Nvidia, a company that achieved significant market value due to overwhelming demand for its Graphics Processing Unit (GPU)—the core component for developing artificial intelligence—has expanded its offerings. The chip giant is now introducing its own Central Processing Unit (CPU) designed specifically for AI tasks, named Vera. New information detailing the specifications and architectural benchmarks of this processor has been released for potential customers to thoroughly evaluate.
Market Shift: Entering the CPU Battleground
Nvidia’s move into the CPU market initiates a new competitive phase within AI servers, directly challenging established industry players like Advanced Micro Devices (AMD) and Intel. Previously, the GPU was considered the most critical component in AI infrastructure. However, the increasing complexity of “agentic AI”—systems capable of running independently with minimal human oversight—has redirected focus back to the CPU, which is necessary for managing and feeding data to these agents.
On Tuesday, Nvidia provided details regarding its Vera data center CPU, including technical specifications and performance metrics. Furthermore, representatives indicated that Vera chips had already been supplied to major clients, including OpenAI, Anthropic, and SpaceX, in June.
Nvidia’s Strategy for Vertical Integration
While Nvidia is known for setting industry direction, entering the CPU space positions it as a challenger against Intel and AMD, both of which possess deep connections with cloud computing giants. This investment exemplifies Nvidia’s strategy to vertically integrate its systems, allowing the company to manufacture more components and technology within its own ecosystem annually.
Instead of merely selling individual chips, Nvidia aims to sell entire computational racks. According to the company, this approach will help engineers maximize the performance derived from their GPUs, helping keep Nvidia’s system solutions favored by leading AI research laboratories amidst rising competition from both custom silicon and AMD.
Technical Specifications and Performance Claims
Nvidia stated that Vera is unique because it is the first server CPU designed entirely in-house, rather than utilizing an off-the-shelf design provided by Arm, which typically requires less engineering effort. The company claims Vera was specifically engineered to resolve bottlenecks inherent in AI agents, asserting that its chip offers 50% better performance for such agents compared to traditional x86 chips used by Intel and AMD.
Nvidia explained that while previous CPU designs from competitors emphasized core count, the Vera chip and its Olympus core are optimized for single-core speed. Hannah Coutand, a Vera product marketer at Nvidia, noted during a presentation that “the chip focused on per-core speed, high memory bandwidth, and latency, so that agents can return to their GPUs as quickly as possible and keep those GPUs, which are a very expensive and a highly valuable asset in the AI factory, as highly utilized as possible.”
The CPU is power-intensive, requiring between 250 watts and 450 watts of electricity. It also incorporates substantial low-power memory—similar to that used in phones and laptops—supporting up to 1.5 terabytes of memory per chip.
Market Analysis and Industry Outlook
Industry analysts suggest Nvidia has introduced a novel class of CPUs that neither Intel nor AMD currently have a direct countermeasure for. Karl Freund, founder of Cambrian AI Research, stated that Nvidia’s new processor is intended exclusively for intensive AI workloads, not traditional tasks like serving websites.
In terms of market size, Nvidia projects the entire server CPU market could eventually reach $200 billion in value, significantly surpassing a prior Bernstein estimate which valued the mature server CPU market at approximately $37 billion in 2025. Wolfe Research forecasted in May that the average selling price for a Vera chip would be around $5,000 and projected Nvidia to ship about 1.3 million units this year.
Regarding existing market share, Gartner analyst Kevin Knox noted that AMD reportedly holds roughly 33% of the server CPU market, while Intel accounts for 66.8%. However, Knox also observed that AMD has been successfully increasing its market presence and maintaining strong ties with hyperscale cloud providers.
Nvidia confirmed that Vera will be sold independently as well as when paired with the company’s GPUs. It will be available in a liquid-cooled rack of 256 Vera chips, or configured with two Vera chips within a single server unit, and also through a complete system known as Vera Rubin.
Despite its advanced capabilities, Nvidia acknowledged that adoption is still early. While the company listed Oracle among its partners, it also disclosed that OpenAI intends to deploy substantial quantities of Vera chips starting in the current quarter.
