Nvidia Enters CPU Battlefield: Vera Chips Delivered to OpenAI, Performance Exceeds x86 Chips by 50%, Challenges AMD and Intel

Nvidia has released the specifications, benchmarks, and architectural details of its Vera data center CPU.

Nvidia Vice President and General Manager Ian Buck provided the most direct confirmation at a headquarters briefing: "We are absolutely in full production, and it is being deployed across all major customers." Computing systems based on Vera Rubin technology have been delivered to major AI enterprises, with CoreWeave, Google Cloud, Microsoft Azure, Meta, and Dell among those on the adoption list.

Vera is Nvidia's first server CPU designed from the core up, rather than using an off-the-shelf Arm design.

On the other hand, while Intel and AMD's cloud CPUs focus on core counts, Nvidia's Olympus core focuses heavily on single-core frequency. Nvidia claims that Vera is designed to address AI agent bottlenecks, delivering 50% higher performance in AI agent tasks compared to the x86 chips used by Intel and AMD.

Hannah Coutand, head of Vera product marketing, explained the underlying logic in a demonstration: the chip focuses on per-core speed, high memory bandwidth, and low latency, "allowing agents to return to the GPU as quickly as possible, keeping these extremely expensive, extremely high-value assets in the AI factory at the highest possible utilization."

In terms of product form, Vera offers highly flexible distribution options: it can be sold individually, as dual-Vera single-server configurations, in liquid-cooled racks of 256 Vera chips, or as Vera Rubin systems paired with GPUs. Regarding specifications, the CPU's power consumption ranges between 250 and 450 watts, with a single chip supporting up to 1.5TB of memory—utilizing the same type of low-power memory found in laptops and smartphones, which is used in massive quantities.

Competition in the existing market. Gartner analyst Kevin Knox pointed out that AMD is currently "the one to beat" in the enterprise AI server CPU space—it holds about a 33% share in the server CPU market, while Intel holds 66.8%, but AMD is gaining share and has deep relationships with hyperscale cloud providers. "AMD has done an outstanding job of building an ecosystem around its chips."

Are cloud providers buying in? Nvidia admitted that the adoption of Vera is still in "early stages," and no other major cloud providers are on the partner list besides Oracle. Persuading cloud giants to abandon the mature Intel/AMD ecosystem is no easy task, even for Nvidia. Moreover, AMD and Broadcom are looking to grab market share, and major customers like Amazon are developing their own chips to reduce their reliance on Nvidia.

Expectation gap in market size. Nvidia claims the entire server CPU market could eventually reach $200 billion, while Bernstein's estimates earlier this year showed that the total size of the mature server CPU market in 2025 will be only about $37 billion. This massive gap implies that Nvidia is not betting on replacing the installed base, but rather on AI agents expanding the market itself multiple times over.

This content was translated using AI and reviewed for clarity. It is for informational purposes only.