SpaceXAI takes Nvidia's agent CPU as Mistral explores HUMAIN compute in Saudi Arabia
Nvidia said SpaceXAI will deploy Vera for agent orchestration and put Grok and a Starmind satellite on Vera Rubin. Mistral announced a HUMAIN collaboration of hundreds of millions of euros, without published contract terms.
Artificial Intelligence··Night
Vera sits between the inference calls
Nvidia said SpaceXAI will deploy its Vera CPU for the tool coordination, code execution and data processing that sit between inference calls. On Nvidia's own figures Vera carries 88 Olympus cores, reaches 1.2 terabytes per second of memory bandwidth over LPDDR5X and finishes agentic workloads up to 1.8 times faster than x86 CPUs; the measurements are the company's own. SpaceXAI president Mike Nicolls says the CPU takes over orchestration and data processing so the GPUs can stay on their own work.[1]
Grok and Starmind share the same architecture
The same announcement says Grok's infrastructure will expand on the Vera Rubin platform and that the first Starmind satellite will be based on that architecture. The first Starmind satellite is based on a Vera Rubin NVL72 rack system adapted for orbit.[1]
Mistral explores local compute with HUMAIN
The French model builder Mistral announced a collaboration with Saudi-based HUMAIN on the scale of hundreds of millions of euros. The partnership covers local compute in the region, frontier models optimised for Arabic and a joint route to market in regulated industries. The announcement says Mistral will explore using HUMAIN's data-centre infrastructure for its local compute needs, and that joint model work will focus on cybersecurity and voice applications. The target sectors are listed as financial services, manufacturing, telecommunications, cybersecurity and the public sector. The text defines sovereign AI as keeping data, intelligence, compute and operations under the customer's control. No investment figure, timetable or binding contract terms were published.[2]
