NVIDIA has made an announcement that its Vera Rubin platform, which is tailored for "gigascale AI factories," is undergoing a global expansion in its deployment. At present, prominent cloud service providers such as CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud have all embraced the platform's NVL72 system.

The Vera Rubin platform is a sophisticated integration of diverse hardware components. It has successfully established its presence across more than 350 factory nodes worldwide, covering a vast geographical expanse of 30 countries.

Tests carried out by CoreWeave have revealed some remarkable findings. When compared to the Grace Blackwell NVL72, the Vera Rubin platform demonstrates a tenfold surge in Token throughput per megawatt. This significant improvement effectively boosts the efficiency of both AI training and inference processes. Moreover, it leads to a reduction in infrastructure costs and an enhancement in energy utilization.

DeepInfra benchmark tests further highlight the superiority of the NVIDIA Vera CPU. It is over twice as fast as other CPUs. Under the same service quality conditions, it can support up to 1.6 times more concurrent AI agents. Additionally, its collaboration speed is up to 2.2 times higher. This performance makes it well-suited to meet the stringent requirements of cost-effectiveness, low-latency, and high throughput for production-grade AI agents.

NVIDIA emphasizes that as AI agents are entrusted with increasingly complex tasks, the scheduling efficiency of CPUs during model invocation becomes a pivotal factor influencing the performance of AI infrastructure. The Vera CPU is specifically engineered to handle agentic workloads. By doing so, it assists cloud service providers in optimizing infrastructure utilization and enhancing cost efficiency.