Sequoia partner David Cahn, who three years ago first quantified the revenue needed to justify Silicon Valley's spending on AI infrastructure, has updated his calculations to reflect the industry's continued expansion. He now estimates that AI infrastructure spending will reach $1.5 trillion in 2026, and that the sector will ultimately need to generate around $3 trillion in revenue to pay back its investment in chips and data centres. The figure matters because it highlights a widening gap between what the industry is spending and what it is currently earning, with Cahn suggesting even $3 trillion may prove an underestimate given rising memory and construction costs.

Against that required revenue, Anthropic is thought to have reached $60 billion in annual recurring revenue, while OpenAI reportedly earned $13 billion in 2025 (having claimed a $20 billion ARR run rate by November 2025), leaving a substantial shortfall. Apollo chief economist Torsten Slok notes that hyperscalers Google, Meta, Microsoft and Amazon are all forecasting sharp accelerations in free cash flow by 2028 as their chip investments pay off, but warns of risks: organisations are increasingly turning to cheaper, often Chinese, open-weight models, and token prices are falling, with OpenAI's latest model said to be 54% more token-efficient on coding tasks. Slok cautions that if the hyperscalers miss their cash-flow targets, the fallout could be severe enough to tip the economy into recession and the S&P 500 into a correction, given how much rides on so few companies.