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 titanic spending on AI infrastructure.
In 2023was reacting to Nvidia’s reported annual GPU revenue of $50 billion. Starting with that number and adding in the implicit costs of running the data centers and margins for their operators, he concluded that $200 billion in revenue would be required to recoup the initial investment.
He took it as a challenge, asking entrepreneurs to find AI products and services to make use of and monetize all this infrastructure. Fast forward to today, adding up to three years of hyper-escalation, and Cahn has one new number for AI infrastructure spending for 2026: $1.5 trillion;
In total, he estimates the AI industry would need to earn $3 trillion to justify all those chips and other data center spending. And that’s probably an underestimate – the rising cost of memory and the increasing use of exotic or inference-specific chips will increase that number. “Recently,” he writes, “required revenue per GW CapEx has risen sharply due to this dynamic bottleneck and rising construction costs.”
On the other side of the ledger, Anthropic is believed to have struck $60 billion in ARRwhile OpenAI reportedly won 13 billion dollars in 2025 (although in November 2025, it said it was at $20 billion ARR) and probably earns more this year. But there is clearly a big gap that needs to be filled.
Someone interested in this gap is Torsten Slok, chief economist at Apollo, the giant asset manager. In one recent notepoints out that the hyperscalers — Google, Meta, Microsoft and Amazon — are projecting huge accelerations in their free cash flow in 2028. That is, they expect to see all those chips they bought pay off.
What if they don’t? Slok notes a risk we’re currently seeing across AI use: more organizations are turning to cheaper open-weight models, often Chinese, rather than those made by frontier labs, and overall token prices are falling. OpenAI’s latest model, according to CEO Sam Altman, is 54% more efficient token in coding tasks. This is good for users who are concerned about the cost of their AI agents, but it can be bad for companies building token factories if users don’t wildly increase their overall token usage with them.
Slok worries that if hyperscalers fail to meet their cash flow targets, the market backlash could be severe —
“With so many names,” he writes, “a slower payback wouldn’t just be an industry problem, it would risk sending the economy into recession and the S&P 500 into a correction.”
Just something to keep in mind as you shop your AI agents towards cheaper brands.
When you purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.
