SpaceX ramps compute push as Microsoft demand surges

SpaceX is being positioned as a major supplier of near-term AI compute after Elon Musk said the company could add 6-8GW in 2027, with potential to exceed +10GW. SemiAnalysis sees SpaceX on track to build about 10GW by year-end 2027, despite the implied $300-500B in capex at 50B per GW.

The economic case rests on scarcity: large-scale, near-term compute can command up to $50B/GW/year, while frontier AI labs could generate over $100B/GW/year from API inference on GB300 clusters. SemiAnalysis models around $12B/GW/year of annual cost using $3/GPU-hr rental pricing, making inference highly profitable for companies with leading models.

Microsoft is described as the most likely large offtaker because its access to OpenAI models lets it pursue similar revenue per MW without bearing training costs. The April 2026 OpenAI deal removed a 20% revenue share, while Azure growth could accelerate from ~42% to over 100% if Microsoft can procure enough capacity.

Financing is expected to rely on Nvidia vendor support and premium pricing for fast delivery. SpaceX’s claimed advantage is speed: clusters delivered with 3-5 months lead time and priced at 30-50M/MW/year could support a path to $300B of ARR by the end of 2027 if only 50% of incremental compute is monetized.