① NVIDIA has disclosed details of its data center CPU Vera. The chip was delivered to customers including OpenAI in June this year, optimized for single-core performance in AI agents, offering approximately 50% better performance on AI agent tasks compared to x86 architecture CPUs.
② NVIDIA's launch of Vera is a reflection of its vertical integration strategy. The company plans to sell complete rack-scale computing systems to customers in the future to enhance GPU utilization and solidify its leading position in the AI field.
Cailian Press, July 22 (Editor Zhao Hao) On Tuesday (July 21), NVIDIA disclosed more details about its data center CPU "Vera," including technical specifications, performance test results, and chip architecture information, to help potential customers comprehensively evaluate the product.
NVIDIA stated that the Vera chip has been delivered to customers including OpenAI, Anthropic, and SpaceX in June this year.
NVIDIA has long been viewed as a bellwether in the information technology industry but remains a challenger in the CPU arena, needing to compete against two established rivals—Intel and AMD.
These two companies have long dominated the server CPU market and have established deep cooperative relationships with hyperscale cloud computing providers. However, some cloud service providers may choose to deploy NVIDIA CPUs in the future, opening a new front.
Increasing its focus on CPUs is another manifestation of NVIDIA's vertical integration strategy. The company aims to independently develop more chips and technologies needed internally each year and ultimately sell complete rack-scale computing systems to customers, rather than selling chips individually.
NVIDIA believes this strategy can further unleash GPU performance, helping its AI systems remain the preferred platform for leading AI labs in the face of competition from AMD and various custom AI chips.
Focusing on Single-Core Performance
NVIDIA stated last week that Vera is the company's first server processor independently designed from the CPU core, rather than using off-the-shelf core designs provided by Arm. Consequently, the R&D investment is higher, but the degree of independent control is also stronger.
The company stated that Vera is specifically optimized for performance bottlenecks in AI agent operations. Compared to CPUs using the x86 architecture (i.e., the architecture used by Intel and AMD), it offers approximately 50% better performance on AI agent tasks.
NVIDIA pointed out that while cloud computing CPUs previously emphasized core count, the Olympus core used in Vera focuses more on single-core performance.
Hannah Coutand, head of product marketing for NVIDIA's Vera, stated that Vera prioritizes optimization of single-core speed, memory bandwidth, and latency performance, "enabling AI agents to hand tasks back to the GPU as quickly as possible, thereby maximizing the utilization of the GPU—the most expensive and critical asset in the AI factory."
NVIDIA stated that Vera can be sold as a standalone CPU or paired with the company's GPUs. Product forms include:
Standalone Vera CPU for sale;
Liquid-cooled rack-scale systems composed of 256 Vera chips;
Server configurations equipped with two Vera CPUs;
Vera Rubin systems combined with NVIDIA GPUs.
Vera has a relatively high power consumption, ranging from approximately 250 watts to 450 watts. It also supports up to 1.5TB of Low-Power Double Data Rate (LPDDR) memory, a type typically used in laptops and smartphones.
Market Views
Before the rise of the AI wave, the CPU was the most important and expensive component in a server.
When ChatGPT was launched in 2022, first-generation AI servers typically adopted an architecture of "one CPU paired with up to eight GPUs," marking a shift in computational focus towards NVIDIA GPUs.
However, with the rise of Agentic AI, which can operate autonomously with minimal human intervention, the importance of the CPU has increased again. The CPU is responsible for managing agent operations and continuously feeding them data, thus regaining the spotlight.
The capital market has already taken note of this trend. As of 2026, AMD and Intel have risen 128% and 149% respectively, becoming the two best-performing chip stocks this year, while NVIDIA has risen only 8% over the same period.
Ian Buck, Vice President of NVIDIA's Hyperscale Computing business, stated last week that agents have made the CPU "even more indispensable," especially the speed at which the CPU can answer individual requests, which is becoming increasingly important.
NVIDIA estimates that the total server CPU market size could reach $200 billion in the future. Bernstein estimated earlier this year that the mature server CPU market size in 2025 is approximately $37 billion.
Wolfe Research estimated in May this year that the average selling price of the Vera chip is about $5,000, with NVIDIA's shipments expected to be around 1.3 million units this year. NVIDIA declined to comment on product pricing.
Gartner analyst Kevin Knox stated that AMD remains the strongest contender in the enterprise AI server CPU market. According to reports, AMD currently holds about 33% of the server CPU market share, while Intel holds 66.8%, but AMD is continuously expanding its share and maintains close cooperation with hyperscale cloud computing customers.
Knox said: "AMD has done an excellent job building an ecosystem around its chips."
Some analysts believe that NVIDIA has essentially created a new CPU category for which AMD and Intel do not yet have corresponding products.
Karl Freund, founder of Cambrian AI Research, stated that Vera will not be used for traditional server tasks like website hosting but will be specifically targeted at high-intensity AI computing.
However, even for NVIDIA, fully adopting this CPU across large cloud service providers remains a challenge.
NVIDIA's Coutand stated that Vera is still in the early stages of market introduction. Apart from Oracle, the partner list does not yet include other major cloud service providers. However, OpenAI plans to begin large-scale deployment of Vera chips starting this quarter.
Freund stated: "Launching a CPU is essentially about freeing customers from dependence on Intel and AMD CPUs while capturing that revenue stream as well. What they have chosen to build is a unique CPU that is not currently offered by other players in the market."
Comments