Nvidia’s Vera CPU Debuts Into a Market Still Owned by AMD and Intel

Nvidia published full specifications for its first custom built server chip on Tuesday, a processor called Vera that it says beats Intel and AMD by 50% on AI agent workloads. The chip is already running inside servers at OpenAI, Anthropic and SpaceX, after Nvidia representatives said the hardware shipped to those customers in June.

AMD and Intel are the two companies Vera is built to dethrone. Both are having their best stock years in over a decade. Nvidia’s own shares are barely moving by comparison, and that gap sits at the center of the Vera story.

A Chip Built From Scratch for AI Agents

Nvidia’s data center processor carries 88 custom Olympus cores and 176 processing threads on a single die, with 164 megabytes of shared cache, according to architectural details the company released this week. It draws between 250 and 450 watts and can address up to 1.5 terabytes of low power memory per chip, the same LPDDR5X family used in laptops and phones rather than the DDR5 memory found in most servers.

That memory choice matters. Nvidia claims roughly three times the memory bandwidth per core and up to five times the bandwidth per watt of a standard DDR5 server platform, per architectural disclosures reported this week.

Nvidia also shared workload benchmarks pulled from real customers instead of standard lab tests. The company said Vera completed jobs about 1.5 times faster than x86 chips in a Perplexity sandboxing environment, cut streaming latency as much as sixfold in New York Stock Exchange market data processing, and ran select agentic simulations at Los Alamos National Laboratory up to seven times faster.

Nvidia designed the Olympus core to maximize instructions per cycle for concurrent AI workloads, according to the company’s own technical blog. Conventional cloud CPUs from Intel and AMD have chased higher core counts for years. Nvidia went the other way, betting that single core speed and low latency matter more once an agent needs an answer fast enough to keep an expensive GPU busy.

Hannah Coutand, a Vera product marketer at Nvidia, said the chip was built around per core speed, 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.”

Why Do AI Agents Need a Faster CPU?

Agentic AI, software that runs in the background with little human input, leans harder on the CPU than earlier chatbot style tools ever did. Each agent still needs a central processor to fetch data, manage memory and hand instructions to the GPU doing the heavy computation, and a slow CPU leaves that GPU sitting idle.

The first AI servers built after ChatGPT’s 2022 debut paired as many as eight GPUs to a single CPU, a ratio that reflected how little the processor mattered once the GPU became the star of the system. That has since shifted.

Ian Buck, Nvidia’s vice president of hyperscale, said agents have made the CPU “much more integral,” especially “how fast a CPU can answer one question,” speaking at a presentation last week.

The Stock Market Still Bets on AMD and Intel

Investors have not treated Vera as an existential threat to the incumbents. AMD shares are up 128% so far in 2026. Intel is up 149%. Nvidia, the company launching the chip built to unseat them both, has gained just 8%.

The muted reaction extended to launch day itself. Nvidia shares were up about 1% at $205.43 within hours of Tuesday’s disclosure, a subdued response for a stock that has moved sharply on smaller news, according to TradingKey’s market coverage.

Compare that with June 1, when Nvidia unveiled RTX Spark, an Arm based laptop chip built with MediaTek and Microsoft. Intel fell 6% and AMD fell 5% that day, while Nvidia rose about 4%, according to a Forbes analysis. Vera, arguably the bigger threat to both companies’ core data center business, barely moved either stock seven weeks later.

Two Incumbents Still Hold Nearly All the Business

Vera is entering a market Intel and AMD have split almost entirely between themselves. Mercury Research, the industry’s benchmark tracker of processor shipments, measured AMD’s server CPU unit share at 33.2% in the first quarter of 2026, with Intel holding 66.8%.

Here is how the three chips compare heading into Vera’s commercial push:

Kevin Knox, a Gartner analyst who covers data center hardware, said AMD is currently the company to beat in enterprise AI server processors, pointing to relationships with hyperscalers built over a decade. “AMD’s done a great job building their ecosystem around their chips,” Knox said.

Nvidia’s Biggest Customers Are Hedging Their Bets

Nvidia’s own list of Vera partners is short. Oracle is the only major cloud provider the company names so far, and Coutand described Vera as being in “early innings” of adoption.

Nvidia counts a handful of marquee names already running the chip:

• OpenAI – received Vera hardware in June and plans large scale deployment starting this quarter
• Anthropic – among the first companies to get Vera chips in June
• SpaceX – also confirmed as an early June recipient of the hardware
• Oracle – the only cloud service provider Nvidia names as a partner so far

Even as Nvidia tries to pull hyperscalers away from Intel and AMD, some of those same customers are building their own chips to cut reliance on Nvidia too. Amazon continues developing in house processors for its cloud business, a dynamic flagged in TradingKey’s coverage of the Vera launch, reducing how much any single hyperscaler depends on one outside supplier for either GPUs or CPUs.

Nobody Agrees on How Big This Market Is

Analysts cannot even agree on the size of the prize Nvidia is chasing. Wolfe Research estimated in May that Nvidia would charge around $5,000 per Vera chip and ship about 1.3 million of them this year. Multiply those two figures together and Vera alone could generate somewhere close to $6.5 billion in chip sales in 2026, though Nvidia declined to comment on pricing.

Where the estimates sharply diverge is on how large the category becomes over time.

• Nvidia says the total server CPU opportunity could eventually reach $200 billion
• Bernstein estimated the entire mature server CPU market at about $37 billion in 2025
• GF Securities, a China based brokerage, projects the category hitting $211 billion by 2030 as agentic AI inference scales

Those three numbers describe roughly the same market, yet span a range of more than $170 billion.

Nvidia has a clear commercial reason for building its own CPU.

The CPU is something they’ve done to kind of unhook their customers from using Intel or AMD CPUs, and they covet that revenue.

Karl Freund, founder of the semiconductor analysis firm Cambrian AI Research, said Nvidia built a category no rival currently matches. “What they’ve done is they decided to focus on a unique CPU that isn’t available in the market from anyone right now,” he said.

Vera is not Nvidia’s only run at Intel and AMD’s core business. The company’s N1X chip aimed at the laptop market applies a similar playbook to Windows PCs, pairing an Arm based Nvidia processor with its own GPU instead of buying either part from an outside supplier.

Vera still has to win the data center one contract at a time, against two rivals that just delivered the best stock returns of their careers.

Frequently Asked Questions

Is Nvidia’s Vera CPU Built on Arm or x86 Architecture?

Vera uses the Armv9.2 instruction set with a fully custom Olympus core that Nvidia designed itself, rather than licensing an off the shelf Arm Neoverse design or building on x86 like Intel’s Xeon and AMD’s EPYC chips.

How Does Vera’s Olympus Core Compare to Nvidia’s Earlier Grace Chip?

The Olympus core inside Vera delivers up to 50% higher instructions per cycle than Grace, Nvidia’s previous data center chip, according to the company’s technical blog. That is separate from the 50% edge Nvidia claims over x86 rivals like Intel and AMD.

Which AMD Chip Is Nvidia Benchmarking Vera Against?

Nvidia’s white paper compares Vera directly against AMD’s Epyc 9755, built on AMD’s Turin architecture. In a PageRank benchmark reported by Tom’s Hardware, Vera held onto most of its single core performance through 32 cores, while the Epyc chip’s performance flattened out around 20.

Will Vera Replace Nvidia’s GPUs in AI Servers?

No. Vera is meant to sit alongside Nvidia’s GPUs, not instead of them. Karl Freund of Cambrian AI Research said the chip is not built to run websites or ordinary server jobs; it is reserved for intense AI work like agent processing and reinforcement learning.

What Happened to AMD and Intel’s Stock the Last Time Nvidia Unveiled a Similar Chip?

On June 1, when Nvidia introduced RTX Spark, an Arm based laptop chip built with MediaTek and Microsoft, Intel fell 6% and AMD fell 5% in a single session, while Nvidia gained about 4%, according to a Forbes analysis. Vera’s July spec release did not trigger a similar selloff in either stock.