Nvidia’s Vera CPU pushes it deeper into AI servers

Nvidia is giving investors a clearer look at Vera, its new central processing unit for artificial intelligence servers, and the move puts it in more direct competition with AMD and Intel. For retail investors watching chip stocks, the key point is that Nvidia is trying to sell more than the graphics chips that made it the market’s AI leader.

The company released specifications, benchmark information and architecture details for Vera on Tuesday, saying the material gives potential customers what they need to assess the chip. Nvidia representatives said Vera chips were delivered in June to customers including OpenAI, Anthropic and SpaceX.

Nvidia became the world’s most valuable company on demand for its graphics processing units, or GPUs, the chips used to train and run many AI models. A central processing unit, or CPU, handles general computing tasks and helps coordinate data movement inside a server. In AI systems, that means the CPU can affect how quickly expensive GPUs are kept fed with work.

Why Nvidia wants the CPU slot

Nvidia’s broader strategy is to control more of the AI server, rather than sell only individual chips. The company wants to offer full racks of computing power that combine its own processors, networking and system design.

That approach could help Nvidia improve performance in systems built around its GPUs, according to the company. It also gives Nvidia a shot at revenue that has traditionally gone to Intel and AMD, which have long relationships with major cloud providers.

The CPU has gained fresh attention as agentic AI grows. Agentic AI refers to software that can run tasks with less direct human prompting, often staying active in the background. Ian Buck, Nvidia’s vice president of hyperscale, said at a presentation last week that agents have made CPUs “much more integral,” especially in how fast a CPU can answer one question.

Nvidia said the server CPU market could eventually reach $200 billion. Bernstein estimated earlier this year that the mature server CPU market was worth about $37 billion in 2025.

AMD and Intel still own the field

Nvidia is entering a market where AMD and Intel are entrenched. Gartner analyst Kevin Knox said AMD is currently the company to beat in enterprise AI server CPUs. AMD reportedly holds about 33% of the server CPU market, while Intel has 66.8%, according to figures cited by CNBC, though AMD has been gaining share.

“AMD’s done a great job building their ecosystem around their chips,” Knox said.

Chip stocks show how investors have been treating the CPU story this year. CNBC reported that AMD and Intel were among the two best-performing chip stocks so far in 2026, rising 128% and 149%, respectively, compared with an 8% gain for Nvidia.

What Vera is built to do

Nvidia said Vera is its first server CPU designed from the core, rather than built around an off-the-shelf Arm design. The company said Vera targets AI-agent bottlenecks and delivers 50% better AI-agent performance than x86 chips, the processor architecture used by Intel and AMD.

Nvidia said Vera emphasizes single-core speed, memory bandwidth and latency, which is the delay before data moves or a task begins. Hannah Coutand, a Vera product marketer at Nvidia, said the goal is to return agents to GPUs quickly and keep those GPUs highly used.

Vera will be sold as a standalone chip and also paired with Nvidia GPUs in a system called Vera Rubin. Nvidia said it will offer configurations including a liquid-cooled rack with 256 Vera chips and a two-Vera server design.

The chip uses between 250 watts and 450 watts of power and supports as much as 1.5 terabytes of low-power memory per chip, according to Nvidia. Wolfe Research said in May that it expected an average selling price of about $5,000 per Vera chip and forecast 1.3 million shipments this year. Nvidia declined to comment on pricing.

Adoption is still early, according to Coutand. Nvidia named Oracle among cloud partners and said OpenAI plans to deploy Vera chips in large quantities starting this quarter. Cambrian AI Research founder Karl Freund said the chip is aimed at heavy AI workloads rather than traditional server jobs such as serving websites.

This story draws on original reporting from CNBC.