out corporation behind in the private sphere. Wait for public market euphoria, re-IPO the junk, and repeat the cycle. In a low-rate era, this logic held perfectly. Average households carried negative-equity mortgages, struggling with payments and incapable of increasing consumption. PE masters didn’t want to scale production or improve products; they just wanted to maintain legacy cash flows, slash costs, and extract dividends for their own limited partners.
U.S. Total Credit as % of GDP. Orange represents private market credit, blue represents government-related credit. Post-2008, private credit steadily declined while government credit rose continuously, reflecting a structural shift in debt from the private to the public sector.
Assets Under Management (AUM) for Private Equity and Venture Capital (PE & VC). Despite surviving multiple recessions, PE & VC has expanded continuously since 2000, surpassing $15 trillion by 2025. Gray shading marks NBER recession periods.
In the post-pandemic era, rising capital costs dealt a severe blow to PE fund returns via diminishing marginal utility. Because financing was cheap, securing deals required paying inflated valuations for cash-flow assets, compressing PE fund yields. To close the next fund, PE bosses turned to long-term capital pools indifferent to short-term redemptions — enter insurance companies. Insurers sell life and annuity policies: policyholders pay premiums, insurers invest the capital for returns, and payouts occur decades later. This is the perpetual capital pool PE dreamers crave, perfect for pouring into funds stuffed with overvalued private companies and high-yield private credit.
Thus, PE titans acquire insurance companies, serve as investment managers, and package toxic assets to sell to unwitting policyholders. This is captive insurance.
The most absurd part of this deal is how captive insurers satisfy legally mandated capital buffers. Asset prices fluctuate, and regulators require insurers to hold capital buffers to guarantee policy payouts. This spawned the reinsurance industry to absorb the primary insurer’s payout risks. Normally, primary and reinsurance firms are completely separate, with reinsurers pricing risk at fair market value. But this doesn’t work for the PE insurance scam, whose core objective is finding a counterparty without the PE firm having to deploy substantial equity. Consequently, PE-controlled insurers set up affiliated captive reinsurance entities. The parent company only needs to inject minimal capital to secure reinsurance coverage.
These are regulated entities requiring periodic public disclosure. Would policyholders buy in if they knew the insurer operated such a system? To conceal the fraud, the U.S. capital system sided with PE. Regulations in states like Vermont blatantly contradict nationwide prudential standards. Primary insurers and affiliated reinsurers can privately establish reinsurance risk assets, freely sizing their capital buffers. Once approved by state regulators, all relevant documentation is sealed and kept confidential.
There’s more detail. Before dissecting this audacious scheme further, I’ll draw an analogy using a crypto example. Remember Terra/LUNA? LUNA collapsed because USDT holders dumped the stablecoin, breaking the USD peg. What if Do Kwon had the resources of those septuagenarian New York PE titans? How would the ending play out?
As USDT’s price fell, Luna acquired an insurer named Alameda Insurance. Luna used premium funds to buy USDT, attempting to stabilize the peg. Alameda sold life insurance to Californians, managing billions in capital. Alameda couldn’t directly buy shitcoin stablecoins, but it could purchase investment-grade corporate bonds. Luna bribed Moody’s analysts to rate its own debt as investment grade. To attract buyers like Alameda, Luna offered coupon rates 5% above 10-year Treasuries. What attractive yields! Alameda registered a reinsurer in Vermont called Three Daggers. For every $100 in reinsurance assets, Alameda pledged only $1 in equity. Then Alameda told regulators: the $10 billion Luna investment-grade debt purchased with policyholder premiums is safe, and Three Daggers will cover payouts if things go south. Everything looks clean. But USDT continued to bleed, and LUNA crashed. Weeks later, Luna defaulted on bond coupons. Even if Moody’s analysts hid their Rolex watches to suppress ratings, they couldn’t ignore the default and were forced to downgrade the debt to junk.
Once ratings are downgraded, the whole structure collapses. Regulated insurers must replenish capital after debt is downgraded. But the catch is that this reinsurance asset was fake from day one. Three Daggers never transferred hundreds of millions in cash to the parent company to meet capital requirements. The scam unravels, Alameda’s books show insolvency, and regulators are left cleaning up the mess.
Policyholders suffer massive losses. Most U.S. states cap insurance protection at only $250,000 to $300,000. If your policy should have paid out millions, the shortfall goes uncompensated. Worse still, insurance guarantee funds are paid ex-post by surviving insurers, fundamentally different from the banking system. The FDIC mandates banks pay premiums upfront. This ex-post funding mechanism effectively incentivizes extreme institutional risk-taking, as entities never have to front-load crisis costs.
Back to TradFi. Replace shitcoin debt with private credit to software companies impacted by AI, alongside AI data center debt, where asset values rely entirely on the three major AI labs continuously purchasing compute.
The “Affiliated Reinsurance” column in the table represents this false capital buffer. PE giants like Apollo, KKR, and Brookfield rely on it to mask all their AI-related investments. The true quality of captive insurer assets is unknowable because they are deliberately incorporated in jurisdictions/state regulations that hide real financial data. However, from public news, almost every large AI data center debt issuance has these mega PE firms behind it. Simultaneously, they operate the largest private credit funds, investing heavily in numerous SaaS companies. These private credit funds have already restricted investor redemptions because these illiquid loans cannot be quickly liquidated at a discount.
Nick Nameth estimates the size of these fake captive reinsurance assets reaches $1.54 trillion. We can’t verify the exact number, but he points to Brookfield’s reinsurance assets: booked at $1.48 billion, yet the reinsuring entity reported to regulators that its actual payout liability is zero.
The private credit and AI debt markets are starting to crack under the weight of massive outstanding obligations. Before the market fully grasps that PE-backed captive insurers are insolvent, we’ve already seen hairline fractures emerge. The trigger will be a ratings downgrade on securitized AI data center investment-grade debt. Insurers originally bought the highest-rated tranches to enjoy yields exceeding comparable Treasuries. Once top-tier labs fail to purchase compute at expected volumes, generating insufficient cash flow to service data center debt, rating agencies will inevitably downgrade the debt, triggering parent capital replenishment demands that affiliated reinsurers simply cannot fund. For investors profiting from monetary expansion, this multi-trillion-dollar hole on insurers’ balance sheets inevitably warrants a bailout. Baby boomer policyholders hold the ballots; they will vote to force relief measures. In 2008, AIG acted as the ultimate bagholder, absorbing massive subprime CDO trash, prompting government rescue. Don’t forget: after TARP funds plugged AIG’s hole, capital flowed straight to Goldman Sachs, which distributed record bonuses in 2009. Main street received foreclosure notices while elites walked away with fat checks.
History repeats. But Bessent and Walsh aren’t stupid; they won’t repeat the pure free-market rhetoric of old. Paulson and Bernanke drew a hard line back then: failed investments mean bankruptcy. They let Lehman fail, a mistake that exposed to the public how financial institutions loot the masses. This time, Bessent and Walsh will absolutely not permit large insurers to collapse into a living The Big Short disaster film. They will keep printing money to avoid this reckoning. Because post-2008, populist political forces have risen, and the public won’t submit passively like before. Obama, ostensibly a progressive Democrat, rode anti-crisis and anti-banker sentiment to victory, yet once in office still approved bailouts without halting mass foreclosures. By 2028, AOC won’t tolerate such compromises. Therefore, Walsh and Bessent must preempt any public credit catastrophe.
If Trump opts not to become the ultimate compute buyer, and rating agencies downgrade AI data center debt, money printing will roll out in staggered, gradual batches to prevent the market from fully realizing the insurance sector’s insolvency.
History Repeats
“Safety first” won’t instantly trigger massive money printing. That’s a call for Trump to make, but as Bitcoin and crypto investors, no matter his choice, more monetary injection is inevitable. This article should convince you that the recent crypto market volatility following the small late-August rally will resolve quickly. The total dollar supply will continue expanding, driving prices higher for Bitcoin and select altcoins.
As the founder of the AI/crypto project Flop Network, this macro environment is highly favorable for me. The U.S. government will never let free markets halt data center construction; compute wholesale costs will fall, spot compute supply will overflow, accelerating AI agent adoption. Furthermore, massive new USD liquidity will drive capital into crypto assets — which historically outperform during monetary expansion cycles.












