SpaceX Prices IPO While Locking Large AI Compute Deals

SpaceX filed for an initial public offering that Reuters reports will price at $135 per share, aiming to raise $75 billion. Economic Times reports secondary coverage indicating investor demand near $150 billion. Regulatory filings and multiple news outlets show SpaceX has announced large compute contracts ahead of the IPO. CNBC reports a filing that Google will pay $920 million per month for access to about 110,000 NVIDIA GPUs and related compute from October 2026 through June 2029. TechCrunch and DatacenterDynamics report a separate agreement giving Anthropic access to significant capacity at SpaceX data centers for roughly $1.25 billion per month. Economic Times cites SpaceX financials showing $18.7 billion in 2025 revenue and $6.6 billion in adjusted EBITDA, and reports that xAI recorded approximately $6.4 billion in operating losses last year that were moved onto SpaceX's balance sheet after their merger, according to the prospectus.

Per filings covered by CNBC and TechCrunch, the Google agreement covers roughly 110,000 NVIDIA GPUs, plus CPUs, memory, and related components, with capacity ramping through September 2026 at a reduced fee and contractual termination/cancellation clauses tied to delivery milestones. TechCrunch notes the Anthropic arrangement centers on capacity at the Colossus 1 data center; reporting also indicates CEO Elon Musk has previously suggested reserving a Colossus 2 facility for xAI. Wired and TechCrunch reporting further notes heavy capex related to data-center infrastructure, including commitments to purchase gas turbines; Wired reports a recent $2.8 billion purchase of turbines to power data centers, as disclosed in IPO documents.

Editorial analysis: Companies across cloud and AI infrastructure have been racing to secure compute capacity for large models, and observers framed SpaceX's commercial deals as part of that pattern. Major cloud customers have used multi-year capacity contracts and captive facilities to manage shortages and pricing volatility; the reported deals with Google and Anthropic fit established patterns where buyers pay premiums for immediate, guaranteed capacity. Industry coverage also places the space-based data-center language in SpaceX's prospectus as a speculative, long-term infrastructure idea that several outlets have highlighted as a differentiator in the story around the IPO.

Editorial analysis: For ML engineers and infra teams, the headline items to extract from reporting are the scale and contractual structure of capacity access. The reported 110,000-GPU figure and multi-month, high-dollar agreements underscore continued demand pressure for high-memory, high-bandwidth GPU pools. Practitioners working on model deployment, cost forecasting, and capacity planning should view these deals as signals that large consumers will continue to secure bespoke capacity arrangements outside traditional hyperscaler spot markets.

Editorial analysis: Observers will monitor the IPO pricing and subscription dynamics cited by Reuters and Economic Times for indications of how public markets value vertically integrated compute propositions. Reporting also leaves open delivery risk and operational execution as observable indicators, since filings cited by CNBC and TechCrunch include clauses allowing contract termination if GPU delivery milestones are missed. Finally, practitioners should watch whether the market for long-term reserved GPU capacity tightens or loosens as these large contracts come online, and whether colocation and bespoke data-center builds become more common among major AI model developers.