Three years ago, Sequoia partner David Cahn was among the first to calculate the consequences of Silicon Valley's massive spending on AI infrastructure. In 2023, he was responding to Nvidia's reported annual GPU revenue of $50 billion. Starting from that figure and factoring in the implied costs of running data centers along with their operators' margins, he concluded that $200 billion in revenue would be necessary to recoup the initial investment. He framed this as a challenge, urging entrepreneurs to develop AI products and services that could utilize and generate revenue from all that infrastructure.

Now, after three years of hyperscaling, Cahn has a new estimate for AI infrastructure spending in 2026: $1.5 trillion. Overall, he calculates that the AI industry will need to earn $3 trillion to justify all those chips and other data center expenditures. And that number is likely an underestimate—rising memory costs and the growing use of exotic or inference-specific chips will push it even higher. "Recently," he writes, "the required revenue per GW of CapEx has sharply increased due to these bottleneck dynamics and rising costs of construction."

On the other side of the equation, Anthropic is believed to have reached $60 billion in annual recurring revenue, while OpenAI reportedly earned $13 billion in 2025 (though in November 2025, it claimed $20 billion in ARR) and is presumably earning more this year. Still, there is clearly a large gap to close. One person monitoring that gap is Torsten Slok, chief economist at Apollo, the massive asset manager. In a recent note, he points out that the hyperscalers—Google, Meta, Microsoft, and Amazon—are all predicting massive accelerations in their free cash flow by 2028. That is, they expect to see the payoff from all those chips they purchased.

But what if that payoff doesn't materialize? Slok highlights a risk currently visible across AI usage: More organizations are turning to cheaper open-weight models, often from Chinese developers rather than frontier labs, and overall token prices are falling. OpenAI's latest model, according to CEO Sam Altman, is 54% more token-efficient on coding tasks. That's good news for users worried about the cost of their AI agents, but it could be bad for companies building token factories if users don't dramatically increase their overall token consumption with them.

Slok worries that if hyperscalers fail to meet their cash flow targets, the market reaction could be severe. "With so much riding on so few names," he writes, "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."

Just something to keep in mind as you steer your AI agents toward cheaper tokens.