Why we're watching this: Nvidia moving into agentic CPUs is a direct challenge to Intel, AMD, and cloud-native chip programs at Amazon and Google. If Vera captures even a fraction of the $200B TAM Huang is claiming, it reshapes enterprise infrastructure buying decisions before the end of 2026.
Key Takeaways
- Nvidia’s Vera CPU is purpose-built for agentic AI, delivering 2x efficiency and 50% faster throughput than traditional rack-scale CPUs, per Nvidia
- $20 billion in standalone Vera CPU sales already recorded in 2026, according to Jensen Huang on Nvidia’s Q1 earnings call
- Huang claims Vera opens a brand new $200 billion TAM for Nvidia, a market the company has never previously addressed
- Nvidia posted $81.6 billion in Q1 revenue and forecasts $91 billion for Q2; Vera joins a product line already breaking records
Nvidia has launched Vera, a CPU it says is the world’s first processor purpose-built for agentic AI, and CEO Jensen Huang used this week’s earnings call to claim it opens a brand new $200 billion total addressable market the company has never competed in before.
The claim is large even by Huang’s standards, but it arrives alongside real numbers. Nvidia reported $81.6 billion in Q1 fiscal 2027 revenue and guided for $91 billion in Q2, continuing a run of record-breaking quarters. Huang said Nvidia has already recorded $20 billion in standalone Vera CPU sales this year.
The logic behind Vera is architectural. Traditional CPUs are designed around cores, built to run multiple application instances simultaneously. Vera, by contrast, is optimized for token throughput, processing AI agent tasks as fast as possible rather than managing multi-tenant workloads.
Huang’s argument is that as AI agents proliferate from millions to billions, each running its own tools, the demand for this class of CPU will be enormous.
“The world is going to have billions of agents, and those billions of agents will all use tools. And those tools are going to be like PCs, just like us humans using PCs today.” — Jensen Huang, Founder and CEO, Nvidia
Vera features 88 custom NVIDIA-designed Olympus cores, LPDDR5X memory delivering up to 1.2 TB/s of bandwidth (twice the bandwidth at half the power of general-purpose CPUs, per Nvidia), and the second-generation NVIDIA Scalable Coherency Fabric.
When paired with Nvidia’s Rubin GPU in the Vera Rubin NVL72 platform, it connects via NVLink-C2C at 1.8 TB/s of coherent bandwidth, seven times the bandwidth of PCIe Gen 6.
Early adopters include Meta, Alibaba Cloud, ByteDance, Oracle Cloud Infrastructure, CoreWeave, and Cloudflare, alongside system makers Dell, HPE, Lenovo, and Supermicro. Cursor, the AI-native coding tool, is also deploying Vera to improve throughput for its coding agents.
National laboratories including Los Alamos, Lawrence Berkeley, and the Texas Advanced Computing Center have announced plans to deploy Vera.
The skeptic case is straightforward. Amazon Web Services last month announced a major contract with Meta for its own homegrown AI CPUs, and AWS CEO Andy Jassy has been explicit that he believes Amazon can build AI chips, both GPUs and CPUs, as well as or better than Nvidia.
Google, Microsoft, and other hyperscalers are all running parallel chip programs. The $200B TAM Huang is projecting assumes Nvidia wins the agent CPU market before those programs mature.
Vera is in full production. Partner systems from Dell, HPE, Lenovo, Supermicro, and others will be available in the second half of 2026.
Vera is Nvidia’s clearest signal yet that it intends to own the entire AI compute stack, not just the GPU layer. Whether enterprise buyers standardize on Vera or treat it as one option among several will define whether this $200B claim ages as vision or as one of the rare Huang predictions that missed, for teams tracking AI infrastructure decisions via Relve, an AI tools intelligence platform.
