Nvidia Vera Rubin Bundles CPUs With Rubin GPUs
.jpg)
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Nvidia unveiled performance benchmarks for its Vera Rubin system, pairing 36 Vera CPUs with 72 Rubin GPUs per NVL 72 rack, just days before rival AMD's annual product event in San Francisco.
- Ian Buck, Nvidia's VP of accelerated computing, framed the platform as part of a roadmap to "crank out new architectures, not just GPUs but CPUs" — a notable expansion beyond Nvidia's GPU roots.
- OpenAI already has one Vera Rubin rack in use; early customers also include Microsoft and Oracle, with full production and shipments slated for the second half of this year.
- Vera Rubin NVL72 is marketed as "cable-free compute" and "hot-swappable," with Nvidia claiming it processes 10x more tokens per watt than Grace Blackwell and cuts rack installation time from hours to minutes.
- Vera Rubin abandons the chiplet architecture common in x86 data center CPUs in favor of a monolithic design, which Nvidia argues eliminates the bandwidth tax of stitching chiplets together.
- AMD revealed more details about its competing Helios AI chip rack on Sunday as both companies compete for multiyear supply contracts with hyperscalers Meta and Amazon and AI labs OpenAI, Anthropic, and SpaceXAI.
- The Vera CPU will also be sold as a stand-alone product, with Nvidia reportedly telling Chinese customers it could be ready as soon as August.
Why it matters: Nvidia's dominance has rested on selling the best GPU; with AI workloads shifting toward agentic systems that lean heavily on CPUs for orchestration, the company is now making a concrete move to lock customers into Nvidia-only racks. The competitive stakes are immediate — AMD's Helios rack is being unveiled this week, and Meta, Amazon, OpenAI, Anthropic, and SpaceXAI are all in the market for multiyear chip supply deals.
