Nvidia's Vera Rubin: Targeting Every Chip in AI Data Centers
Original: Nvidia Wants to Own Every Chip Inside AI Data Centers
Why This Matters
Nvidia's CPU expansion signals a bid to dominate full-stack AI infrastructure, intensifying competition with AMD and Intel.
Nvidia revealed new performance benchmarks for its Vera Rubin chip system, combining CPUs and GPUs in a single platform. The NVL72 system delivers 10x tokens-per-watt vs. Grace Blackwell. OpenAI is already using one rack, and Nvidia plans standalone CPU sales, possibly to China by August 2025.
Nvidia held a technical workshop at its Santa Clara headquarters to brief journalists on the Vera Rubin platform, its successor to the Grace Blackwell hybrid superchip system. The Vera Rubin NVL72 is a liquid-cooled rack unit pairing 36 Vera CPUs with 72 Rubin GPUs, designed around a 1-CPU-to-2-GPU ratio. Nvidia claims the system delivers 10 times more tokens per watt than Grace Blackwell and outperforms competing CPUs from AMD and Intel on agentic AI workloads—though the benchmarks reportedly used slightly older competitor chip generations.
The strategic push into CPUs marks a significant expansion for Nvidia, traditionally a GPU-focused company. As AI systems shift toward agentic architectures—requiring orchestration of data flows, networking, and software tasks—CPU demand is rising. Nvidia VP of Accelerated Computing Ian Buck stated: 'We're on a road map to crank out new architectures, not just GPUs but CPUs. We're going to keep innovating, because it's do this or die in Silicon Valley.'
Nvidia is also selling the Vera CPU as a standalone product and has reportedly indicated availability to Chinese customers as early as August. OpenAI already has one Vera Rubin rack deployed. CEO Jensen Huang was absent from the briefings, as he was in Japan announcing robotics AI partnerships.