SpaceXAI Puts NVIDIA's Vera CPU to Work — and a Vera Rubin NVL72 Is Heading to Orbit

In a week where most silicon headlines were about accelerators, NVIDIA dropped a CPU story with an orbital twist. On August 24, 2026, alongside its Hot Chips 2026 news, NVIDIA announced that SpaceXAI will deploy NVIDIA Vera CPUs — billed as the first CPU built for AI agents — to accelerate its next generation of agentic AI applications. The same announcement confirmed two bigger structural moves: SpaceXAI is expanding the AI infrastructure behind Grok on the NVIDIA Vera Rubin platform as it scales toward gigawatts of computing capacity, and it plans to extend that architecture into space, with a first-generation Starmind AI satellite based on an optimized NVIDIA Vera Rubin NVL72 rack-scale system.

Why a CPU story matters in the accelerator era

The framing is easy to miss but worth unpacking. Agentic AI applications don’t just run model inference — between model calls, agents orchestrate tools, execute code, process data, and run simulations. That surrounding work lands heavily on general-purpose CPUs, and if the CPU can’t keep up, the whole agent loop stalls and expensive GPUs sit idle waiting to be fed. NVIDIA’s pitch is that Vera attacks exactly this bottleneck.

“Agentic AI requires a new kind of computing system — one built not only to generate answers, but to take action,” said Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA. “Vera gives AI agents the CPU performance to act in real time — executing code, processing data and coordinating complex tasks. SpaceXAI is taking this architecture from massive AI factories to the next frontier of computing in orbit.”

Mike Nicolls, president of SpaceXAI, put the practical case bluntly: “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. That means higher-performance AI agents and more useful work from every watt of compute.”

What’s inside Vera

NVIDIA positions Vera as the first CPU purpose-built for the CPU-intensive work that surrounds model inference. The headline specifications:

• 88 NVIDIA-designed Olympus cores with NVIDIA Spatial Multithreading technology
• High-bandwidth LPDDR5X memory delivering up to 1.2TB/s of bandwidth
• Up to 1.8x faster task completion compared with x86 CPUs across agentic AI, reinforcement learning, and data processing workloads

Those figures track what NVIDIA showed when it first detailed the platform: 176 threads, 1.8TB/s of NVLink-C2C bandwidth, 1.5TB of system memory, and 227 billion transistors. As StorageReview noted, the 1.8x claim is a vendor figure that independent testing has yet to validate — but even partial real-world confirmation would make Vera a serious alternative to x86 in agent-heavy fleets.

SpaceXAI will use Vera specifically to handle the orchestration, code execution, data processing, and simulation tasks that surround AI model inference — helping agents act faster while keeping GPUs fully utilized.

Scaling Grok on Vera Rubin

The second layer of the announcement is infrastructure. SpaceXAI plans to build out the compute behind Grok on NVIDIA Vera Rubin, NVIDIA’s codesigned platform that spans compute, networking, and software rather than shipping as discrete parts. The stack brings together NVIDIA accelerated computing, NVLink interconnect technology, Spectrum-X Ethernet networking, BlueField data processing, and NVIDIA software in one integrated architecture, with the stated goal of maximizing performance, power efficiency, and utilization at scale — and driving down cost per token.

As SpaceXAI expands toward gigawatts of computing capacity, Vera Rubin gives it a common architecture to scale its next generation of AI factories. That “common architecture” phrase is doing real work here, because it’s also the bridge to the third and most speculative part of the announcement.

From AI factories to orbit

SpaceXAI is developing AI computing infrastructure for orbit, where power, thermal management, bandwidth, reliability, and physical integration impose constraints that look nothing like a terrestrial data center. The planned first-generation Starmind AI satellite will be based on an optimized NVIDIA Vera Rubin NVL72 rack-scale system — extending the same accelerated computing architecture powering next-generation AI factories on Earth into space.

This isn’t NVIDIA’s first step toward orbital compute. At GTC 2026, the company showed a Space-1 Vera Rubin module aimed at orbital deployment. Nor is it happening in a vacuum: Starcloud raised $250 million at a $2.3 billion valuation just days earlier to build its own orbital data centers, and SpaceX itself is reportedly developing an AI satellite called AI1 with a 230-foot wingspan. The difference here is the software story — NVIDIA and SpaceXAI say they are adapting the Vera Rubin foundation to orbital requirements while preserving a common architecture and software ecosystem, so the same stack runs in the data center and in the sky.

Worth noting: no timeline, launch date, or capacity figure accompanies the Starmind plan, and NVIDIA files the entire payload under forward-looking statements. Radiation hardening, launch mass, thermal rejection in vacuum, and inter-satellite link bandwidth all remain open engineering questions that today’s press release doesn’t answer.

The takeaway

Strip away the space spectacle and this is a coherent land-and-expand play. One computing foundation now spans three environments: Vera CPUs accelerating increasingly sophisticated AI agents, Vera Rubin powering the infrastructure behind Grok and gigawatt-scale AI factories on Earth, and NVIDIA accelerated computing extending into orbital AI infrastructure via Starmind.

For the broader industry, the signal is that the agent era is reshaping demand well beyond GPUs. If agents really do spend most of their wall-clock time on CPU-bound orchestration, then CPU architecture becomes a competitive lever in AI deployments — and NVIDIA, which spent two decades as the accelerator company, is now making a direct run at x86’s home turf with SpaceXAI as its flagship customer. The version of this story worth watching is whether Vera’s 1.8x claim survives independent benchmarks, and whether a Vera Rubin NVL72 actually survives launch, orbit, and operations. Both would be firsts.