Nvidia details its next-generation Vera CPU for AI, setting up challenge to AMD and Intel
CNBC ● Covered by 8 sources
Nvidia just published full specs for Vera, its first from-scratch server CPU, and it's already shipping to OpenAI, Anthropic and SpaceX. That's Nvidia going after AMD and Intel's turf, not just GPUs.
Based on reporting by CNBC — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Nvidia didn't build its trillion-dollar valuation on CPUs. It built it on GPUs, the chips that train and run AI models while CPUs handled the boring plumbing. That's changing. On Tuesday the company laid out the architecture and benchmarks for Vera, its data center CPU, confirming that chips already reached OpenAI, Anthropic and SpaceX back in June.
The timing tracks with a bigger shift in how AI systems actually work. Back in 2022, when ChatGPT launched, a single CPU might sit next to eight GPUs, basically a bystander. Agentic AI changed that math. Agents that run for minutes or hours with little human oversight need something fast babysitting them, feeding data, answering the constant stream of small questions that keep a GPU busy. Ian Buck, Nvidia's VP of hyperscale, put it bluntly last week: CPUs are now "much more integral," especially in how fast they can answer a single question.
Wall Street noticed the shift before most of us did. AMD is up 128% this year and Intel 149%, both crushing Nvidia's comparatively modest 8% gain. Nvidia itself is pitching the entire server CPU market as eventually worth $200 billion, dwarfing Bernstein's $37 billion estimate for the mature market in 2025. Wolfe Research figures Nvidia could ship 1.3 million Vera chips this year at roughly $5,000 apiece, though Nvidia won't confirm pricing.
What makes Vera different isn't raw core count, the metric Intel and AMD have chased for years. Nvidia designed its Olympus core from scratch instead of licensing an off-the-shelf Arm design, and optimized for single-core speed, memory bandwidth and low latency. The goal, according to Nvidia product marketer Hannah Coutand, is getting agents back to the GPU as fast as possible so that expensive silicon never sits idle. The chip draws 250 to 450 watts and can pack in up to 1.5 terabytes of low-power memory, the kind normally found in phones and laptops, not servers.
Still, specs alone don't win contracts. AMD holds about a third of the server CPU market and has spent years cultivating hyperscaler relationships; Intel still commands roughly two-thirds. Nvidia's own partner list leans heavily on Oracle, with OpenAI promising large-scale Vera deployment this quarter. Analyst Karl Freund frames it as Nvidia trying to unhook customers from Intel and AMD entirely by offering a CPU built for a job nobody else is targeting. Coutand herself admits adoption is in "early innings." Whether cloud giants bite depends less on benchmarks than on how much they want another chip vendor besides Nvidia calling the shots.
My take — AI-written commentary, not fact-checked reporting
Nvidia isn't building CPUs because it suddenly loves general-purpose computing, it's building them so no one else gets to touch its rack. That's the vertical-integration playbook, and it's smart business but bad for anyone hoping for real competition in AI infrastructure. Watch OpenAI's actual deployment numbers this quarter, not the marketing slides, that's where you'll see if Vera is a genuine threat to AMD and Intel or just Nvidia hedging its own supply chain.
Read more about this at: CNBC