Overview
Speaker: Jensen Huang
The focus of Jensen’s address is on NVIDIA Omniverse DSX, a platform for designing building and operating gigascale AI factories using digital twins. It also highlights the significance and advancements in open-source AI models ecosystem integrations across clouds and SaaS new partnerships in cybersecurity and enterprise analytics and the computing stack and partners enabling “physical AI” and advanced robotics including a Foxconn factory case study.
Omniverse DSX blueprint and digital-twin workflow
- DSX collaborates with NVIDIA’s AI infrastructure stack to co-design buildings power and cooling systems beginning with an Omniverse Digital Twin.
- Jacobs Engineering optimizes compute density and layout to maximize token generation under power constraints.
- They aggregate SIEM ready OpenUSD assets from Siemens, Schneider Electric, Trane, and Vertiv into PTC’s product lifecycle management, then simulate thermals and electricals with CUDA accelerators.
- NVIDIA partners like Vectel and Vertiv deliver prefabricated, factory-built and tested modules, shrinking build time and accelerating time to revenue.
- Once the physical factory is operational, the digital twin serves as its operating system. Engineers then use AI agents from Phydra and Emerald AI – both trained on the digital twins – to optimise power consumption and minimise grid strain.
- Impact: For a one‑gigawatt AI factory, DSX optimizations can deliver billions of dollars in additional annual revenue across Texas, Georgia, and Nevada.
- NVIDIA is building an AI factory research center in Virginia using DSX to test and productize Vera Rubin, from infrastructure to software.
- Approach: design, plan, optimize, and operate entirely as digital twins long before physical deployment.
Open-source models and NVIDIA’s contributions
- Open-source models have become highly capable due to reasoning, multi-modality, and efficiency from distillation—now essential for developers and startups embedding domain expertise.
- Open source is critical for researchers, developers, companies, and national leadership; the United States “has to lead in open source.”
- NVIDIA states it leads in open-source contributions, with 23 models on leaderboards across language, physical AI, robotics, and biology; claims “number one” speech, reasoning, and physical AI models, enabled by NVIDIA’s own supercomputers.
Ecosystem, clouds, and SaaS integrations
- Startups build on NVIDIA due to a rich ecosystem, tools that work across all NVIDIA GPUs, and GPU availability everywhere—clouds and on‑prem—including enthusiast PCs.
- GPU cloud providers serving startups include CoreWeave, Nscale, Nebius, Lambda, and Crusoe.
- Integrations: AWS, Google Cloud, Microsoft Azure, and Oracle integrate NVIDIA GPUs, libraries (including CUDAx), and open-source AI models so workloads run consistently across clouds.
- Agentic SaaS momentum: NVIDIA libraries are integrated into ServiceNow (Bill McDermott) and SAP (Christian Klein), including CUDAx, Nemo, and Nemotron; work with Synopsys (Sassine) and Cadence (Anirudh) accelerates CAE/CAD/EDA and develops AI agents, with an aspiration to employ AI agent ASIC designers alongside human designers.
New partnerships and announcements
- CrowdStrike: joint effort to achieve cybersecurity at “speed of life,” deploying cybersecurity AI agents in the cloud and on‑prem/edge to detect threats within moments.
- Palantir Ontology facilitates collaboration to accelerate data processing for government, national security and enterprises. This includes structured, human-recorded and unstructured data, all integrated with NVIDIA for rapid processing and insightful discovery.
Physical AI computing stack
- Language models are trained, evaluated and inferred on a large GB200 system. Physical AI requires three computers: a training computer (either a GB or a GB200), an Omniverse computer for digital twin simulation and a robot computer like Thor or Jetson Thor, used in self-driving cars and robots. Some robots may use two robot computers. All three computers run CUDA, allowing AI to understand the physical world, laws of physics causality and permanence.
Case study: Foxconn Houston robotic factory
- Context: U.S. reindustrialization and reshoring. In Houston, Texas, Foxconn is building a state-of-the-art robotic facility for manufacturing NVIDIA AI infrastructure systems.
- Born digital in Omniverse: Foxconn engineers assemble the virtual factory in a Siemens digital twin solution built on Omniverse Technologies; all mechanical, electrical, and plumbing systems are validated before construction.
- Siemens plant simulation runs design-space exploration to find optimal layouts; when bottlenecks appear, engineers update the layout via Siemens Teamcenter.
- In Isaac Sim, the same digital twin trains and simulates robot AIs.
- Assembly flow: BANIC manipulators build the GB300 tray module; By manual manipulators from FII and skilled AI install bus bars; AMRs shuttle trays to test pods.
- Large-scale sensor simulation in Omniverse enables robot AIs to learn to work as a fleet.
- Overwatch: Vision AI agents built on NVIDIA Metropolis and Cosmos monitor fleets of robots and workers, alert engineers to anomalies and quality issues, and power interactive AI coaches for employee onboarding.
- Vision: people and robots working together define the future of manufacturing.
Robotics ecosystem highlights
- The factory is described as a robot orchestrating robots; digital twins are essential given software complexity.
- Caterpillar (Joe Creed) is incorporating digital twins in manufacturing.
- Figure (Brett Adcock): working with NVIDIA on training the AI, simulating the robot, and the robotic computer; described as “worth almost $40 billion today.”
- Humanoid robots are likely to become one of the largest consumer electronics and industrial equipment markets; “my friend Elon is also working on this.”
- Peggy Johnson and Agility: collaborating on robots for warehouse automation.
- Johnson Johnson surgical robots are presented as enabling completely non-invasive surgery with unprecedented precision.
- Disney Research: developing a new framework and simulation platform based on “Noom.” The Noom simulator trains robots in a physically aware, physically based environment. Demonstration shown was a simulation in Omniverse, not animation.
Next steps and actions
- Stand up the Virginia AI factory research center using DSX to test and productize Vera Rubin.
- Enable partners worldwide to design, build, bring up, and operate AI infrastructure faster—completely in digital before physical.
- Execute partnerships with CrowdStrike (cybersecurity AI agents) and Palantir Ontology (accelerated data processing and insights).
- Continue expanding NVIDIA’s open-source model contributions and leaderboard presence across domains.