$NVDA's sales per gigawatt keep rising, so AI budgets buy less capacity

Our read: on its next chip platform, Nvidia expects about $40B of its own sales per gigawatt of AI data center, which means the all-in cost of a Rubin-era gigawatt will sit well above our $40.6B central estimate, which is based on today's spending.

What it means for a regular investor: the same big-tech budgets buy less AI capacity than the headlines suggest. Each gigawatt (roughly a large power plant's worth of electricity) costs more as chips get pricier.

The numbers:
- Nvidia on its August earnings call: its revenue opportunity went from about $18B per gigawatt (Hopper) to $25B (Blackwell) to $40B (Vera Rubin, its next chip platform). That's Nvidia's own share, not the whole building.
- Jensen Huang at GTC Taipei 2026, all-in: a gigawatt AI factory costs "$50, $60 billion, and soon … $80, $100 billion per gigawatt."

So what does ~$820B of 2026 spending by the five biggest spenders buy? (our calc, scenarios not guidance)
- At $50–60B per gigawatt (today, per Huang): about 14–16 GW
- At $80–100B (Rubin era, per Huang's "soon"): about 8–10 GW; even with 50% more spend in 2027, about 12–15 GW

What would prove us wrong

Set before publication. We report back in public when one trips.

• all-in costs per gigawatt stay near today's levels as Rubin ships, or a lot of capacity gets funded through leases that don't show up in these budgets. Open · next check: $NVDA's next earnings call (Nov 17) and Rubin ship dates

Method and sources

Figures marked "our calc" are our own arithmetic on the cited primary sources (company filings, earnings releases, public datasets). Every figure was checked against its source before publication.

Original post on X → Download data note

General research, the same for every reader. Not investment advice.