Nebius (NBIS.US) has announced that it will begin offering NVIDIA's (NVDA.US) Vera Rubin NVL72 graphics processing units (GPUs) in the United States and Europe starting in the second half of 2026. According to the Netherlands-based AI infrastructure provider, it will deploy the NVIDIA Rubin platform through its Nebius AI Cloud and Nebius Token Factory, positioning itself as one of the first AI cloud suppliers to offer this computing platform. Nebius stated it will integrate the Vera Rubin NVL72 into its full-stack infrastructure housed in data centers across the US and Europe, enabling customers to build next-generation AI applications with regional availability and enhanced control. "By integrating Vera Rubin into the Nebius AI Cloud and our inference platform, Nebius Token Factory, we are providing the infrastructure for AI innovators and enterprises to develop agentic and inference-based AI systems more rapidly and efficiently," said Arkady Volozh, Founder and CEO of Nebius. The company described Nebius Token Factory as a post-training inference platform tailored for enterprise use. Nebius pointed out that the Rubin accelerated computing platform will complement its existing NVIDIA GB200 NVL72 and NVIDIA Grace Blackwell Ultra NVL72 capacity, thereby expanding the range of platform choices available to its customers. On Monday, NVIDIA CEO Jensen Huang indicated that the next-generation computing platform, Vera Rubin, which succeeds Grace Blackwell, has now entered full-scale production. Dion Harris, Senior Director of High-Performance Computing and AI Infrastructure Solutions at NVIDIA, characterized Vera Rubin as "a single AI supercomputer composed of six chips." This platform is built upon six core components: the Vera CPU, Rubin GPU, sixth-generation NVLink switch chips, ConnectX-9 network cards, BlueField 4 DPU, and the Spectrum-X 102.4T CPO, targeting next-generation AI workloads for the cloud and large-scale data centers. The Rubin GPU chip features a third-generation Transformer engine and delivers 50 PFLOPS of NVFP4 inference computing power, which is five times greater than NVIDIA's previous-generation Blackwell GPU. At the overall architecture level, the Vera Rubin platform can complete the training of ultra-large-scale Mixture of Experts (MoE) models in the same amount of training time while requiring only one-quarter of the GPUs previously needed, and it reduces the training cost per token to one-seventh of the original. NVIDIA also emphasized that Vera Rubin will support third-generation confidential computing technology and is set to become the industry's first rack-scale trusted computing platform, designed for AI scenarios with stringent requirements for security isolation, data privacy, and multi-tenant environments.