Nvidia–SpaceX Partnership: What Vera Rubin and Starmind AI1 Mean for Orbital AI

The Nvidia–SpaceXAI partnership is bigger than a conventional chip-supply agreement. SpaceX says it will build its AI infrastructure exclusively on Nvidia systems, using Vera CPUs and Rubin GPUs for agentic workloads on Earth and for a proposed orbital-computing platform called Starmind. 27

The near-term significance is Nvidia’s expansion from GPUs into the CPU, networking and rack-scale layers of AI infrastructure. The longer-term bet is that some data-center computing can move into orbit. That second idea remains a plan: the first launch and large-scale deployment have not yet demonstrated commercial performance in space.

What SpaceXAI is buying from Nvidia

Vera is Nvidia’s CPU platform for the work surrounding an AI model, including code execution, tool use, sandboxing, retrieval, data processing and orchestration. Those tasks are particularly important for agentic systems, which do more than generate a single response and instead perform multi-step actions.

SpaceXAI is expected to use Vera for terrestrial agentic-AI infrastructure associated with its AI services, including the computing layer supporting Grok-related workloads. The partnership also covers Vera Rubin rack-scale systems rather than isolated processors.

Nvidia’s standard DGX Vera Rubin NVL72 configuration combines 36 Vera CPUs and 72 Rubin GPUs, along with high-speed networking and data-processing components. The design is intended to keep CPU and GPU resources closely integrated inside a single AI system.

What is Starmind AI1?

Starmind AI1 is described as the first satellite in a planned constellation designed to run AI workloads directly in orbit. Nvidia and SpaceX are co-designing its compute payload, with each satellite expected to carry Rubin GPUs and Vera CPUs in an orbital adaptation of the Vera Rubin architecture. 515

Musk has said the first space-optimized Vera Rubin NVL72 system is targeted for launch in the fourth quarter of 2027, with “significant scale” anticipated in 2028. 3913

Those dates describe an intended deployment schedule, not a completed system. The public announcements do not establish how many satellites will launch, what workloads will be commercially available, or how the economics will compare with ground-based data centers after launch, operations and maintenance costs are included.

Why Musk says orbital AI could be different

Musk has characterized satellite-based data centers as simpler, lower-cost, denser and lighter than conventional terrestrial racks. 3 The appeal is straightforward: an orbital platform could place compute closer to some space-based data sources while avoiding the physical constraints of a traditional land-based facility.

The claim still needs to be tested in practice. A production orbital system would need to show that its compute hardware, power supply, thermal management, communications and reliability can work together over an extended mission. The partnership announcement provides a target architecture and launch plan, but not independent evidence that the proposed cost or density advantage has been achieved.

What Vera’s specifications show—and do not show

Nvidia says Vera contains 88 custom Olympus cores and supports up to 1.2 TB/s of LPDDR5X memory bandwidth. Nvidia also reports more than 1.8× higher sandbox performance than competing x86 CPU infrastructure across the agentic workloads used in its testing.

These specifications explain the role Vera is intended to play: handling the CPU-heavy inner loop around AI inference while helping keep GPUs supplied with work. They do not, by themselves, establish how Vera will perform inside Starmind, under space conditions, or in a complete customer workload. The 1.8× figure is a vendor-reported comparison rather than an independent benchmark.

SpaceX’s exclusive Nvidia commitment

Musk told investors that SpaceX would build its AI infrastructure exclusively on Nvidia systems. 27 That commitment could make SpaceX a prominent customer across Nvidia’s portfolio, including Vera CPUs, Rubin GPUs and integrated rack-scale infrastructure—not merely a purchaser of discrete graphics processors.

It also gives Nvidia a high-profile endorsement as it tries to own more of the AI-factory stack. Nvidia’s Vera materials position the CPU as both a standalone processor and a component inside Vera Rubin systems and storage platforms. 18 Its broader platform includes CPUs, GPUs, networking and data-processing hardware designed to operate as a coordinated system.

How Oracle fits into Nvidia’s CPU expansion

SpaceX is not the only early Vera customer. Nvidia identifies Oracle Cloud Infrastructure among the organizations adopting Vera, while Oracle has described an OCI Supercluster based on the Vera Rubin platform, including Rubin GPUs, Vera CPUs, BlueField DPUs and ConnectX networking. 18

Oracle has also said it plans to deploy hundreds of thousands of Vera CPUs beginning in 2026. That is a planned hyperscale rollout, not evidence that the final deployment has already reached that size. Still, it illustrates Nvidia’s broader objective: make Vera a core building block for large AI factories, whether the infrastructure is operated by a cloud provider, an AI lab or SpaceXAI.

Why SpaceX shares fell despite the announcement

The Nvidia partnership arrived alongside concerns about how aggressively SpaceX is spending on AI. Reuters reported $18.4 billion in quarterly capital spending, including $15.8 billion for AI infrastructure, and said SpaceX shares fell about 12% and moved below the company’s $135 IPO price.

That reaction suggests investors were evaluating the partnership as part of a much larger capital-allocation decision. Strong revenue or an important hardware relationship does not automatically make a large AI buildout profitable. Investors still need evidence about utilization, customer demand, financing and the time required to recover the investment.

Market reactions were not uniform across every session: Nvidia initially benefited when Musk described SpaceX as an exclusive Nvidia customer, while SpaceX fell as capital-spending concerns dominated its post-IPO trading. 2 The orbital opportunity is long dated, whereas the spending is immediate.

What Nvidia investors were watching

For Nvidia, the partnership is positive evidence that customers want complete AI systems spanning CPUs and GPUs. But it does not by itself establish near-term revenue, margin impact or the commercial viability of orbital data centers.

Investors were also focused on Nvidia’s upcoming fiscal second-quarter results scheduled for August 26. The earnings report was expected to provide more immediate information about demand, margins and management’s outlook than the still-developing Starmind program.

Bottom line

The partnership has three layers:

1. Immediate infrastructure: SpaceXAI plans to deploy Nvidia Vera CPUs for CPU-intensive agentic-AI work and use Vera Rubin systems on the ground.
2. Exclusive supply relationship: SpaceX says its AI infrastructure will use Nvidia systems exclusively. 27
3. Long-term orbital experiment: Nvidia and SpaceX are designing a space-optimized Vera Rubin NVL72 payload for Starmind AI1, targeted for launch in Q4 2027 and significant expansion in 2028. 3913

The first two layers strengthen Nvidia’s platform strategy today. The third could become a new model for AI infrastructure, but it remains an ambitious engineering and business plan until Starmind satellites are launched, operated and shown to deliver reliable computing at a competitive cost.