Nvidia VP of hyperscale and high-performance computing Ian Buck personally hand-delivered the first production Vera CPU systems to Anthropic, OpenAI, SpaceXAI and Oracle Cloud Infrastructure, moving the chip from announcement to revenue-bearing shipments. The first three AI labs received hardware Monday, May 18 at Anthropic's San Francisco office, OpenAI's Mission Bay headquarters and SpaceXAI's Palo Alto site; Oracle followed Wednesday, May 20 in Santa Clara. Oracle committed to deploying hundreds of thousands of Vera CPUs beginning in 2026, becoming the first cloud provider to lock in at that scale.
What's inside Vera
Vera is Nvidia's first fully custom CPU and the successor to Grace. The part ships with 88 custom Olympus cores and 1.2 TB/s of memory bandwidth, and Nvidia claims roughly 50% higher per-core performance under sustained load versus comparable existing designs. It is purpose-built for concurrent, agentic workloads — the orchestration layer, tool use, retrieval, code generation and reinforcement-learning pipelines that increasingly sit between the user and the GPU — rather than burst response generation.
"When AI models are posed a question, the models actually have to generate some Python code to arrive at the correct answer," Buck said during the rollout. "That's why we are seeing the demand for CPUs skyrocket."
Who took delivery
Anthropic's head of compute James Bradbury accepted the first system in San Francisco and framed the chip in operational terms: "Scaling compute is an important accelerant for the growth of models." OpenAI's head of compute infrastructure Sachin Katti received the hardware at Mission Bay. At SpaceXAI's Palo Alto office, Elon Musk inspected the system in person — asking about cores, memory layout and cooling — and reacted on X with "Vera nice, Vera nice …". SpaceXAI plans to evaluate Vera on reinforcement-learning workloads and simulation pipelines. Oracle product management lead Karan Batta and chief customer and partner success officer Gary Miller signed for the Santa Clara delivery.
The infrastructure thesis
Vera is Nvidia's clearest move yet to capture the CPU half of the agentic-AI rack, where x86 silicon from Intel and AMD has historically owned the orchestration plane. By pairing custom CPUs with Rubin GPUs and Nvidia networking, the company is selling whole-system economics into hyperscalers already stretching their AI capex budgets — Microsoft, Meta, Alphabet and Amazon have collectively guided to roughly $700-725 billion for 2026. Oracle's commitment to hundreds of thousands of Vera units sets a procurement reference point competitors will be pressed to match.
Nvidia shares closed at $222.32, down 1.33% on the day, ahead of the company's Wednesday earnings report — where analysts will look for evidence that the agentic-CPU pivot widens Nvidia's data-center moat rather than cannibalizing existing GPU spend.
