Operating profit per gigawatt
Operating profit per gigawatt
One-line summary: The unit metric for AI-compute monetization — not installed capacity, not capex, but the operating profit a counterparty extracts per gigawatt deployed — which re-ranks the hyperscalers by deal economics rather than by size and makes speed-to-energization a margin lever rather than a schedule detail.
The insight
Gigawatts are how AI compute is now counted, but capacity says nothing about return. gavin-baker proposes the denominator that matters: what operating profit does a gigawatt actually throw off, given the price of the deal signed against it?
The claim, from gavin-baker in 2026-06-11-podcast-bg2-pod-the-spacex-ipo-fable-5-ai-capex-update-market: "xai's deal with Google for cloud computing generates more operating profit per gigawatt than Anthropic, than Meta, than Google, than OpenAI."
Deal prices are dispersed and rising. andrew-fox in the same source: "The implied monetization rate on that number is something like $14 billion per gigawatt per year for the AI business. They just signed Anthropic at 22 to 23. They just signed Google at 50." A gigawatt sold to Google at ~$50B is not the same asset as a gigawatt sold to Anthropic at ~$22-23B.
Two consequences follow.
Speed becomes margin. Because the cost base is heavily fixed and labor-loaded during construction, the time to energize a gigawatt is a direct cost. gavin-baker: "We do know from Jensen that Elon brings data centers up faster than anyone. 122 days speed is literally cost because every day you're paying electricians and plumbers that's cost." This is the mechanism by which a construction advantage becomes a P&L advantage.
Incremental revenue is near-pure margin. clark-tang in the same source, on monetization rising into a fixed base: "All of that is very heavy fixed cost base." Rising $/GW against a fixed cost base flows disproportionately to operating profit — which is why the metric moves faster than capacity does.
The chain
Deal prices per gigawatt are dispersed ($14B implied, Anthropic ~$22-23B, Google ~$50B) → the operator who energizes fastest carries less construction cost per gigawatt → operating profit per gigawatt diverges sharply across counterparties even at identical installed capacity → the ranking of AI-cloud businesses by economics differs from the ranking by size → SpaceX/xAI screens as the highest-return operator. Canonical: elon-web-services-to-spacex-hyperscaler-rerate.
Why it matters to stock-market
It supplies the unit of account for the AI-cloud leg of the capex thesis. Most coverage counts gigawatts and capex dollars; this metric asks what each gigawatt earns. If it holds, it re-ranks spacex above incumbents on economics, and it is the load-bearing input to elon-web-services-to-spacex-hyperscaler-rerate.
It also bears directly on ai-inference-revenue-run-rate-dispute: a high operating profit per gigawatt is hard to reconcile with rajiv-jain's claim that "the economics are really bad."
Weaknesses
- Every figure here comes from investors long the asset, on their own podcast. None is a disclosed number.
- "Operating profit per gigawatt" is not a reported line item anywhere; it is reverse-engineered from announced deal values and assumed cost bases.
- The metric flatters whoever signed the most recent, highest-priced deal. It is a snapshot, not a run-rate — the same objection the falsifiers on elon-web-services-to-spacex-hyperscaler-rerate record.