The Trillion-Dollar Bet Against Payroll

Follow the money this month, because the money stopped being subtle.

Nvidia and OpenAI announced a letter of intent for up to $100 billion of investment tied to ten gigawatts of AI datacenters. Oracle's stock had its best day since 1992 on the strength of enormous multi-year AI compute contracts. Add Stargate, add Meta's Louisiana campus, add Anthropic's own buildout — and the industry's committed infrastructure spend now runs comfortably into the trillions when you extend the pledges.

Ask the only question that matters for this series: what revenue justifies a trillion dollars of specialized computers that depreciate in five years? Consumer subscriptions don't. Advertising doesn't. There is exactly one pool of spending on Earth large enough to underwrite this: the roughly sixty trillion dollars the world pays every year for human labor. The capex supercycle only makes sense as a bet that a meaningful slice of global wages converts into spending on machine cognition. The investors know it. The labs know it — Anthropic's own Economic Index published usage data this month showing adoption concentrating in exactly the white-collar tasks you'd expect, with automation-style usage rising against augmentation-style.

This is not a complaint

I want to be precise: converting labor cost into abundance is the good future, if the winnings circulate. Star Trek's economy is also "post-payroll." The difference between that future and a dystopia is not the technology — it's who owns the machines and how the surplus flows. A trillion dollars of private compute with no distribution mechanism is how you get the hourglass economy I wrote about in July, at planetary scale.

Actionable Steps

Policymakers: the tax base is going to migrate from wages to capital returns on compute — get ahead of it; study sovereign-wealth-style public stakes in AI infrastructure, and the data-dividend proposals while they're still cheap to design. Investors and founders: the durable margin will sit with whoever owns distribution and trust, not raw tokens — those are being commoditized by the gigawatt. Workers: watch where the compute lands geographically; datacenter regions are getting new tax bases and new jobs — a modern version of following the railroads.

Sixty trillion in wages. One trillion in silicon, and counting. The bet has been placed; the only open question is who's at the table when it pays out.