Nvidia Vera Rubin Platform Ramps Production With 10x Efficiency Gain Over Grace Blackwell

Nvidia on July 21, 2026, announced that its Vera Rubin NVL72 platform is entering volume production, with racks already running at partners CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure. The platform spans more than 350 factory sites across 30 countries, making it the largest rack-scale supply chain Nvidia has ever assembled.

CoreWeave's first benchmark on the DeepSeek-R1 model showed 10x more throughput per megawatt than the Grace Blackwell NVL72, a metric that directly addresses power constraints in AI factories. The result underscores the platform's focus on performance per watt and lowest token cost.

Vera CPU and Olympus Core Architecture

At the heart of Vera Rubin is the Vera CPU, Nvidia's first data center processor built with a custom core design. The Olympus core delivers 2x single-threaded performance, 3x core-to-core bandwidth, and 40% lower memory latency compared to competing chiplet designs, according to Nvidia. The CPU features 88 cores and 176 threads, with general release scheduled for the second half of 2026.

The platform integrates seven chips and five rack trays codesigned as a single system: Vera Rubin NVL72, Vera CPU rack, Groq 3 LPX, Spectrum-6 SPX, and Vera BlueField-4 STX. Key specifications include:

• Sixth-generation NVLink scale-up networking with 2x throughput, 3x lower latency, and 10x higher packet rates versus off-the-shelf Ethernet.
• Spectrum-X Ethernet with 102.4T Spectrum-6 switches and 1.6T ConnectX-9 SuperNICs, enabling 1.6x higher RDMA bandwidth.
• NVIDIA Photonics with co-packaged optics for scale-out, delivering 5x lower power and 10x higher mean time between interruptions versus pluggable transceivers.
• A cable-free, fan-free, hose-free compute tray design that cuts assembly time from hours to one minute.
• 45-degree Celsius liquid cooling inlet temperature enabling chiller-free dry-cooler operation, saving millions of gallons of water per megawatt annually.

European Infrastructure and Partner Deployments

Vera Rubin is also powering a new multibillion-dollar partnership between Microsoft and Mistral to expand AI infrastructure in Europe. Mistral will draw on thousands of Vera Rubin GPUs to increase compute availability for customers, providing a shared platform for training, inference, and large-scale deployment. The partnership aims to deliver sovereign-ready AI across public cloud, cloud-connected, and fully disconnected private cloud environments.

Nvidia said the platform's full-stack architecture combines accelerated computing, networking, and software to support agentic workloads that can consume up to 15x more tokens than traditional AI applications. With production ramping now and general availability of the Vera CPU expected in H2 2026, the Vera Rubin platform is positioned to define the next generation of AI factory infrastructure.