Nvidia's data center CPU business hit $5 billion in trailing twelve-month sales as of its fiscal second quarter 2027 report, and the company says revenue in the category will more than double in the next fiscal year. The figure matters because it marks the first time the GPU leader has disclosed meaningful scale in a market where Intel and Advanced Micro Devices have long set the terms.
Grace establishes a foothold
The Grace family of processors, not the newer Vera line, produced the $5 billion run rate. Nvidia had largely ignored server CPUs for a decade, content with margins from graphics chips that carry the heavy lift in AI training and inference. A shift toward more CPU-intensive inference workloads, plus the sheer volume of deployed models, changed the calculus.
Vera enters production
Full production of the Vera CPU is now underway. The chip arrives alongside the Vera Rubin GPU family, and Nvidia is selling the combination as an integrated platform as well as a stand-alone processor. The company states that Vera delivers performance and efficiency significantly exceeding any other data center CPU in key areas, a claim that, if borne out, would let it take share from the incumbents quickly.
Integrated advantage
Control of the high-end GPU stack gives Nvidia a structural lever: it can co-design CPU and GPU silicon for joint optimization, something neither Intel nor AMD can replicate at the same level of GPU dominance. The Vera Rubin pairing is the first full expression of that strategy.
What to watch
The next fiscal year will test whether the projected doubling materializes from actual deployments or remains a slide-deck target. Intel's Xeon roadmap and AMD's Epyc refreshes are both in motion, and neither vendor has ceded the socket count without a fight.