First author of SemiAnalysis SpaceX 10GW article; leads inference economics modeling (Inference X) and supply chain tracking for AI data centers.
no scored calls yet — needs a stated position or a categorical verdict, with a matured window vs SPY
SpaceX can build 10GW of data centers by 2027 using warehouse sites, Chinese electrical equipment, and mobile turbines, monetizing at $50-100M/MW/year via 90-day cancellation on-demand contracts. Operating cash flow from initial 2GW ($4B/month) plus Nvidia vendor financing funds expansion. Execution risk is low given Memphis/Mississippi precedent of 2GW built faster than industry using 3x less labor.
Microsoft has 7GW offtake with OpenAI but paused data center builds in H2'24/H1'25, creating capacity gap in late 2026/H1'27. Now accelerating via pre-leasing (7GW YTD), neocloud deals (Nscale), and behind-the-meter (Chevron 2.7GW). Must buy SpaceX on-demand capacity at $50M/MW/year to monetize GPT tokens at $100M/MW/year. 90-day cancellation makes it balance-sheet light.
Anthropic and OpenAI combined add $20-30B ARR/month at 85% gross margins, validating $100M/MW/year token economics that underpin SpaceX pricing power.