Nvidia is repositioning itself from a GPU supplier to a provider of integrated accelerated computing systems, unveiling an AI factory reference design and detailing factory costs and equipment demand. The company frames the shift as necessary to scale artificial intelligence workloads across chips, software and operations.
CEO Jensen Huang set out the strategy in his GTC keynote and a subsequent Stratechery interview, saying accelerated computing "requires understanding applications" and broader systems work, the company said in an announcement.
Huang warned that poor cross-functional integration forces over-design and inefficiency when building large AI facilities.
Nvidia released the Vera Rubin DSX AI Factory reference design and an Omniverse DSX Blueprint, with DSX Max-Q, DSX Flex, DSX Exchange and DSX Sim as core software components.
The company lists partners including Dassault Systèmes, Cadence, Schneider Electric, Siemens, Jacobs, PTC, Procore, Switch, Nscale, Vertiv, Trane Technologies and Eaton. Nvidia says it has a $300 billion equipment backlog and more than 200+ gigawatts of demand waiting.
The company also defended its Vera CPU effort as complementary to GPUs, citing single-thread performance tuning and "bandwidth-per-core three times higher than existing CPU designs" to avoid GPU throttling.
Nvidia said it has restarted H200 chip manufacturing for Chinese customers after receiving U.S. export approvals, noting sales will share 25% of revenue with the U.S.; Chinese regulatory approval for imports remains pending before large-scale deliveries.
Related reading
- From Vera Rubins, Blackwell, Nemotron, NemoClaw, to Isaac and Cosmos - here's a whistlestop guide to NVIDIA's…
- Samsung and AMD sign memory partnership deal to power next generation of AI chips
- Nvidia launches NemoClaw to bring security infrastructure to always-on AI agents
Nvidia additionally unveiled space computing modules, the Nemotron model family updates and a NemoClaw single-command installer for agent deployments.
Partners are integrating the blueprint and Nvidia is offering tools to validate facility designs, while regulatory and customer approvals will determine delivery timing for chips and factory projects.
The recap
- Nvidia shifts to integrated accelerated computing systems globally.
- Nvidia reports $300 billion equipment backlog and 200+ gigawatts.
- Chinese regulatory approval pending before large-scale H200 deliveries.
