# xAI Takes Vera CPUs for Agents — and Plans a Vera Rubin Rack in Orbit

Times of AI Desk · 2026-08-24 · Work

[https://timesof.ai/2026/08/spacexai-nvidia-vera-cpu-starmind](https://timesof.ai/2026/08/spacexai-nvidia-vera-cpu-starmind)

> Nvidia said xAI will deploy Vera CPUs — the 88-core Olympus CPU Nvidia bills as built for agents — to orchestrate Grok’s next agentic workloads, scale Grok infrastructure on Vera Rubin toward gigawatts, and base its first-generation Starmind AI satellite on an optimized Vera Rubin NVL72 rack. No matching x.ai/news post — Nvidia primary is the source. Starmind is planned, not a launch date.

The Grok stack is becoming a **full Nvidia factory**: Rubin for tokens, **Vera** for the agent runtime, **LPX** (separate SKU) for interactive decode — and now a **satellite** that is the same rack, thermally restated.

**NVIDIA** newsroom: **xAI** will deploy **NVIDIA Vera** CPUs to accelerate **orchestration, code execution, data processing, and simulation** around **Grok** agents, keeping GPUs fed. xAI is expanding Grok infrastructure on the **Vera Rubin** platform as it scales **toward gigawatts**. Same post: the planned first-generation **Starmind** AI satellite will be based on an **optimized Vera Rubin NVL72**. Quotes: **Ian Buck** (NVIDIA) and **Mike Nicolls** (xAI president). **No matching x.ai/news post** in the window — the NVIDIA primary is the source.

## What was announced

| Item | NVIDIA newsroom Aug 24 |
|------|------------------------|
| **CPU** | **Vera** — **88 Olympus** cores; **Spatial Multithreading**; **LPDDR5X** up to **1.2 TB/s** |
| **Claim** | Up to **1.8×** faster task completion vs **x86** on agentic / RL / data-processing mixes |
| **Customer use** | Agent loop **between** model calls — tools, code, data, sim |
| **Factory path** | Grok infra on **Vera Rubin**, **gigawatt-scale** |
| **Orbit** | **Starmind** gen-1 satellite on **optimized NVL72** — planned, not a launch date |

Vera launched as “the CPU for agents” at **GTC Taipei (May 31)** with xAI already on the **exploring** customer list. Today is a **named deployment + orbital SKU**, not a dual-date of the May unveil.

Not August 12 **Grok 4.6**, not August 19 **Build / Bedrock**, not August 21 **Bot seats / Google Agent Platform**, not same-day **Groq 3 LPX** (inference accelerator, Nebius). CPU vs LPU vs GPU are **three sockets**.

## Limits

- **No matching x.ai/news post** — Nvidia-only primary.
- Starmind is **planned**, not a launch window.
- **1.8×** vs x86 is Nvidia’s claim on agentic/RL mixes.
- Whether **Colossus-class** power numbers get attached to Vera Rubin rather than Blackwell is open.

## Sources

- [NVIDIA: xAI Adopts NVIDIA Vera CPU to Accelerate Agentic AI at Massive Scale (August 24, 2026)](https://nvidianews.nvidia.com/news/spacexai-adopts-nvidia-vera-cpu-to-accelerate-agentic-ai-at-massive-scale)
- [NVIDIA on X: SpaceX deploying Vera (August 24, 2026)](https://x.com/nvidia/status/2091920680317046847)
- Times of AI: `nvidia-groq-3-lpx-full-production` (August 24); `spacexai-grok-4-6` (August 12)