Nvidia announces new storage platform and confidential computing features for Vera Rubin NVL72.
Nvidia announced its Rubin GPU platform at CES 2026 through a keynote by CEO Jensen Huang, marking the launch of the highly anticipated follow-up to its fast-selling Blackwell Ultra products. While the company said Rubin is in "full production," availability from partners won't begin until the second half of this year. Nvidia executives have recently countered concerns about an AI data center bubble by citing expected revenues of $500 billion from Blackwell and Rubin products between the start of last year and the end of this year, pointing to ongoing demand for generative, agentic and physical AI solutions. The platform has garnered support from major tech companies including Amazon Web Services, Microsoft, Google Cloud, CoreWeave, Cisco, Dell Technologies, HPE, and Lenovo.
The company will initially deliver Rubin through two platforms: the Vera Rubin NVL72 rack-scale platform, which connects 72 Rubin GPUs and 36 of its custom Arm-compatible Vera CPUs, and the HGX Rubin NVL8 platform, which connects eight Rubin GPUs for servers running on x86-based CPUs. The rack-scale platform was originally called the Vera Rubin NVL144 when revealed at Nvidia's GTC 2025 event last March, with the 144 number reflecting GPU dies per rack, but Nvidia opted to adopt the NVL72 nomenclature used for Grace Blackwell to reflect GPU packages instead. According to Dion Harris, senior director of high-performance computing and AI infrastructure solutions at Nvidia, "Essentially we're just being consistent with how we've deployed and talked about it for Blackwell, and we're carrying that forward for Vera Rubin as well."
The Vera CPU features 88 custom Olympus cores, 176 threads with Nvidia's new spatial multi-threading technology, 1.5 TB of system LPDDR5x memory, 1.2 TBps of memory bandwidth, and confidential computing capabilities including a 1.8 TBps NVLink chip-to-chip interconnect to support coherent memory with the GPUs. Harris said the CPU's confidential computing feature enables Vera Rubin to deliver the "first rack-scale Trusted Execution Environment, maintaining data security across CPU, GPU and the NVLink domain [to protect] the world's largest proprietary models, training data and inference workloads." Compared to Nvidia's Grace GPU, the Vera offers double the performance for data processing, compression and code compilation.
The Rubin GPU delivers 50 petaflops for NVFP4 inference computing, five times faster than Blackwell, and 35 petaflops for NVFP4 training, which is 3.5 times faster than its predecessor. Its HBM4 high-bandwidth memory achieves 22 TBps bandwidth, 2.8 times faster than Blackwell, while NVLink bandwidth per GPU reaches 3.6 TBps, double that of Blackwell. The platform incorporates the liquid-cooled NVLink 6 Switch with 400G SerDes, 3.6 TBps of per-GPU bandwidth for GPU-to-GPU communication, a total bandwidth of 28.8 TBps and 14.4 teraflops of FP8 in-network computing, alongside Nvidia's ConnectX-9 SuperNIC and BlueField-4 DPU for scale-out networking.
The Vera Rubin NVL72 platform achieves 3.6 exaflops of NVFP4 inference performance, five times greater than Blackwell, and 2.5 exaflops of NVFP4 training, 3.5 times higher than its predecessor. The platform features 54 TB of LPDDR5x capacity, 2.5 times higher than Blackwell, and 20.7 TB of HBM4 capacity, 50 percent more than Blackwell, with HBM4 bandwidth reaching 1.6 PBps, 2.8 times greater, and scale-up bandwidth of 260 TBps, double that of the Blackwell NVL72 platform. Harris noted, "That's more bandwidth than the entire global internet." The platform also features the third generation of Nvidia's NVL72 rack resiliency technologies, including a cable-free modular tray design enabling 18 times faster assembly and service, along with NVLink Intelligent Resiliency for "zero downtime" maintenance capabilities.