2026-08-27 — views
NVIDIA Ships First Custom CPU Built for Agentic AI Agents
Read this because Builders must watch how new CPUs handle orchestration and tool-calling alongside GPUs in the AI factory.
NVIDIA delivers its first custom Vera CPU to AWS, Oracle, and leading AI labs as agentic AI infrastructure scales.
A New Era for CPU Infrastructure
NVIDIA has officially begun shipping its first custom-built CPU, named Vera, specifically designed for the age of agentic AI. This development marks a significant shift as AI models move from simply answering questions to acting on tasks. Ian Buck, NVIDIA Vice President of Hyperscale and HPC, hand-delivered the initial systems across the ecosystem. The delivery included AWS receiving its first Vera CPU server and Vera Rubin GPU in Seattle.
Previously, Buck delivered systems to Oracle Cloud Infrastructure and three leading AI labs: Anthropic, OpenAI, and SpaceXAI. Ian Buck stated that agentic AI is creating a new CPU moment because every agentic sandbox, tool call, and orchestration layer requires CPU work. He emphasized that Vera is purpose-built to keep this work moving at scale.
Technical Specifications and Performance
Vera represents a new class of CPU designed with the realities of concurrent, real-time tasks as its starting point. The processor packs 88 custom NVIDIA-designed Olympus cores and provides 1.2TB/s of memory bandwidth. It delivers up to 1.8x faster per-core performance on agentic AI workloads compared to traditional designs.
Under constant load, Vera ensures work completes more quickly, increasing the efficiency of the entire AI factory. The system is part of NVIDIA’s extreme codesign story alongside the Rubin GPU and BlueField-4 DPU. In configurations like the NVL72, Vera acts as the host processor for Rubin GPUs via second-generation NVIDIA NVLink-C2C.
Adoption Across the Ecosystem
AWS is the latest stop in Vera’s journey, joining Oracle Cloud Infrastructure which became the first cloud provider to deploy Vera at hyperscale. OCI plans to deploy hundreds of thousands of these CPUs beginning in 2026 because agentic AI demands sustained performance at massive scale.
SpaceXAI is evaluating Vera for reinforcement learning workloads and agent-based simulation pipelines. At OpenAI, Sachin Katti thanked Buck for the delivery, noting that scaling compute is an important accelerant for model growth. James Bradbury of Anthropic expressed excitement to see Vera emerge as a promising part of the ecosystem when solving for agentic workloads.
Practitioner note
Developers should monitor how Vera handles orchestration, control, and data movement needed to feed GPUs at 2x the energy efficiency of traditional infrastructure. Builders must ensure their agentic sandboxes and tool-calling layers leverage this new CPU architecture for sustained performance at massive scale.