NVIDIA Vera CPU: SpaceX Deploys World’s First Agentic AI Processor at Gigawatt Scale

NVIDIA has announced that SpaceX is deploying its Vera CPU platform to accelerate the orchestration, code execution and data processing that underpin large-scale agentic AI systems. The move marks one of the most ambitious early deployments of a processor designed not for general-purpose computing, but specifically to keep AI agents running fast and fed with work.

The announcement came via NVIDIA’s official social media account, which posted details of SpaceX’s adoption of Vera as an example of agentic AI operating at what the company calls “gigawatt AI factories” — a phrase that gives some sense of the sheer scale of power these systems now consume.

What Is the Vera CPU — and Why Does It Matter?

Vera is built around 88 custom-designed Olympus cores, compatible with the Armv9.2 architecture. NVIDIA has engineered it from the ground up for agentic AI workloads: the kind of AI that doesn’t just answer questions but plans, uses tools, writes and runs code, and carries out multi-step tasks with minimal human input.

That’s a different job from what traditional server CPUs do. In most AI data centres, the GPU does the heavy lifting — training models, running inference — while the CPU handles supporting tasks. But agentic AI changes that equation. When an AI agent is constantly calling tools, scheduling tasks, moving data and managing dozens of concurrent workflows, the CPU can become the bottleneck. Vera is NVIDIA’s answer to that problem.

NVIDIA claims Vera delivers around 50% faster performance and roughly twice the energy efficiency of traditional rack-scale CPUs for agentic AI workloads. The company also cites figures of up to ten times the agent throughput and ten times more tokens per megawatt compared with previous generations. These are vendor-stated benchmarks and have not yet been independently verified, so they should be read with that caveat in mind.

SpaceX and the Race to Build AI Factories

SpaceX’s AI infrastructure — linked in industry coverage to systems including Grok — is reported to be scaling towards gigawatt-level computing capacity. That makes it one of the most demanding environments on the planet for AI hardware, and an early proving ground for Vera.

The goal, as NVIDIA frames it, is to ensure that GPU clusters are never sitting idle. Vera CPUs handle the orchestration layer — spinning up agent environments, routing data, managing tool calls — so that the GPUs are constantly supplied with work. A Vera rack can hold up to 256 liquid-cooled processors and is designed to support tens of thousands of concurrent CPU environments simultaneously.

SpaceX is not alone. Anthropic and OpenAI are also listed among early adopters of the Vera Rubin platform — NVIDIA’s broader Rubin-generation AI stack, which combines Vera CPUs with GPU infrastructure, networking and storage into an end-to-end system for large AI factories.

Jensen Huang, NVIDIA’s chief executive, said: “The age of agentic AI is here, and Vera is the CPU the world’s AI factories have been waiting for.”

A Platform Built for 2026 and Beyond

Vera-based systems are not yet on general sale. NVIDIA has confirmed that full commercial availability is expected in the second half of 2026, with pre-production and evaluation hardware already in the hands of select partners including SpaceX’s AI operations.

Server manufacturers including Dell Technologies, Hewlett Packard Enterprise, Lenovo and Supermicro are all expected to offer Vera-based systems, as are major cloud providers. That means enterprises and developers worldwide — not just the handful of AI labs with direct access today — will be able to run agentic AI workloads on this infrastructure once it reaches the market.

But the scale of these deployments is drawing scrutiny. Critics point to the environmental cost of gigawatt-scale AI factories: the electricity demand, the carbon footprint, and the water consumed by liquid-cooling systems like those used in Vera racks. There are also wider questions about what more powerful agentic AI means for workers in roles that involve scheduling, administration, coding and customer service — tasks that agentic systems are specifically designed to handle.

Some researchers and commentators also raise concerns about concentrating this level of AI computing power in a small number of global companies, and about the risks of deploying increasingly autonomous AI agents at scale before safety and governance frameworks have fully caught up.

The performance figures NVIDIA is promoting — the ten-times throughput claims, the efficiency gains — are, for now, the company’s own. Independent benchmarks will matter when they arrive.

What This Means for Kent Residents

No Vera CPUs are being deployed in Kent, and none are expected to be. But as cloud providers integrate NVIDIA’s Vera Rubin infrastructure from late 2026, organisations across the county — including Kent County Council, NHS Kent and Medway ICB, local businesses and universities — may find that the AI tools they access through national and global cloud services become faster and more capable. For workers in logistics, administration, customer service and public services, the spread of more powerful agentic AI through mainstream software platforms is worth watching, as it will shape how those tools are used — and what skills employers look for.