Arthur Hayes Says an AI Bust Would Force Washington to Print — and Bitcoin Would Soak Up the Liquidity

Arthur Hayes is back with a new macro thesis, and this time the target is not a central bank balance sheet but the artificial intelligence buildout itself. In a Sept. 22 essay titled Safety First, the former BitMEX CEO argued that the sudden safety-consciousness of major AI labs may conceal a simpler economic problem: demand for expensive AI compute is weakening, and the debt stack built on that demand is starting to look fragile. His verdict is blunt. “Safety First is by definition compute demand destruction,” Hayes wrote, before compressing his policy forecast into four words: “Trump has a choice, print or print.”

The argument: AI debt is the pressure point

Hayes’s case rests on the financing behind the AI infrastructure boom. Data centers, chip purchases and private credit arrangements all assume robust future cash flows from training and inference demand. If AI labs are slowing frontier development because customers will not pay for the compute, those cash-flow assumptions weaken, and the credit built on top of them comes under pressure. Independent research confirms the scale of the financing, if not the crisis Hayes predicts. Apollo said in August that AI-related issuance accounted for nearly 40 percent of longer-duration investment-grade corporate bond supply, and its economists estimated the AI ecosystem could support more than 2 trillion USD in additional investment-grade debt — while public markets may absorb less than 1 trillion USD of it through 2030. Apollo separately estimated that roughly 5 trillion USD could be spent on AI infrastructure through 2030, and that businesses and consumers would need to spend around 2 trillion USD annually on AI services to justify that outlay. A Sept. 21 Apollo note added that consensus forecasts assume operating cash flow at five major hyperscalers — Alphabet, Amazon, Meta, Microsoft and Oracle — will rise from roughly 600 billion USD to 2 trillion USD by 2030. Weaker cash-flow growth, Apollo warned, could mean wider credit spreads and lower capital expenditure.

The insurance leg: 1.54 trillion USD in affiliated reinsurance

The second pillar of Hayes’s scenario involves insurers and private credit. Drawing partly on research by Nick Nemeth, Hayes points to affiliated reinsurance structures across the U.S. life and annuity industry, which Nemeth estimates at 1.54 trillion USD. His argument is that some of those reinsurance assets may offer less economic protection than statutory accounting suggests, leaving insurers exposed if AI-linked debt is downgraded and marked lower. Regulators are watching the neighborhood, if not yet confirming the fire. The National Association of Insurance Commissioners identifies private credit as an area requiring continued supervision, citing its weaker liquidity, less frequent valuation and reduced price transparency relative to public debt. The NAIC also counts 139 private-equity-owned U.S. insurers as of June 2025, with affiliated investment management and cross-border reinsurance under ongoing review. Current industry data, however, does not establish that insurers are insolvent because of AI. A recent Moody’s survey reported by the Wall Street Journal estimated direct U.S. insurer exposure to data centers at up to 20 billion USD — far below the systemic figures in Hayes’s thesis, which depend on indirect exposure through private credit, reinsurance and structured financing.

Print or print: the two rescue scenarios

If AI infrastructure economics deteriorate, Hayes sketches two possible government responses. In the first, Washington becomes a “compute buyer of last resort,” using public spending or offtake agreements to keep demand for AI capacity alive. In the second, the government backstops insurers whose private-credit losses threaten policyholder claims. No U.S. authority has announced either policy — but Hayes argues both would require expanded government borrowing, banking-system liquidity or direct monetary support, and that such a liquidity wave would raise demand for Bitcoin and other scarce assets. The immediate monetary backdrop runs against him. The Federal Reserve raised its target range by 25 basis points on Sept. 16 to 3.75–4.00 percent, its first hike since July 2023, with a unanimous vote and officials noting inflation remained elevated. The New York Fed has scheduled no reserve-management purchases for the Aug. 14–Sept. 14 and Sept. 15–Oct. 14 operating periods, although reinvestment purchases continue. Fed H.8 data shows commercial-bank credit growing from 19.74 trillion USD in July to 19.87 trillion USD by the week ending Sept. 9 — balance-sheet growth, but not the deliberate stimulus program Hayes describes.

The AI labs tell a different story

The companies themselves attribute the slowdown to safety, not weak demand. OpenAI said in August it temporarily slowed parts of frontier model development after cybersecurity concerns and stronger internal safeguards became necessary, and its GPT-6 Astra work proceeded after the model was classified at a critical cybersecurity capability level. Anthropic CEO Dario Amodei has called for the industry to pace development so safety controls catch up with capabilities — and Reuters reported on Sept. 19 that Anthropic is weighing another model release despite that request. Financing activity also remains substantial. Nvidia announced in August that Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR were working on independent AI-compute financing platforms intended to mobilize more than 500 billion USD in third-party capital over time. SoftBank began marketing more than 11 billion USD of high-yield bonds this week to help finance its OpenAI investment, according to the Financial Times. Hayes, for his part, has history on his side as a storyteller if not always as a forecaster: he previously linked a potential AI credit crisis to a monetary response favoring Bitcoin, and argued in August that Treasury buybacks and banking liquidity could support crypto prices even without conventional quantitative easing. Bitcoin traded near 85,700 USD early on Sept. 22 after climbing more than 6 percent in the previous session, and changing hands around 85,958 USD at press time — up roughly 5.9 percent on the day, with Ethereum at 2,751 USD and Solana at 117 USD as the market weighs whether the print-or-print scenario is prophecy or provocation.