Arthur Hayes has laid out a stark thesis: Anthropic, OpenAI, and SpaceX claim to be slowing down artificial general intelligence (AGI) development out of a commitment to safety, but the real motive is to hide their brutal economic reality of unprofitability. The narrative of a so-called silicon god built by these companies is now crumbling, as the market's genuine demand lies with Chinese pricing models offered at a fraction of the cost of American technology. This mismatch between supply and demand is shrinking compute requirements, which in turn is triggering a massive debt crisis lurking behind the industry.
A deeper look into the financial truth of AI giants reveals that their profitability claims are under direct assault from Chinese pricing pressure. As the third quarter draws to a close, Anthropic's prolonged absence from the public markets has fueled intense skepticism about the authenticity of its financial reports. Hayes is eager to examine the upcoming S-1 filing to clarify the true cost of delivering each token to users and whether the base of profitable customers is expanding or contracting. Yet every question about cost structures and margins gets buried beneath the safety-first excuse.
In reality, market demand for AI is robust, but consumers are gravitating toward Chinese models that cost just one percent of their American counterparts. When U.S. players accuse China of producing subpar goods or relying on distilled models, the market is not buying that argument — it simply wants the cheapest intelligence available. So AI giants have pivoted to claiming they are pausing development out of concern for humanity, which is actually a ploy to secure government regulation and subsidies that will support their inflated pricing structures.
Core crisis: the compute debt loop
The compute debt crisis is spreading quickly, creating a tangled web of trillions in liabilities and off-balance-sheet guarantees. Top AI labs like Anthropic and OpenAI generate enormous demand for computing power, underpinning over one trillion dollars in investment-grade debt alongside hundreds of billions in lower-credit-rated loans. Since these labs collectively produce zero profit, they must lean on profitable tech companies such as Nvidia (NASDAQ: NVDA), Broadcom (NASDAQ: AVGO), Google (NASDAQ: GOOGL), and Microsoft (NASDAQ: MSFT) to provide off-balance-sheet guarantees for the debt tied to data center leases and chip purchases.
Future acquisitions of chips and hardware depend entirely on AI labs continuing to train cutting-edge models and process inference requests for clients. If safety-first becomes the guiding principle, spending on new model training will drop from its highs, and the focus will shift to improving the efficiency of converting electricity into intelligence — directly reducing customer spending on compute. In essence, safety-first is a mechanism to destroy compute demand, thereby destabilizing the very foundation of the entire debt chain.
Who actually holds the leveraged AI debt?
The essence of safety-first is the destruction of demand, which in turn triggers the risk of debt defaults. If AI capital expenditures were funded by operating cash flow, the risk would be manageable, but the reality is that trillions of dollars in debt remain outstanding. Once AI labs stop consuming compute at the expected scale, the prices of such debt will plummet dramatically. The key issue is that the speculators who bought this debt are typically leveraged and hold vast amounts of low-quality paper. This leveraged debt structure makes the market extremely fragile, and any contraction on the demand side could set off a chain reaction.
The real core question is who ultimately bought this debt and whether they used leverage in the process. The answer is obviously yes — speculators piled into these inferior assets propped up by the AI narrative using high leverage, planting the seeds for a future crisis. The final bag holders are millions of American insurance policyholders who have indirectly bet on the AI story while facing the risk of insolvency.
The captive insurance scheme hiding the losses
Nick Nameth has thoroughly exposed this scam: if AI-related debt were marked to market fair value, a large portion of the U.S. insurance industry would already be insolvent. In a global economy dominated by fractional reserve banking, this raises the core investment question. Insurance companies sell life insurance and annuity policies, investing the premium proceeds into AI-linked debt to chase higher yields. But once that debt depreciates, the insurers' balance sheets take a massive hit, and policyholders face enormous losses. This risk-transfer mechanism makes ordinary people the ultimate victims of financial elites' speculative behavior.
The U.S. government now faces a binary choice: either act as the ultimate purchaser of compute under the guise of national security, or print money to bail out struggling insurers. Whichever path it takes, bitcoin holders and crypto investors come out on top. If the government ignores market signals and pours funds into developing a silicon god that lacks commercial viability, it will need to print currency to finance such non-productive spending, inevitably spawning more financial speculation and pushing bitcoin prices higher. If the government opts to rescue the insurance industry, it will print money to absorb bad AI debt, expanding the money supply and again driving bitcoin up. This dilemma makes monetary expansion an unavoidable outcome, providing a solid macro foundation for crypto assets.
The political narrative: countering China and elite interests
The political narrative revolves around countering China and delivering benefits to elites. Invoking China as a threat, the U.S. government can justify nearly any action, much like the global war on terror launched after 9/11. This time, the fabricated adversary is China, which offers affordable AI products. AI moguls have successfully convinced Trump and his advisors to ignore the market evidence that AI businesses are unprofitable and to disregard voters' opposition to new data centers. To defeat China, the U.S. must implement state socialism within the capitalist system, pouring more money into building the silicon god.
The narrative argues that America possesses the most inclusive and fair culture in the world and that AGI must never fall into the hands of a nation outside the Judeo-Christian tradition. Therefore, trillions in taxpayer dollars are handed to Elon, Sam, and Dario for xAI, OpenAI, and Anthropic research. This hubris recalls Icarus flying too close to the sun — it is destined to fail.
Macro data: rate hikes meet balance sheet expansion
Macro data and monetary policy reveal a peculiar situation: rate hikes coexist with bank balance sheet expansion, keeping the liquidity environment accommodative. As of June 2026, U.S. nominal year-on-year GDP growth stands at 6.6%, while the effective federal funds rate is around 3.6%. Treasury Secretary Bessent continues to issue more short-term Treasury bills; the government earns a 3% return on this issued debt, but savers suffer the losses. If the fiscal deficit stays within 3% of GDP, the debt-to-GDP ratio declines. However, for the first time since July 2023, U.S. monetary policy has tightened — the Federal Reserve unanimously voted last week to raise the policy rate by 0.25%.
The Fed's creation of new money is no longer growing, and as of August 14, the RMP short-term Treasury purchase program has been halted. If the government pushes forward with a compute procurement plan but the Fed refuses to lower the cost of funds or expand its balance sheet, massive debt issuance would push interest rates higher, provoking voter anger. Thus, Trump and Bessent will need the support of at least seven FOMC members to ensure the plan's feasibility. At the same time, commercial banks have created hundreds of billions in new money by expanding their total assets, backed by relaxed liquidity regulatory constraints. After the recent 0.25% rate hike, banks holding excess reserves at the Fed earn an additional $7.5 billion per year in interest, which can be used for new loans and financial market speculation. Combined, the net effect remains stimulative, sufficient to support new debt issuance for AI compute buildout.
Echoes of 2008: bailouts and the crypto upside
In 2008, AIG (NYSE: AIG) acted as the final absorber of massive amounts of toxic secondary CDO debt, and the government stepped in to rescue it. After the TARP bailout funds plugged AIG's hole, the money flowed directly to Goldman Sachs (NYSE: GS), which paid record bonuses in 2009. Paulson and Bernanke back then held a line by letting Lehman fail, but that decision was a mistake that revealed to the public how financial institutions prey on the masses. This time, Bessent and Waller will absolutely not allow a major insurer to collapse in a scene reminiscent of *The Big Short*. They will keep printing money to avoid such a reckoning, because since 2008, populist political forces have risen, and the public will not be as compliant as before. Obama approved the bailout without broadly halting home foreclosures. By 2028, AOC will not be so accommodating. Therefore, Waller and Bessent must prevent a public credit catastrophe at all costs.
If Trump declines to act as the ultimate purchaser of compute, and rating agencies downgrade AI data center debt, money printing will occur gradually and in measured increments to prevent the market from fully realizing that the insurance industry is insolvent. Hayes, founder of the AI/crypto project Flop Network, believes this macro environment is highly favorable. The U.S. government will not allow free markets to halt data center construction; the raw cost of compute will decline, and spot compute oversupply will accelerate the adoption of AI agents. Additionally, a flood of new dollars will drive capital into crypto assets, which historically perform best during monetary expansion cycles.
Comments