SpaceX 10GW in 2027 – Why It’s Real, Will Drive $300B ARR for SpaceX, and Why Microsoft Will Be the Largest Offtaker
Jeremie Eliahou Ontiveros
· SemiAnalysis
· August 07, 2026 at 20:08
· ⏱ 12 min read
| Read on Substack ↗
Summary
SpaceX can plausibly build ~10GW of AI datacenter capacity by end-2027 and monetize it at premium prices because near-term large-scale compute is extremely scarce and frontier inference margins are far above cost. This validates an enormous capex supercycle in which Microsoft is likely the anchor offtaker, Nvidia financing underpins SpaceX's buildout, and gas-powered datacenter construction becomes a critical bottleneck.
•Elon Musk 'conservatively' targets 6-8GW incremental build in 2027, potentially +10GW; at $50B/GW that equals $300-500B capex in one year.
•SemiAnalysis estimates SpaceX ends 2026 at only ~2GW but will reach ~10GW by end-2027; its Southaven gas site expanded from 27 turbines (~495MW) in Feb 2026 to 69 turbines (>1.2GW) by July 2026.
•Microsoft signed over 10GW of binding contracts year-to-date — leases, self-build, PPAs, ESAs — worth ~$300B, reversing the leasing pause SemiAnalysis flagged in Dec 2024.
•Microsoft's reworked April 2026 OpenAI deal dropped the old 20% revenue share, so Microsoft can monetize OpenAI models at similar ~$100M/MW/year rates and potentially accelerate Azure growth from ~42% to over 100%.
•SemiAnalysis models say OpenAI and Anthropic can generate over $100B/GW/year in API inference revenue, far above ~$12B/GW/year rental cost for a GB300 cluster.
•SpaceX can fund its buildout with Nvidia vendor financing (the likely reason Elon declared Nvidia exclusivity) plus premium pricing at $30-50M/MW/year for 3-5 month lead-time compute.
The article says Elon declared Nvidia exclusivity likely because Nvidia offered vendor financing, and the entire >$100M/MW inference economics are built on GB200/GB300 clusters — implying sustained Nv
The article says Elon declared Nvidia exclusivity likely because Nvidia offered vendor financing, and the entire >$100M/MW inference economics are built on GB200/GB300 clusters — implying sustained Nvidia demand and pricing power.
Risk: Exposure is tied to SpaceX/xAI's aggressive buildout; any slowdown in AI capex or a shift to custom silicon could pressure the thesis.
Microsoft's April 2026 OpenAI deal dropped the 20% revenue share, it has signed 10GW of contracts, and the article estimates Azure revenue growth could accelerate from ~42% to over 100% by monetizing
Microsoft's April 2026 OpenAI deal dropped the 20% revenue share, it has signed 10GW of contracts, and the article estimates Azure revenue growth could accelerate from ~42% to over 100% by monetizing OpenAI-model inference.
Risk: Large binding commitments (~$300B) and reliance on SpaceX delivery timelines create execution and balance-sheet risk.
The article states SpaceX/xAI actively evaluated alternatives like AMD but likely abandoned them because Nvidia's vendor financing made the financial case; losing a high-profile, high-volume customer
The article states SpaceX/xAI actively evaluated alternatives like AMD but likely abandoned them because Nvidia's vendor financing made the financial case; losing a high-profile, high-volume customer is a competitive setback.
Risk: AMD still has other AI customers; the lost SpaceX opportunity is one datapoint rather than a fundamental product failure.
This newsletter, published August 07, 2026,
features Jeremie Eliahou Ontiveros
discussing NVDA, MSFT, AMD.
3 trade ideas extracted by AI with direction and confidence scoring.