Can artificial intelligence answer the $3 trillion question?

Three years ago, Sequoia partner David Kahn was one of the first people to do the math and put a number on the implications of Silicon Valley’s massive spending on AI infrastructure.

in 2023that was a reaction to Nvidia’s reported annual GPU revenue of $50 billion. Starting with that number, and adding the implicit costs of operating data centers and the profit margins of their operators, he concluded that it would take $200 billion in revenue to repay the upfront investment.

He saw it as a challenge, asking entrepreneurs to create AI products and services to leverage all that infrastructure and generate revenue from it. Fast forward to today, adding three years of hyper-expansion, and Kahn has… New number AI infrastructure spending for 2026: $1.5 trillion.

In total, he figures, the AI ​​industry would have to earn $3 trillion to justify all those chips and other data center expenses. This may be an underestimate, as rising memory costs and the increasing use of exotic or proprietary chips will cause this number to rise. “Recently, the revenue required per gigawatt of capital expenditure has increased sharply due to these bottleneck dynamics and rising construction costs,” he writes.

On the other side of the ledger, Anthropy is thought to have achieved great success $60 billion in ARRwhile OpenAI reportedly gained $13 billion In 2025 (although in In November 2025, it said it had reached US$20 billionHe is supposed to earn more this year. But there is clearly a big gap that needs to be filled.

One person paying attention to this gap is Torsten Slok, chief economist at Apollo, the giant asset management company. In a Recent noteHe points out that the big companies — Google, Meta, Microsoft, and Amazon — all expect a huge acceleration in their free cash flow in 2028. That is, they expect to see the payoff from all those chips they bought.

What if they don’t? Sloak points to the risks we are currently seeing in the use of AI: more institutions switching to cheaper open-weighted models, often Chinese, rather than those made by frontier labs, and generally lower token prices. OpenAI’s latest model, according to CEO Sam Altman, is 54% greater token efficiency In coding tasks. This is a good thing for users who are concerned about the cost of their AI agents, but it could be bad for companies building token factories if users don’t significantly increase their overall token usage with them.

Slok is concerned that if super-expanders don’t meet their cash flow targets, the market reaction could be severe –
“With so much reliance on so few names, a slower return would not only be a problem for the sector, it could risk pushing the economy into a recession and the S&P 500 into a correction,” he writes.

It’s just something to keep in mind as you point your AI agents towards cheaper tokens.