Nvidia Corp. Chief Executive Officer Jensen Huang speaks next to the Vera Rubin Ultra Kyber compute tray and Vera Rubin Ultra Kyber NVLink midplane during the keynote address at the Nvidia GTC conference on March 16, 2026 in San Jose, California.
Bloomberg | Bloomberg | Getty Images
Nvidia has become the most valuable company due to insatiable demand for graphics processing units (GPUs), the primary chips used to create and deploy artificial intelligence.
But semiconductor giants now ship their own central processing units (CPUs), and cloud providers could decide to deploy them instead. advanced micro device and intelopening up a new battlefield for AI servers.
On Tuesday, Nvidia announced new information about a data center CPU called Vera. This includes specifications, benchmark types, and architectural information that future customers will need to fully evaluate the chip. Nvidia representatives said Vera chips have been delivered to the following customers: OpenAI, humanand space xin June.
Nvidia is considered a company that sets the direction for the information technology industry. But in CPUs, Nvidia is once again a challenger, competing with Intel and AMD, two established players with deep ties to hyperscalers and cloud giants.
The company’s investment in CPUs is another example of Nvidia’s strategy to vertically integrate its systems and produce more of the chips and technology inside them each year. Rather than just selling individual chips, the company aims to sell entire systems as full racks of computing power.
Nvidia says this will help engineers squeeze more performance out of the GPU, helping its systems remain the computers of choice for major AI labs as competition from AMD and custom chips increases.
Before the AI boom, the CPU was the most important and most expensive component in a server. The first generation of AI servers, available when ChatGPT was released in 2022, combined up to eight GPUs on a single CPU, marking Nvidia’s move to GPUs.
But the rise of agent AI, which can run independently in the background with minimal human input, has shifted attention back to the CPUs needed to babysit agents and feed them data. Financial markets are also taking notice, with CPU incumbents AMD and Intel becoming one of the two best-performing chip stocks of all time through 2026, up 128% and 149%, respectively, outpacing Nvidia’s 8% rise.
Nvidia, AMD, Intel stock charts.
Ian Buck, Nvidia’s vice president of hyperscale, said in a presentation last week that the agency has made CPUs “more essential.” “Specifically, how quickly can the CPU answer one question?”
Nvidia said the overall server CPU market could eventually reach a value of $200 billion, while Bernstein estimated earlier this year that the mature server CPU market could be worth a total of about $37 billion in 2025.
Wolfe Research said in May that it expects the average selling price per Vera chip to be about $5,000, and that it expects NVIDIA to ship about 1.3 million chips this year. Nvidia declined to comment on pricing.
Gartner analyst Kevin Knox said AMD is the current winner in enterprise AI server CPUs. It is reported that AMD has a market share of approximately 33% and Intel 66.8% for server CPUs, but AMD is increasing its share and has a deep relationship with hyperscalers.
“AMD has done a great job of building an ecosystem around their chips,” Knox said.
Nvidia’s Vera chip configured on the motherboard.
Kif Leswing/CNBC
Single core performance
Nvidia announced last week that Vera is the first server CPU designed from the core up, rather than using an off-the-shelf design provided by Arm, requiring less engineering work.
The company said Vera is specifically designed to address AI agent bottlenecks, and that its chips have 50% better AI agent performance than x86 chips, the type of silicon made by Intel and AMD.
Nvidia said this is because previous cloud CPUs from Intel and AMD focused on core count, while Nvidia’s chips and Olympus cores focus instead on single-core speed.
Hannah Coutand, Nvidia’s Vera product marketer, said in a presentation that the chip focuses on per-core speed, high memory bandwidth, and latency to “enable agents to get back to the GPU as quickly as possible and leverage the GPU, which is a very expensive and very valuable asset in the AI factory, to the highest possible degree.”
The company said it plans to sell the chip on its own as well as in combination with its GPUs. It is also available in water-cooled racks with 256 Vera chips, or in configurations with two Vera chips per server. It can also be used in conjunction with Nvidia’s GPUs in a system called Vera Rubin.
CPUs are power hungry, drawing between 250 and 450 watts of power. The chip uses large amounts of low-power memory, the same kind used in laptops and cell phones, and supports up to 1.5 terabytes of memory per chip.
Some analysts say Nvidia has developed a new class of CPUs that Intel and AMD have not yet covered. Karl Freund, founder of Cambrian AI Research, said Nvidia’s new CPUs will not be used for websites or other traditional server tasks, but will be reserved for intense AI tasks.
But getting cloud providers to buy chips could still be a challenge for Nvidia.
Koutando said Bella is in the “early stages” of adoption. The company did not list any major cloud service providers except for: oracle While on the list of partners, OpenAI said it plans to deploy Vera chips in large quantities starting this quarter.
“The CPU is what they did to free customers from using Intel or AMD CPUs, and they desperately want that revenue,” Freund said. “What they’ve done is they’ve decided to focus on unique CPUs that aren’t currently available on the market anywhere.”
