NVIDIA Releases Vera Rubin Benchmarks, Betting on Full-Stack AI Data Center Supply
Miles Bennett
Nvidia published benchmark data for its next-generation Vera Rubin chip system just ahead of AMD's annual product launch — a move that signals its accelerating shift from GPU specialist to full-stack AI data center supplier, where the real contest is selling entire systems, not single chips.
What exactly is Vera Rubin?
Vera Rubin succeeds the Grace Blackwell hybrid superchip, built on a two-GPU-to-one-CPU architecture.
A single NVL72 superchip system packs 72 Rubin GPUs and 36 Vera CPUs, working together in a fixed 2:1 ratio.
This means → Nvidia is no longer selling GPUs alone — it is bundling its own CPU into the system and selling the whole rack as one product.
Why is Nvidia suddenly making CPUs?
Ian Buck, Nvidia's VP of accelerated computing, said at the technical session: "We're advancing not just GPUs but CPUs. In Silicon Valley, you innovate or you die."
In plain terms = AI is moving from training large models to more complex agentic systems — AI that autonomously handles multi-step tasks — and those systems need CPUs to coordinate data flow, networking, and software.
GPUs remain the core for training and inference, but GPUs alone are no longer enough. Nvidia wants to cover both ends and sell customers a complete system, not a single chip.
Who is already using it, and how hard is deployment?
Nvidia executives disclosed that OpenAI is already running a Vera Rubin rack.
The new NVL72 rack uses a liquid-cooled, integrated platform. Nvidia emphasized its "plug-and-play" design is simpler than prior generations, aiming to lower the deployment barrier for data centers.
This means → Nvidia is not just selling chips — it is solving the "how do I install this" problem for buyers. Reducing deployment friction is itself a competitive moat.
Why does the timing matter?
Nvidia released the benchmarks just before AMD's annual product event in San Francisco — a clear pre-emption move.
The Vera CPU will also be sold standalone; Nvidia has reportedly told Chinese customers it could ship as early as August this year.
This reflects a strategic pivot: from the single-point narrative of "best GPU performance" to the system narrative of "full-stack AI data center supplier." Whether that story converts into real CPU market share is the key test for this strategy.
Content is for reference only, not financial advice.