NVIDIA Vera Rubin Driving Performance Per Watt, Lowest Token Cost for Partners Worldwide

Engineering for Gigascale Efficiency

NVIDIA's Vera Rubin platform represents a significant leap in AI infrastructure, built on an 'extreme co-design' philosophy that integrates seven chips and five rack trays into a single, unified system. At its core is the NVIDIA Vera CPU, featuring custom Olympus cores that deliver twice the single-threaded performance and three times the core-to-core bandwidth compared to competing chiplet designs. This meticulous engineering aims to optimize for 'agentic workloads,' which are increasingly critical for advanced AI applications. The platform also incorporates sixth-generation NVLink for scale-up, offering over double the throughput and three times lower latency, alongside Spectrum-X Ethernet for scale-out, which provides 1.6x higher RDMA bandwidth. These networking advancements, coupled with NVIDIA Photonics' co-packaged optics, significantly reduce power consumption and improve reliability, making the Vera Rubin system a highly integrated and efficient solution for demanding AI tasks.

Powering Global AI Factories and Sovereign AI

The Vera Rubin platform is rapidly being adopted by leading AI infrastructure builders and cloud providers, including CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure, SpaceXAI, and Tesla. This widespread adoption underscores the industry's need for high-performance, scalable AI compute. A notable partnership highlighted in the article is with Microsoft and Mistral in Europe, where Vera Rubin will underpin a multibillion-dollar agreement to expand AI infrastructure. This initiative is particularly focused on enabling 'sovereign-ready AI,' allowing European governments and regulated industries to deploy AI solutions that adhere to regional data control, governance, and autonomy requirements. By providing a robust computing foundation, Vera Rubin is positioned to accelerate Europe's open-model ecosystem and support the deployment of frontier AI models like Mistral Medium 3.5 and OCR 4 within secure, localized environments.

The Economic Impact of Token Cost Reduction

One of the most compelling aspects of the Vera Rubin platform is its promise of dramatically lower token costs and superior performance per watt. Benchmarks, such as CoreWeave's DeepSeek-R1 test, indicate a tenfold increase in throughput per megawatt compared to the Grace Blackwell NVL72, translating to one-tenth the cost per million tokens. This efficiency gain is critical for 'power-constrained AI factories' and for handling the escalating demands of agentic systems, which can consume up to 15 times more tokens than traditional AI applications. Beyond raw performance, the system's design also addresses operational efficiencies, such as reducing compute tray assembly time from hours to minutes and enabling chiller-free dry-cooler operation, saving millions of gallons of water annually. These combined innovations offer substantial economic advantages, making advanced AI more accessible and sustainable for partners worldwide.