Key points
- SpaceXAI is adopting Nvidia's Vera CPU, an 88-core chip built for AI agents, to run the orchestration and code execution behind its Grok models.
- Elon Musk says a space-optimized Vera Rubin NVL72 will launch to orbit in Q4 2027, the first node of the Starmind orbital-compute plan, with scale in 2028.
- The public plays are SpaceX (SPCX) and Nvidia (NVDA), which holds a $21 billion SpaceX stake and reports earnings August 26.
SpaceX (SPCX) and Nvidia (NVDA) said Monday that they are building the compute behind a new generation of orbital artificial intelligence, and attached a date to the most ambitious piece of it. Elon Musk said the two companies have designed a space-optimized version of Nvidia's Vera Rubin NVL72 system for launch to orbit in the fourth quarter of 2027, with what he called significant scale the year after.
The nearer-term part is already shipping inside Nvidia's product line. SpaceXAI, the unit behind the Grok chatbot, is adopting Nvidia's new Vera CPU to run the work that surrounds an AI model rather than the model itself.
What SpaceXAI is actually buying
Nvidia calls Vera the first CPU built for AI agents. It carries 88 of the company's in-house Olympus cores and, by Nvidia's own account, finishes agent tasks up to 1.8 times faster than a standard x86 processor. The chip is not meant to do the heavy model math, which still runs on GPUs. Its job is the connective work between inference calls: tool orchestration, code execution, data processing and simulations. Those are the steps that decide how quickly an AI agent can act.
In Nvidia's words, the aim is "keeping GPUs fed and agents acting fast." SpaceXAI plans to pair the chip with Nvidia's wider Vera Rubin platform as it builds the infrastructure behind Grok toward several gigawatts of computing capacity.
The orbital plan, and where it stops being certain
The more speculative half is Starmind, SpaceX's plan to put AI compute in space. The first Starmind satellite is designed around the same space-optimized Vera Rubin NVL72 that Musk described, and the argument is that a rack built for orbit can be simpler and cheaper than a standard data-center design.
SpaceX, in partnership with Nvidia, has designed a space-optimized Vera Rubin NVL72 system for launch to orbit in Q4 next year, with significant scale in 2028
See Elon Musk's post on X
That timeline is Musk's own, and it is a target rather than a shipped product. Running a data center in orbit means solving power, heat, radiation and communications at the same time, and nothing announced this week shows a working unit in space. Musk has described a longer-term vision of a constellation of AI satellites acting as distributed compute nodes, but the first launch he has named is still more than a year away.
What it means for SPCX and NVDA
SpaceX is unusual among Musk's ventures in that public investors can own it directly, through SPCX, after its listing this summer. The interest has been heavy: we counted twenty hedge funds holding $35 billion of the stock in the latest 13F filings. Nvidia is the other side of the trade, supplying the silicon, and it has disclosed a stake in SpaceX worth about $21 billion, which ties the two companies together on the balance sheet as well as in the product.
The news also lands in front of a major catalyst. Nvidia reports earnings on Wednesday, August 26, the print the whole AI trade is waiting on. By midday Monday, Nvidia traded near $210 and SpaceX near $136, both a little lower on the session.
For now, the split is clean. The Vera is in production, and SpaceXAI says it will deploy the CPU, although neither company disclosed timing or contract value. Starmind remains a 2027 launch target with major engineering risks still unresolved.
Prices as of midday on August 24, 2026. This is general market information, not investment advice.
