Nvidia said its Rubin platform is in full production as it rolled out a series of hardware updates at CEO Jensen Huang's CES 2026 keynote.
Huang said the Rubin platform will offer an integrated hardware and software stack that will provide a 10x reduction in inference token costs and a 4x reduction in the number of GPUs needed to train foundational models relative to Nvidia's Blackwell platform.
The company also noted that Microsoft's next-gen Fairwater AI factories will feature Nvidia Vera Rubin NVL72 rack systems with CoreWeave being among the first to offer the latest platform.
- Nvidia highlights its robotics momentum as Qualcomm makes its platform case
- A look at how Mercedes Benz, Nvidia collaborated on autonomous vehicles
Huang's keynote walked through a lot of backstories for Vera Rubin and its various networking and storage components. "Vera and Rubin are co designed from the start to bi directionally and coherently share data faster and with lower latency," said Huang.
Here's what you need to know about Nvidia's hardware announcements:
- Nvidia's Rubin platform uses "extreme codesign" across six chips including Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU and Spectrum-6 Ethernet Switch.
- Adoption of the Rubin platform is broad with most AI labs and cloud providers on board.
- Huang outlined Nvidia BlueField 4, which is behind the Nvidia Inference Context Memory Storage Platform, an AI-native storage system for inference.
- BlueField 4 is designed for long-context processing agentic AI systems.
- The Inference Context Memory Storage Platform is designed for AI agents long-term memory.
- Nvidia demonstrated how DGX Spark and DGX Station deskside AI supercomputers can leverage 100 billion parameter models on Spark and 1 trillion parameter models on DGX Station.
