There are many who believe that we could be in the agentic era, and NVIDIA has introduced a chip that is optimized for agentic use.
NVIDIA has announced the Vera CPU — its first processor purpose-built for AI agents — at GTC Taipei. Now in full production, the chip delivers 1.8x faster task completion compared with x86 CPUs across agentic AI, reinforcement learning, and data processing workloads. For a company that has become the most valuable in the world on the back of GPU dominance, the Vera CPU marks a deliberate pivot toward owning the full AI infrastructure stack — not just training, but agentic inference.
What Vera Is, and Why It Matters
Vera is powered by NVIDIA’s custom Olympus core architecture — 88 cores designed specifically for the CPU-heavy work that agentic AI demands: Python runtimes, sandboxed code execution, orchestration logic, and analytics pipelines. The chip features a LPDDR5X memory subsystem delivering up to 1.2TB/s of bandwidth, which allows agents to spend less time waiting on CPU-bound steps.
The shift matters because AI factories are increasingly measured in tokens per dollar, not cores per dollar. Vera is NVIDIA’s answer to that new economics. As Jensen Huang put it: “AI agents will be the largest users of computing. Vera is the first CPU designed for that future.”
Vera also integrates tightly into NVIDIA’s broader platform. It serves as the host CPU for NVIDIA Vera Rubin systems via second-generation NVLink-C2C, offering up to 1.8TB/s of coherent bandwidth between CPU and GPU. It also powers the NVIDIA Vera BlueField-4 STX AI storage platform. The result is a chip that can sit anywhere in the AI factory — from standalone CPU servers to tightly coupled accelerated systems.
Who Is Adopting It
The adoption list is significant. Global AI labs planning to deploy Vera include Anthropic, OpenAI, and SpaceXAI, alongside hyperscalers ByteDance, CoreWeave, and Oracle Cloud Infrastructure. Major server manufacturers — Dell Technologies, HPE, Lenovo, and Supermicro — will offer Vera in standalone configurations, marking the first standard CPU option beyond x86 at this scale.
Anthropic, whose agentic coding platform Claude Code has been growing rapidly, is evaluating Vera to scale CPU-intensive agentic workloads. “Scaling compute is an important accelerant for the growth of models,” said James Bradbury, head of compute at Anthropic. “We’re excited to see Vera emerge as a promising part of the ecosystem when solving for agentic workloads.”
Oracle Cloud Infrastructure is also moving quickly. “By deploying NVIDIA Vera CPUs, OCI will support high-throughput reasoning and data processing workloads across next-generation AI environments,” said Mahesh Thiagarajan, EVP of Oracle Cloud Infrastructure.
The NYSE is a notable non-AI adopter in the mix. Working with Redpanda and HPE, the exchange — which processes more than 1.1 trillion messages per day — plans to use Vera to scale capacity and optimize latency for its market infrastructure.
The Bigger Picture
The launch signals a broader architectural shift. As agentic AI moves from answering questions to taking actions — running code, using tools, evaluating results — the CPU has re-emerged as a bottleneck. GPU-centric infrastructure alone can’t support the orchestration and context-switching that agents require at scale.
Benchmarking from Phoronix across code compilation, Python, Java, and database processing found that Vera delivered the fastest overall performance across these agentic workloads. With agent sandboxes running 50% faster on Vera than traditional CPUs, Jensen Huang’s thesis — that the infrastructure layer is where AI economics are actually won — is being stress-tested in the market in real time.
Vera is available now in dense liquid-cooled racks for large-scale agentic environments and two-socket air-cooled systems for enterprise deployments. Whether this reshapes the CPU market the way NVIDIA reshaped the GPU market remains to be seen, but the intent is unambiguous.