NVIDIA Vera Rubin NVL72 Arrives at Oracle and CoreWeave, Signaling a New Era for AI Cloud Computing
NVIDIA’s next major AI platform is moving from anticipation to real-world deployment. The company’s Vera Rubin NVL72 systems are now being delivered to leading cloud providers for validation, marking an important step toward the next generation of large-scale AI infrastructure.
Oracle and CoreWeave are among the first cloud companies to receive and begin testing NVIDIA’s Vera Rubin NVL72 platform. These early systems are being validated in close collaboration with NVIDIA as cloud providers prepare to offer next-generation accelerated computing for demanding AI workloads, including agentic AI, large language models, inference, and mixture-of-experts training.
The NVIDIA Vera Rubin NVL72 rack is a massive AI system built around 72 Rubin GPUs and 36 Vera CPUs. Designed for cloud-scale performance, the platform represents NVIDIA’s follow-up to its Grace Blackwell generation, which has already set major performance records in AI benchmarks. With Vera Rubin, NVIDIA is aiming to push performance, efficiency, and scalability even further.
Oracle Cloud Infrastructure executive Mahesh Thiagarajan shared that OCI is among the first cloud providers bringing up a Vera Rubin NVL72 rack for validation testing. The move reflects Oracle’s broader effort to expand its AI infrastructure and deliver high-performance accelerated computing to enterprise customers at global cloud scale.
CoreWeave has also received and installed one of the first Vera Rubin NVL72 platforms. The system’s arrival highlighted just how large and complex these AI racks have become. Moving a single unit into a facility required multiple people, underscoring the physical scale of modern AI data center infrastructure. When multiplied across hundreds or thousands of racks, the size and power demands of next-generation AI clusters become enormous.
The Vera Rubin NVL72 is not just about GPUs and CPUs. It is part of a much broader AI computing stack. High-performance AI systems depend on advanced power delivery, liquid cooling, high-speed interconnects, networking, rack-level management, and a mature software ecosystem. NVIDIA’s long-developed CUDA platform, along with its AI-focused CUDA-X libraries, remains one of the company’s strongest advantages in the AI market.
CoreWeave is also pairing the new platform with its own infrastructure technologies, including software-defined liquid cooling and unified rack control systems. These technologies are designed to help manage the power, thermal, and operational challenges that come with deploying dense AI hardware at cloud scale.
According to NVIDIA’s positioning, Vera Rubin is expected to deliver major improvements over Blackwell in key AI workloads. The platform is said to offer faster mixture-of-experts training while using only a fraction of the GPUs, along with significantly lower cost-per-token for inference. These improvements could make a major difference for companies building and serving large AI models, where performance and operating costs are critical.
Volume production of NVIDIA Vera Rubin is already underway, and the first operational deployments are expected to begin soon. As Oracle, CoreWeave, and other major cloud providers validate the platform, Vera Rubin is set to become a central part of the next wave of AI infrastructure.
The arrival of NVIDIA Vera Rubin NVL72 shows how quickly the AI hardware race is advancing. As demand for agentic AI, generative AI, and large-scale inference continues to grow, cloud providers are investing heavily in systems that can deliver more compute power with better efficiency. Vera Rubin is positioned to be one of the most important platforms in that shift, potentially shaping the future of AI data centers for years to come.
