Version 5

NVIDIA's Vera Rubin and Cosmos platforms have extended into two additional domains since their July 21 launch: surgical medical robotics, via an open-source Medical Physics Simulation framework with CMR Surgical and Johnson & Johnson MedTech as named partners[7], and chip design, where NVIDIA is using Vera CPU internally to accelerate EDA workloads and claims up to 1.5x improvement on formal verification and functional simulation.[8] Geographic expansion is also visible: South Korea's government, KAIST, SK Group, and SK Telecom signed a cluster of partnerships at a San Francisco AI Summit, including what NVIDIA describes as the first joint AI research lab between a Korean university and any global technology company.[9] Named deployments now span pharma, cloud, sovereign AI, defense education, medical robotics, and domestic manufacturing across North America, Europe, South Korea, and Japan.

The medical robotics and EDA use cases show NVIDIA moving its platform narrative into regulated industries and its own silicon design pipeline. If Vera CPU genuinely accelerates the EDA workloads used to design future NVIDIA chips, each hardware generation compounds the next — a claim that is commercially significant if independently confirmed but is so far asserted only by NVIDIA itself.

At GTC 2026 in March, NVIDIA framed its hardware roadmap around what it called the agentic AI era — a regime in which continuous post-training loops, rather than one-time training runs, define compute demand. The Vera Rubin platform and Jetson Thor line are the two products built around this thesis. For Vera Rubin, NVIDIA argues two efficiency metrics matter most: 'intelligence per dollar,' which accounts for the total cost of continuous post-training, and 'tokens per megawatt,' which NVIDIA says determines whether inference can scale profitably.[1][2] CoreWeave's benchmark on DeepSeek-R1 found a 10x improvement in tokens per second per megawatt on Vera Rubin NVL72 compared to Grace Blackwell NVL72; Google Cloud claims 10x lower inference cost per token and 10x higher token throughput per megawatt on its A5X instances built on the same platform.[2] The Vera CPU separately claims 2x single-threaded performance and support for 1.6x more concurrent AI agents than competing designs.[2]

The commercial deployment record spans distinct verticals. Bristol Myers Squibb is running an eight-system Vera Rubin NVL72 SuperPOD for drug discovery, agentic scientific workflows, and institutional knowledge compounding across therapeutic programs.[3] Google Cloud launched A5X instances on the platform, and Microsoft and Mistral signed a multibillion-dollar deal for European sovereign AI infrastructure using tens of thousands of Vera Rubin GPUs.[2] The Naval Postgraduate School commissioned a DGX GB300 for military AI education covering cybersecurity, ocean modeling, and operational decision-making, with NVIDIA and MITRE building a digital twin simulation framework using Omniverse.[4] Wistron's Fort Worth D1 facility, a $700 million investment, is producing Grace Blackwell Superchips and will add Vera Rubin production, scaling to tens of thousands of boards per month; Jensen Huang has framed domestic AI hardware production as economic reindustrialization comparable to roads or electrical grids.[5] Spectrum-6 networking, at 102.4 Tb/s switching capacity, accompanies the platform and is positioned as a prerequisite for full performance at clusters above 100,000 GPUs.[6]

Two domains added after the July 21 platform launch deepen the picture. On medical robotics, NVIDIA's open-source Medical Physics Simulation framework runs 8,192 parallel robot-training environments, reducing training time from over five hours to under two minutes.[7] CMR Surgical contributed nearly 500 hours of anonymized clinical data from its Versius surgical system; Johnson & Johnson MedTech is using the framework to build digital twins of its MONARCH endoluminal platform for urology.[7] NVIDIA is positioning the open-source release as meeting the transparency requirements for regulatory evidence — developers can inspect models and weights and reproduce results across anatomies.[7] On chip design, NVIDIA says Vera CPU delivers up to 1.5x improvement on Cadence Jasper formal verification and Synopsys VCS functional simulation, and NVIDIA is now using Vera in its own internal EDA workflows — creating, by its account, a feedback loop where each hardware generation helps build the next; a next-generation Rosa CPU built on the NVIDIA Rigel core is planned.[8]

Geographically, South Korea has emerged as a significant partnership cluster. At a San Francisco AI Summit attended by South Korean President Jae Myung Lee, NVIDIA and KAIST announced a joint AI research lab described as the first between a Korean university and any global technology company; SK Group expanded its collaboration with NVIDIA to co-develop memory for AI infrastructure, personal AI, and physical AI; and SK Telecom announced plans to build physical AI and robotics infrastructure.[9] On the edge, Cosmos 3 Edge ranked first on VANTAGE-Bench for vision analytics in the 4-billion-parameter class, and NVIDIA demonstrated MCP integrations for Adobe, Blender, Houdini, and Unreal Engine at SIGGRAPH 2026.[10] Advantech and ADLINK have both announced Jetson Thor-based hardware, with Q1 2027 general availability still ahead.[11][12] The competitive landscape remains one-sided: no named GPU competitor has contested NVIDIA's efficiency metrics or offered product-level counter-benchmarks.