Arthur Hayes has cast the AI infrastructure boom as a credit cycle built around data centers, construction and rapidly aging chips, arguing that its eventual rupture could resemble the housing-finance stress of 2008 more closely than the dot-com collapse of 2000.
In Hayes’s account, markets have treated the surge in spending by major technology companies as a conventional technology-growth story. He argues that the more consequential exposure sits behind the software narrative: debt-financed facilities, power infrastructure, specialized chips and the lenders financing them. A slowdown in the growth of data-center construction or reduced forward capital-expenditure guidance from hyperscalers could therefore hit valuations and credit markets before leading AI companies show a dramatic collapse in earnings.
Hayes’s thesis rests on the idea that the buildout is vulnerable to a change in the rate of growth rather than an outright halt in spending. He expects AI capital-expenditure growth to decelerate in mid-to-late 2027 and sees 2028 as a more clearly defined slowdown period. Credit to the sector could continue expanding after capital-spending growth has already peaked, leaving lenders and borrowers exposed when expectations reset.
A physical buildout behind the AI trade
The infrastructure required to run advanced AI systems differs from a purely digital software expansion. Hyperscalers are building facilities, securing power, installing cooling systems and filling server halls with expensive semiconductor hardware. Those assets can be financed over years, while the technology inside them may be superseded much faster.
Hayes used a numerical example in which a future facility could generate more than 1,000 times the “intelligence” of an earlier data center while consuming less electricity. The point is not that AI demand must fall for older sites to lose value. Rising chip efficiency could make existing installations less competitive, particularly if newer models deliver more computing output for each unit of power.
That creates a mismatch between the lifespan of construction and debt obligations on one side, and the shorter replacement cycle for high-end chips on the other. A data-center owner facing costly hardware upgrades may need to refinance even while AI usage continues to rise. Hayes argues that this dynamic places particular pressure on heavily leveraged developers and operators rather than necessarily on the largest technology groups first.
He describes the process as a “second derivative” trade. Financial markets commonly react to whether growth is speeding up or slowing down, not only to whether a company or sector is still expanding in absolute terms. If planned spending rises more slowly than expected, forward valuation multiples can contract, reducing equity values and weakening the collateral behind debt.
Credit stress could emerge before a visible recession
Hayes compares the prospective sequence to the years before the 2008 financial crisis, when stresses appeared in credit markets before the full scale of the housing downturn was widely understood. He cited the approximately 50% decline in the S&P 500 from its peak before the deeper systemic break was fully recognized, as well as the August 2007 failures of three BNP Paribas-linked credit funds as an early sign of market strain.
His analogy does not depend on AI firms producing no revenue. Instead, it focuses on credit structures: borrowers that financed long-lived assets, lenders holding concentrated exposure and a potential feedback loop between falling equity prices, tighter lending terms and refinancing needs.
Hayes expects the earliest failures to emerge among borrowers with the highest leverage. Stress could then move toward lenders with substantial AI-linked debt holdings, particularly if asset values decline while loan maturities approach. The sequence would resemble a credit event in which individual defaults become more damaging as they reveal shared exposures across the financial system.
He also tied the availability of credit to the shape of the U.S. yield curve. Hayes pointed to the spread between the 10-year Treasury yield and the effective federal funds rate, alongside commercial and industrial loan balances at U.S. commercial banks, as indicators of how a steeper curve can support lending margins. He noted that long-term Treasury yields rose after the Federal Reserve held rates steady, a condition he views as supportive of continued bank credit creation.
Hayes expects policy support to follow
A central part of Hayes’s argument is that policymakers would respond to a major AI-credit disruption with liquidity measures rather than allow a broad unwind. He described a mix of negative real interest rates, central-bank purchases of short-term Treasury bills and government measures encouraging banks to direct lending toward sectors such as AI and defense.
He called the proposed arrangement a modern version of a “Treasury–Fed Accord,” supplemented by “window guidance”-style incentives for banks. The term refers broadly to measures that steer credit through policy preferences and financial-system incentives rather than relying solely on changes to benchmark interest rates.
Hayes also referenced the Federal Reserve Act’s emergency lending provisions, which allow the central bank, under specified conditions, to lend through broad-based facilities with Treasury approval. He suggested that a Treasury-created special purpose vehicle could become a route for market support. Using the Exchange Stabilization Fund’s roughly $28 billion balance as his example, he said a 10-to-1 leverage structure could create around $280 billion in purchasing capacity.
His more contentious scenario involves direct state support for AI-linked equities, which he called “equity QE.” Such an approach could leave a government-backed entity holding large positions in companies whose market liquidity deteriorates if the holder attempts to exit. Hayes argues that this would encourage rolling refinancing and prolonged balance-sheet support rather than quick asset sales.
Bitcoin and Ether in Hayes’s liquidity scenario
Hayes linked his bitcoin outlook to the timing of credit stress and policy intervention. He wrote that bitcoin’s previous-cycle low followed the revelations surrounding Sam Bankman-Fried and the misuse of FTX customer funds, while liquidity conditions began improving in October 2023 as balances declined in the Federal Reserve’s overnight reverse repurchase facility.
In his account, bitcoin peaked in October 2025 before falling 50% as AI-linked equities and credit absorbed newly created fiat liquidity. Hayes expects bitcoin could bottom during the early phase of AI credit misallocation, identifying a possible $60,000 to $70,000 range and allowing for a decline to $50,000.
He also repeated his long-term view that bitcoin could exceed $1 million if future liquidity programs surpass the multi-trillion-dollar policy response he associates with 2009 through 2013. That outcome depends on the policy response he anticipates, rather than on the data-center cycle alone.
For Ether, Hayes expects tokenized real-world-asset systems to use customizable Ethereum Layer-2 networks, including Arbitrum, while relying on Ethereum for securities settlement. He set a $5,000 Ether target by the end of 2026, describing it as roughly a 2.6-fold increase from the price level used in his analysis.
The framework leaves a narrow set of signals to watch: hyperscaler spending guidance, the pace of data-center construction, lending growth and the ability of AI infrastructure borrowers to refinance. Under Hayes’s thesis, those indicators would reveal whether AI remains a high-growth infrastructure expansion or begins turning into a concentrated credit problem.
Concerned about AI-fueled credit risks? Learn how DeFi and TradFi react when liquidity tightens and markets deleverage.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
