Three years ago, Sequoia partner David Cahn calculated that $200 billion in revenue would be needed to recoup the upfront costs of AI infrastructure, based on Nvidia’s $50 billion annual GPU revenue in 2023. Now, with three years of hyperscaling, he estimates AI infrastructure spending will reach $1.5 trillion in 2026, requiring $3 trillion in returns to justify the investment. He warns that rising memory costs and the use of exotic chips will likely push this figure even higher. 'Recently, the required revenue per GW of CapEx has sharply increased due to these bottleneck dynamics and rising costs of construction,' he writes.
Anthropic is thought to have hit $60 billion in annual recurring revenue, while OpenAI reportedly earned $13 billion in 2025, although it later said it was at $20 billion ARR. Despite these figures, a significant gap remains between the current revenue and the projected returns needed to justify the massive capital expenditures. Torsten Slok, chief economist at Apollo, highlights that hyperscalers like Google, Meta, Microsoft, and Amazon are forecasting massive free-cash-flow increases by 2028, indicating they expect to recoup their investments in AI infrastructure.
Slok raises concerns about the risks if hyperscalers fail to meet their cash flow targets. He notes a growing trend of organizations turning to cheaper open-weight models, often Chinese, and the overall decline in token prices. OpenAI’s latest model, according to CEO Sam Altman, is 54% more token efficient on coding tasks. While this benefits users, it may harm companies building token factories if users do not significantly increase their token usage. Slok warns that a slower payoff could risk tipping the economy into recession and the S&P 500 into a correction.
Source: techcrunch
