NVIDIA Vera Rubin Platform Powers Europe’s Open-Model Era with Unmatched Performance and Efficiency

NVIDIA Unveils Vera Rubin: A New Era in AI Supercomputing

NVIDIA has launched the Vera Rubin platform, a rack-scale AI supercomputer designed to deliver high performance per watt and token cost efficiency. The system, now ramping into full production, is being deployed by major cloud providers including CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure. Built with extreme codesign across seven chips and five rack trays, Vera Rubin aims to redefine AI infrastructure by unifying compute, networking, and software into a single, optimized system.

NVL72 System Targets Agentic AI Workloads with 10x Efficiency Gains

The Vera Rubin NVL72 system, the platform’s core component, is engineered to address the growing demands of agentic AI workloads, which can consume up to 15x more tokens than traditional applications. According to NVIDIA, the system delivers up to 10x more tokens per megawatt compared to the GB200 NVL72, a metric critical for power-constrained AI factories. CoreWeave, one of the first partners to validate the platform, reported a 10x improvement in tokens per second per megawatt during benchmarks of the DeepSeek-R1 model. This performance leap, achieved through custom-designed chips like the NVIDIA Vera CPU and advanced networking solutions, positions Vera Rubin as a key enabler for large-scale reinforcement learning and open-model ecosystems.

Tight Integration of Custom Chips and Networking Tech Boosts Performance

The Vera CPU, featuring custom Olympus cores, offers 2x single-threaded performance, 3x core-to-core bandwidth and 40% lower memory latency compared to competing chiplet designs, optimizing efficiency for agentic workloads. Networking is handled by sixth-generation NVLink scale-up and Spectrum-X Ethernet, which provide more than 2x throughput, 3x lower latency, and 10x higher packet rates than off-the-shelf Ethernet. For scale-out, Spectrum-X’s 102.4T switches and 1.6T ConnectX-9 SuperNICs enable 1.6x higher RDMA bandwidth, while NVIDIA Photonics’ co-packaged optics reduce power consumption by 5x and increase MTBI by 10x over pluggable transceivers.

Europe’s AI Strategy Shifts as Microsoft, Mistral Leverage Vera Rubin

Vera Rubin’s deployment is already reshaping AI infrastructure strategies in Europe. Microsoft and Mistral have partnered to expand the region’s open-model ecosystem, leveraging Vera Rubin’s compute power to train and deploy frontier AI systems while adhering to local data governance laws. The collaboration includes a multibillion-dollar agreement to increase GPU capacity, with Mistral’s models now available in Microsoft Foundry and integrated into Copilot Studio. This setup allows governments and enterprises to adopt AI on their own terms, balancing innovation with regulatory compliance.

Google Cloud Launches A5X Instance, Cuts Inference Costs 10x

Google Cloud has also adopted Vera Rubin for its new A5X instance, targeting startups like Ineffable Intelligence, which develops “superlearner” systems through reinforcement learning. The A5X instance, built on Vera Rubin’s rack-scale architecture, offers 10x lower inference costs per token and 10x higher token throughput per megawatt than the prior generation. Google’s Virgo networking further enhances scalability, enabling clusters that can scale to tens of thousands of NVIDIA Rubin GPUs within a single site and up to nearly a million GPUs across multisite configurations.

Liquid Cooling and Assembly Streamlining Cut Costs, Water Use

NVIDIA claims Vera Rubin’s design cuts compute tray assembly time from hours to one minute, eliminating cables, fans, and hoses in the tray. Its 45-degree Celsius liquid cooling inlet temperature allows chiller-free dry-cooler operation, saving millions of gallons of water annually per megawatt. These efficiency gains are critical for AI factories seeking to reduce both operational costs and environmental footprints. DeepInfra’s benchmarks further highlight the platform’s advantages, showing the Vera CPU supports up to 1.6x more concurrent AI agents and up to 2.2x faster orchestration compared to alternative CPUs, improving infrastructure utilization and cost efficiency.

Vera Rubin’s Global Expansion Hinges on Adapting to Agentic AI Demands

As Vera Rubin scales globally, its impact on AI development will depend on its ability to meet the evolving needs of agentic systems. With partners like CoreWeave, Google Cloud, and Microsoft accelerating adoption, the platform is poised to set benchmarks for performance, sustainability, and sovereignty in AI infrastructure.