Jensen Huang, chief govt officer of Nvidia Corp., speaks subsequent to a Vera Rubin Extremely Kyber compute tray and a Vera Rubin Extremely Kyber NVLink midplane throughout a keynote handle on the Nvidia GTC convention in San Jose, California, March 16, 2026.
Bloomberg | Bloomberg | Getty Photos
Nvidia grew to become probably the most useful firm due to insatiable demand for its graphics processing unit, or GPU, the first chip used for creating and deploying artificial intelligence.
However the chip big is now delivery its personal central processing items, or CPUs, which cloud suppliers may resolve to deploy rather than these from Advanced Micro Devices and Intel, opening up a brand new battleground in AI servers.
On Tuesday, Nvidia launched new details about its knowledge middle CPU, referred to as Vera, together with specs and the type of benchmarks and architectural info that potential prospects want to completely consider the chip. Nvidia representatives stated Vera chips had been delivered to purchasers, together with OpenAI, Anthropic, and SpaceX, in June.
Nvidia is seen as the corporate that units the path for the knowledge expertise business. However in CPUs, Nvidia is the challenger as soon as once more, competing towards two well-established gamers in Intel and AMD, which have deep ties to hyperscalers and cloud giants.
The corporate’s funding in CPUs is one other instance of Nvidia’s technique to vertically combine its methods and produce extra of the chips and expertise inside them yearly. It goals to promote all the system as a full rack of computing energy, as an alternative of merely promoting chips by themselves.
It is a technique that Nvidia says will assist engineers squeeze extra efficiency out of their GPUs, serving to the corporate’s methods stay the computer systems of selection for main AI labs as competitors from AMD and custom chips heats up.
Earlier than the AI increase, the CPU was crucial and costliest half in a server. The primary technology of AI servers obtainable when ChatGPT was launched in 2022 paired as many as eight GPUs to at least one CPU, signaling a shift in the direction of Nvidia’s GPUs.
However the rise of agentic AI, which may run independently within the background with minimal human enter, has returned consideration to the CPU, which is required to babysit and feed knowledge to an agent. Monetary markets have observed, and CPU incumbents AMD and Intel are among the many two best-performing chip shares up to now in 2026, up 128% and 149% respectively, besting Nvidia’s rise of 8%.
Nvidia, AMD and Intel inventory chart.
Brokers have made CPUs “way more integral,” Ian Buck, Nvidia’s vice chairman of hyperscale, stated at a presentation final week. “Significantly how briskly a CPU can reply one query.”
Nvidia stated the entire server CPU market may ultimately be value $200 billion, whereas a Bernstein estimate from earlier this yr stated the mature server CPU market in complete was value about $37 billion in 2025.
Wolfe Analysis stated in Could that it anticipated the common promoting worth to be about $5,000 per Vera chip, and forecast that Nvidia would ship about 1.3 million of them this yr. Nvidia declined to touch upon pricing.
Gartner analyst Kevin Knox stated that AMD is at present the corporate to beat in enterprise AI server CPUs. AMD reportedly holds about 33% of the marketplace for server CPUs, and Intel has 66.8%, however AMD is gaining share and has deep relationships with hyperscalers.
“AMD’s completed an incredible job constructing their ecosystem round their chips,” Knox stated.
Nvidia’s Vera chip configured on a motherboard.
Kif Leswing/CNBC
Single-core efficiency
Nvidia stated final week that Vera is the primary server CPU that it has designed from the core, versus utilizing an off-the-shelf design supplied by Arm, which requires much less engineering work.
The corporate stated Vera was particularly designed to deal with bottlenecks in AI brokers, and that its chip has 50% higher efficiency for AI brokers than x86 chips, the kind of silicon that Intel and AMD make.
That is as a result of earlier CPUs for clouds from Intel and AMD centered on core depend, however Nvidia’s chip and Olympus core focuses on single-core velocity as an alternative, Nvidia stated.
Hannah Coutand, a Vera product marketer at Nvidia, stated in a presentation that the chip centered on per-core velocity, excessive reminiscence bandwidth, and latency, “in order that brokers can return to their GPUs as shortly as potential and preserve these GPUs, that are a really costly and a extremely useful asset within the AI manufacturing facility, as extremely utilized as potential.”
The corporate stated that it’s going to promote the chip by itself, along with being paired with the corporate’s GPUs. It would additionally are available in a liquid-cooled rack of 256 Vera chips tied collectively, in addition to a configuration with two Vera chips in a single server. It would even be obtainable paired with Nvidia’s GPUs in a system called Vera Rubin.
The CPU is power-hungry, utilizing between 250 watts and 450 watts of energy. The chip additionally makes use of an enormous quantity of low-power reminiscence, the identical sort utilized in laptops and telephones, supporting as much as 1.5 terabytes of reminiscence per chip.
Some analysts have stated that Nvidia has created a brand new class of CPUs that Intel and AMD do not have a response to but. Cambrian AI Analysis founder Karl Freund stated Nvidia’s new CPU will not be used to serve web sites or different conventional server duties, however will likely be reserved for intense AI duties.
However getting the cloud suppliers to purchase the chips may nonetheless be a problem, even for Nvidia.
Coutand stated that Vera was in “early innings” of adoption. The corporate did not checklist any main cloud service suppliers besides Oracle on its checklist of companions however stated OpenAI plans to deploy Vera chips in massive portions beginning this quarter.
“The CPU is one thing they’ve completed to type of unhook their prospects from utilizing Intel or AMD CPUs, and so they covet that income,” stated Freund. “What they’ve completed is that they determined to concentrate on a singular CPU that is not obtainable available in the market from anybody proper now.”
