Vera, Nvidia's first server CPU, is its push into Intel and AMD's market, with a $20B sales projection for this fiscal year and early adopters including Anthropic, Oracle, and OpenAI.
Nvidia built its business on graphics processors. The chips that actually run software on a server, central processing units or CPUs, belonged to Intel and AMD. AI search startup Perplexity confirmed it will deploy Vera, Nvidia's first server CPU line, marking the chipmaker's first real move into a market it has historically left to its rivals.
Vera is built for a workload Intel and AMD's general-purpose server chips were not designed for: nonstop AI agents. An AI agent is an autonomous program that runs coding or task loops without a human at the keyboard, dispatching tools, writing code, and orchestrating other software through the day. Standard server CPUs spend much of their time waiting on those loops, per Nvidia's unveiling of Vera as "the CPU for Agents." Perplexity's own benchmarking showed Nvidia's architecture handled agentic coding tasks about 1.5x faster than standard server CPUs, per Perplexity Vice President for Computer Enterprise and Infrastructure Nate Kupp, as reported by Yahoo Finance and US News.
Nvidia is not chasing Intel and AMD into CPUs because margins are easy; it is doing so because its core GPU franchise is under structural threat. Elite AI labs including OpenAI and DeepSeek have been designing their own AI accelerators. If the largest model builders internalize the silicon layer, Nvidia's GPU business becomes a much smaller business. Vera is partly a hedge. By owning the CPU that runs the agent alongside the GPU that trains the model, Nvidia stays in the rack even when customers start designing their own training chips. The implication is straightforward: a GPU company is now a CPU company, and it is moving on Intel and AMD's territory at the same moment its biggest customers are weighing how to make Nvidia less central to their stacks.
The startup is part of an early enterprise cohort that also includes Anthropic, Oracle, and OpenAI, according to News.Az's coverage of Reuters reporting and Mezha.net's writeup. Nvidia has not disclosed the size or financial terms of Perplexity's order. The cohort matters as signal: the buyers testing Vera are exactly the labs whose custom-silicon ambitions Nvidia is hedging against. If Anthropic, Oracle, or OpenAI follow Perplexity onto Nvidia's CPU rack, the chipmaker's agent-class CPU becomes the default for the next server build cycle, and Intel and AMD's data-center roadmaps face a new kind of competitor.
The financial weight Nvidia is putting behind the line is large. The company expects Vera to generate roughly $20 billion in sales by the end of this fiscal year, per Reuters reporting cited by News.Az. That figure is a sales projection, not booked revenue, and it originates from a single Reuters dispatch; investors should treat it as guidance until it shows up in Nvidia's own 10-Q, earnings call, or investor relations release. Vera sits inside the broader Vera Rubin platform that Nvidia announced around GTC 2026, with third-party coverage citing 336 billion transistors and a generational leap over the Blackwell era, figures Nvidia has yet to publish on its own spec sheet.
The next test is Nvidia's quarterly earnings call, where investors will press for a Vera-specific revenue line rather than the bundled data-center figure. A number close to the $20B projection would tell the market the CPU bet is real. As of today, Perplexity is the only customer with its name on the record; Intel and AMD have not responded on the record about the new entrant.