SpaceXAI adopts NVIDIA Vera CPUs to scale agentic AI toward gigawatts of compute, including an orbital deployment

NVIDIA and SpaceXAI (Elon Musk's combined AI and space venture, formerly xAI) announced that SpaceXAI will deploy NVIDIA's new Vera CPU across its AI infrastructure, with plans to scale toward gigawatts of computing capacity on the Vera Rubin platform — including a version bound for orbit.

What's new

NVIDIA's announcement states that "SpaceXAI is taking this architecture from massive AI factories to the next frontier of computing in orbit," with SpaceXAI's first-generation Starmind AI satellite set to use an optimized NVIDIA Vera Rubin NVL72 system. On the ground, SpaceXAI will use Vera to accelerate agentic AI workloads: NVIDIA describes Vera as featuring 88 Olympus CPU cores with up to 1.2TB/s of memory bandwidth, delivering "up to 1.8x faster task completion compared with x86 CPUs."

Ian Buck, NVIDIA's vice president, said "Vera gives AI agents the CPU performance to act in real time — executing code, processing data and coordinating complex tasks." Mike Nicolls, president of SpaceXAI, said "Vera gives us the CPU performance and memory bandwidth to run enormous amounts of orchestration, code and data processing while keeping GPUs doing what they do best," adding that the result is "higher-performance AI agents and more useful work from every watt of compute." NVIDIA says that as SpaceXAI scales toward gigawatts of computing capacity, Vera Rubin gives it "a common architecture to efficiently scale its next generation of AI factories."

Context

The announcement lands the same day NVIDIA confirmed its Groq 3 LPX inference accelerator entered full production, and a week after NVIDIA said it would guarantee SB Energy's PORTS-Pike technology campus in Ohio to exclusively host NVIDIA AI compute — part of a broader run of NVIDIA infrastructure announcements this month. For SpaceXAI/xAI specifically, this follows a string of Grok-related product moves in August, including Grok 4.6's general availability and Grok Bot's expansion across more subscription plans; this NVIDIA deal is the infrastructure layer underneath that faster product cadence, and the orbital angle ties directly into SpaceX's existing satellite and launch infrastructure, distinguishing SpaceXAI's compute ambitions from purely terrestrial data-center buildouts pursued by other frontier labs.

Why it matters

A CPU built specifically for AI-agent orchestration — rather than raw model training throughput — reflects where the industry's marginal compute demand is shifting: agentic workloads spend enormous cycles on code execution, tool calls, and multi-step coordination that traditional GPU-centric training infrastructure isn't optimized to handle efficiently. NVIDIA positioning Vera as complementary to (not a replacement for) its GPUs, freeing GPUs to focus purely on model computation while Vera handles orchestration, is a bet that agent-heavy workloads need a different hardware split than pure training did.

The orbital compute detail is the more unusual signal here: extending accelerated AI infrastructure into space via a satellite is a distinctly SpaceXAI-specific move that other frontier labs, none of which control their own launch and satellite infrastructure, cannot easily replicate. Whether "AI factories in orbit" becomes a genuine infrastructure category or stays a one-off flagship project, it marks NVIDIA's chip architecture explicitly following SpaceXAI's ambitions beyond terrestrial data centers for the first time.