SpaceX puts AI build-out exclusively on Nvidia hardware

SpaceX plans to build future artificial intelligence computing capacity exclusively with Nvidia hardware, including the company’s Vera Rubin architecture, according to chief executive Elon Musk. The commitment places SpaceX’s largest capital spending area with a single supplier.

SpaceX allocated US$15.8 billion to AI computing infrastructure in the second quarter, out of total capital expenditure of US$18.4 billion. AI-related spending rose from US$749 million a year earlier to US$7.7 billion in the first quarter before reaching its second-quarter level.

Installed computing capacity stood at 1.4 gigawatts at the end of June. Management expects capacity to exceed 2 gigawatts by the end of 2026 and said it may approach 10 gigawatts by the end of 2027.

The scale of the expansion leaves SpaceX exposed to Nvidia’s pricing and supply conditions. Nvidia has reported supply constraints, while production shipments of its Vera Rubin architecture only recently began. Musk said SpaceX expected to receive a “very significant” percentage of Nvidia’s GPUs next year, without specifying a quantity or binding public commitment.

SpaceX disclosed US$28 billion in non-cancellable purchase commitments at the end of June, including US$22.2 billion maturing in 2027. The commitments relate largely to AI infrastructure, cloud capacity and spectrum. Agreements with Anthropic and Google also specify Nvidia GPUs, although both contain termination or delivery-related conditions.

The exclusivity statement was made on an earnings call, while SpaceX’s 10-Q reportedly does not mention Nvidia and no subsequent filing cited in the source material confirms the arrangement. SpaceX is also planning a chip plant with partners, but its prospectus contains no definitive agreements.