
An investor warns that AI's capital cycle has overextended, built on self-reinforcing financing rather than proven demand.
Hyperscalers hold $1.8 trillion in future commitments while borrowing costs surge globally.
If AI customers disappoint, the entire financing structure—which props up semiconductor valuations and keeps capital flowing—risks collapse, potentially destabilizing markets by end of 2026.
What happened
An investor argues that the AI capital cycle has overextended beyond what economics can support, with hyperscalers and Nvidia now carrying roughly $1.8 trillion of off-balance-sheet commitments (including about $982 billion in purchase obligations and $822 billion in leases not yet commenced), while long-term borrowing costs have surged—Japan's 10-year JGB reached 2.93% (its highest since 1996) and the U.S. 30-year Treasury hit 5.27% (not seen since 2007).
Why it matters
The author contends that AI's financing structure has become self-reinforcing: companies raise capital at high valuations to buy compute, which generates revenue for semiconductor suppliers and data center operators, supporting higher valuations that enable more financing. This circular flow masks whether the underlying demand for AI infrastructure can actually justify the trillions being deployed. If this cycle breaks—say, if AI customers disappoint and cut compute purchases—semiconductor earnings would fall at the exact moment investors compress valuations, potentially triggering cascading losses across the ecosystem.
What to watch
The author predicts the AI boom may begin to unravel around the end of 2026 and beginning of 2027, citing five key reasons including rising household debt (American households carried roughly $18.8 trillion of debt as of the end of June 2026, up about $400 billion from a year earlier) and elevated delinquencies (4.7% of all outstanding household debt was in some stage of delinquency as of Q2 2026). Goldman Sachs has predicted hyperscaler AI capex could reach as much as $1.4 trillion in 2027.
Ask the AI about this article →
The author frames the AI boom not as a technology problem but as a capital cycle problem. While acknowledging that AI is genuinely transformative, the piece argues that investors have extrapolated AI's eventual dominance backward into today's valuations and financed a decade or more of buildout in advance—a pattern the author notes has recurred with railroads, fiber optic cable, and the internet itself. In each case, the technology was revolutionary but investments far outpaced sustainable returns, destroying shareholder wealth despite long-term economic benefit. The specific danger with AI, the author contends, is the structural circularity: Nvidia and other suppliers are now financing their own customers, meaning revenue growth depends on capital availability rather than underlying demand. This creates reflexive dynamics where minor disappointments cascade—if AI customers spend less because financing tightens, compute orders fall, semiconductor earnings decline, and the multiple investors pay for those earnings compresses simultaneously, amplifying losses. The timing compounds the risk: the AI boom is reaching its most aggressive phase precisely when households are carrying record debt loads (American households held roughly $18.8 trillion as of the end of June 2026, with credit card and auto loan balances near all-time highs), as asset prices that have propped up household balance sheets face renewed pressure from higher rates, and as bond markets globally are signaling that risk is underpriced. The author's analogy—a poker table where the same chips circulate to create the illusion of wealth—suggests that even without fraud or irrationality, the system becomes fragile once the flow stops. Goldman Sachs estimates hyperscaler AI capex could reach $1.4 trillion in 2027, while Morgan Stanley calculates roughly $1.8 trillion in existing commitments, but these forecasts rest on the financing conditions remaining supportive, a assumption the author doubts will hold through the end of 2026.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
Israeli startup DataAgent Ltd
Taoyuan is positioning itself as a northern hub for AI data centers (AIDC), citing the Tatan area and an LNG c…

SK Hynix presented a custom HBM concept at SEMICON Taiwan 2026, where compute functions are placed in the base…

The U.S. Department of Defense announced on August 31 that it has deployed ChatGPT Mil, a customized version o…

Nvidia reported earnings that were both remarkable and boring, reflecting its focus on avoiding a consolidated…

Anthropic has agreed to a $35bn cloud-computing contract with Lambda, a Nvidia-backed cloud provider
