Vera is here. NVIDIA announced that its first custom CPU, built from the ground up for agentic AI, is shipping — and that Vice President of Hyperscale and HPC Ian Buck personally hand-delivered the opening systems to three of the world’s leading AI labs.
The Deliveries
NVIDIA Vice President of Hyperscale and HPC Ian Buck hand-delivered Vera CPU systems across the AI ecosystem on Friday: Anthropic in San Francisco, OpenAI at Mission Bay, and SpaceXAI in Palo Alto, followed by a delivery to Oracle Cloud Infrastructure in Santa Clara the following Monday.
- Anthropic’s head of compute, James Bradbury, took the handoff from a conference room near the Bay
- At OpenAI, head of compute infrastructure Sachin Katti got a walkthrough of the server on an open-air balcony
- At SpaceXAI, Elon Musk asked detailed questions about cores, memory layout, and cooling during the handoff
Why a CPU for Agents
The pitch is simple: AI agents do not run on GPUs alone. Every agentic sandbox, every tool call, every orchestration layer, and every long-context retrieval operation is CPU work. A gauntlet of concurrent, real-time tasks puts pressure on CPUs in ways traditional core-density-focused designs were never built to prioritize.
“Agentic AI is creating a new CPU moment in the AI factory — as models move from answering to acting, Vera is purpose-built to keep that work moving at scale,” Buck said.
Vera at a Glance
The specs underline the new class:
- 88 custom NVIDIA-designed Olympus cores
- 1.2 TB/s of memory bandwidth
- Up to 1.8x faster per-core performance on agentic AI workloads
- Handles orchestration, tool-calling, RL workloads, data analytics, agent sandboxing, and long-context state management
- Host processor for the Vera Rubin NVL72, paired with Rubin GPUs via second-generation NVLink-C2C
What Vera’s First Customers Said
SpaceXAI is evaluating Vera for reinforcement learning workloads and the agent-based simulation pipelines behind its training stack. OCI — the first cloud to deploy Vera at hyperscale — plans to deploy hundreds of thousands of Vera CPUs beginning in 2026. “Vera’s architecture is purpose-built for high-throughput reasoning workloads,” said OCI’s Karan Batta.
What This Means
Vera marks NVIDIA’s entry as an x86 and Arm competing CPU designer and its bet that agentic AI — not just training — is the next infrastructure battleground. For the AI labs now running it, the question is whether the CPU that keeps up with the AI factory becomes the default host for the age of agents.