SpaceX Goes All-In on Nvidia, Reserving a Big Chunk of 2027 GPU Supply

SpaceX has committed to using Nvidia chips exclusively for its AI workloads, a decision that locks in a major customer for Nvidia's next-generation Vera Rubin platform and gives SpaceX priority access to a significant share of Nvidia's 2027 GPU supply.

CEO Elon Musk made the announcement on SpaceX's first post-IPO earnings call on August 4, 2026. "Going forward, we've decided to build exclusively on Nvidia because we think the Vera Rubin architecture is the best architecture," Musk said. "We think it's the best AI computer, and we greatly value our close cooperation and partnership on many levels with Nvidia. So we're exclusive to Nvidia."

The commitment covers both terrestrial and orbital AI compute. SpaceX is partnering with Nvidia to design the Starmind AI1 satellite compute payload, with each satellite carrying Nvidia Rubin GPUs and Vera CPUs for "datacenter-class space compute." Musk described the Starmind satellite as "essentially an optimized Vera Rubin NVL72 computer." First launches are expected in 2027.

Vera Rubin NVL72: The Core of the Deal

The primary system is Nvidia's Vera Rubin NVL72 rack-scale design, which combines 72 Rubin GPUs with 36 Vera CPUs in a single rack. SpaceX views this architecture as superior to standard rack-style alternatives, according to the company's remarks. Nvidia also offers a space-certified variant, the Space-1 Vera Rubin module, which packs up to four Rubin GPUs and two Vera CPUs and claims up to 25 times the AI compute of the prior H100 generation for orbital workloads.

The deal positions SpaceX and its AI affiliate xAI among the world's largest AI infrastructure operators. Musk said SpaceX expects to receive "a very significant percentage" of Nvidia's GPUs next year, a statement that implies a substantial allocation of Nvidia's 2027 production capacity.

Orbital Data Centers as a Workaround

SpaceX frames the Starmind constellation as a way to bypass terrestrial constraints on land, power, and cooling. By placing compute in orbit, the company can avoid the physical and regulatory bottlenecks that have slowed ground-based data center expansion. The orbital payloads are designed to deliver datacenter-class performance in space, a technical challenge that Nvidia's space-certified modules are meant to address.

The partnership is not new. Nvidia hand-delivered the first DGX-1 supercomputer to SpaceX in 2016. In 2025, Jensen Huang personally delivered a DGX Spark system to Musk at Starbase. In May 2026, Nvidia hand-delivered early Vera CPU systems to SpaceXAI, alongside Anthropic, OpenAI, and Oracle. Nvidia has publicly congratulated SpaceX on its IPO and highlighted the multi-year collaboration.

Market Reaction and Competitive Implications

Nvidia shares rose sharply in the sessions following the call, with reports of gains of 3 to 4 percent or more. AMD shares came under pressure, as the exclusive commitment reduced the perceived multi-vendor opportunity. SpaceX had previously been viewed as open to both Nvidia and AMD in some earlier commentary, so the exclusivity removes a potential growth avenue for AMD in high-performance AI compute.

The deal also supports SpaceX's broader AI ambitions, including xAI's Grok models and the recent Cursor acquisition. The Starmind constellation adds an orbital dimension to that strategy, potentially creating a new class of space-based AI services.

Analysts will watch how Nvidia allocates its 2027 GPU supply, given that SpaceX's "very significant percentage" claim suggests a large reservation. Nvidia has not disclosed specific allocation numbers, and the exact terms of the exclusivity agreement remain private. The deal could strain Nvidia's ability to meet demand from other hyperscalers and AI startups, though Nvidia has not indicated any supply constraints.

SpaceX's move is a bet on Nvidia's roadmap and a signal that the company sees orbital compute as a viable extension of its satellite business. Whether the Starmind satellites can deliver on the promise of datacenter-class performance in orbit remains to be tested, but the partnership gives Nvidia a high-profile showcase for its space-certified hardware.