NVIDIA just landed one of its most ambitious AI deployments yet. SpaceXAI, the AI arm behind Grok, will deploy NVIDIA Vera CPUs to power its next generation of agentic AI applications — and the partnership doesn’t stop on Earth.
What Makes Vera Different
Most AI infrastructure conversations centre on GPUs, but Vera is something new: NVIDIA is calling it the first CPU built specifically for AI agents. Agentic AI workloads increasingly depend on CPUs to orchestrate tools, execute code, process data, and run simulations between model calls — tasks that sit outside the raw inference work GPUs handle. SpaceXAI will use Vera to accelerate exactly that kind of orchestration, keeping its GPUs fully fed and utilized rather than sitting idle between tasks. For more on how AI infrastructure is evolving, check our AI and technology coverage.
Vera’s Specs at a Glance
| Feature | Detail |
|---|---|
| Cores | 88 NVIDIA-designed Olympus cores |
| Memory technology | LPDDR5X |
| Memory bandwidth | Up to 1.2TB/s |
| Performance gain vs x86 | Up to 1.8x faster task completion |
| Target workloads | Agentic AI, reinforcement learning, data processing |
That bandwidth figure matters more than it might seem — agentic workloads often bottleneck on how fast a CPU can move data around while coordinating multiple simultaneous processes, and 1.2TB/s gives Vera considerably more headroom than typical server CPUs.
Scaling Toward Gigawatts With Vera Rubin
Beyond Vera itself, SpaceXAI is expanding its broader AI infrastructure for Grok on NVIDIA Vera Rubin, a platform that combines accelerated computing, NVLink interconnects, Spectrum-X Ethernet networking, and BlueField data processing into one integrated architecture. As SpaceXAI scales toward gigawatt-level computing capacity, Vera Rubin is designed to give it a common architecture that stays efficient at that scale while driving down cost per token. Full technical details on the platform are available via NVIDIA’s official Vera Rubin page.
The Genuinely Unusual Part: AI Computing in Orbit
Here’s where this partnership gets unusual. SpaceXAI plans to extend an optimized Vera Rubin NVL72 rack-scale system into space, powering its first-generation Starmind AI satellite. Orbital computing brings dramatically different constraints than a terrestrial data center — power delivery, thermal management, bandwidth, and physical reliability all behave differently once you’re off the ground — and NVIDIA says it’s working with SpaceXAI to adapt the same architecture and software ecosystem to meet those requirements while preserving compatibility with its Earth-based systems.
What the Companies Are Saying
NVIDIA’s VP of hyperscale and high-performance computing, Ian Buck, framed agentic AI as needing “a new kind of computing system — one built not only to generate answers, but to take action.” SpaceXAI president Mike Nicolls echoed that framing, noting Vera’s CPU performance and memory bandwidth let the company run heavy orchestration and data processing “while keeping GPUs doing what they do best.”
Why This Matters
This deal effectively gives NVIDIA one continuous architecture spanning three very different environments — Vera CPUs accelerating individual AI agents, Vera Rubin powering gigawatt-scale AI factories on Earth, and the same underlying technology extending into orbital computing. It’s a meaningful signal of how seriously both companies are betting on agentic AI as the next major infrastructure category. Keep following our AI infrastructure coverage section as this rollout progresses.
