Three years ago, Sequoia Capital partner David Cahn posed what became known as “AI’s $200 billion question” — the gap between what the industry was spending on Nvidia GPUs and what it needed to earn back. Today, that question has grown by an order of magnitude, and the answer is no clearer.
Cahn’s updated analysis puts AI infrastructure spending in 2026 at US$1.5 trillion (approximately £1.2 trillion). To pay back that investment, the industry must generate US$3 trillion in revenue — and Cahn warns that figure is likely an underestimate, given rising memory costs and the growing expense of specialized inference chips.
“Recently, the required revenue per GW of CapEx has sharply increased due to these bottleneck dynamics and rising costs of construction,” Cahn wrote.
The revenue gap
Current AI-industry revenue tells a mixed story. Anthropic has reached approximately US$60 billion in annualized revenue. OpenAI reported US$13 billion in revenue for 2025 and claimed a US$20 billion annualized run rate by November 2025.
But against a US$3 trillion target, even these eye-catching numbers are a fraction of what is needed. The hyperscalers — Google, Meta, Microsoft, and Amazon — all project massive free-cash-flow payback arriving by 2028, having collectively accelerated infrastructure spending far beyond their traditional cloud buildouts. Big Four combined capital expenditure is expected to reach US$725 billion in 2026, up 77 percent from 2025, with roughly three-quarters of that tied directly to AI infrastructure.
Risks to the payback scenario
Three structural risks complicate the industry’s path to breakeven:
1. Commoditization. Cheaper open-weight models, many developed in China, are gaining adoption at the expense of proprietary frontier models. This compresses the pricing power of incumbents.
2. Falling token prices. Sam Altman noted that OpenAI’s latest model is 54 percent more token-efficient on coding tasks than its predecessor. Efficiency is good for users, but it reduces revenue per query unless total token volume grows proportionally — a dynamic that has already prompted debate about whether AI companies are effectively competing against their own previous pricing.
3. Systemic market risk. Torsten Slok, chief economist at Apollo Global Management, warned that the concentration of AI investment in a handful of hyperscaler names creates a single-point-of-failure risk for the broader market. “With so much riding on so few names, a slower payoff wouldn’t just be a sector problem, it would risk tipping the economy into recession and the S&P 500 into a correction,” Slok wrote.
Echoes of the dot-com era
The parallels to the late-1990s telecom buildout are hard to ignore. During the dot-com boom, telecom operators laid more than 130 million kilometers (80 million miles) of fiber optic cable based on inflated demand forecasts — much of which was never lit. Today’s US$600 billion-plus AI infrastructure cycle follows a similar playbook: building enormous capacity for demand that has yet to materialize in any form approaching the scale of the investment.
There are differences. The Nasdaq-100 trades at approximately 28 times forward earnings today, versus 89 times at the 1999 peak. And the hyperscalers generating this spending are real, profitable businesses — not speculative startups. But a physical infrastructure bubble is harder to unwind than a stock bubble. Hardware depreciates, and next-generation Nvidia GPUs will accelerate the obsolescence of current-generation accelerators, compounding losses if revenue does not catch up.
The unanswered question
The AI industry is not necessarily heading for a crash. But the central question Cahn posed three years ago remains unanswered: who is going to pay for all this? Until the revenue side of the equation catches up with the capital expenditure, the gap is a risk that grows with every new data center and GPU cluster the hyperscalers bring online.
Sources: Can AI answer the $3 trillion question? (TechCrunch, July 9, 2026); The $600 Billion Question (iDX Insights); AI’s $600B Question (Sequoia Capital)