Overview
- Vera Rubin is presented as a platform, not a single chip, integrating CPU Vera, GPU Rubin, NVLink 6, ConnectX-9 SuperNIC, BlueField-4 DPU and Spectrum-6 Ethernet for end-to-end AI workloads.
- Nvidia reports up to 5× higher inference performance, up to 10× lower inference token cost, 4× fewer GPUs for MoE training and roughly 50 petaflops of inference on the Rubin GPU, with figures pending independent validation.
- CPU Vera is described as optimized for agentic reasoning with 88 custom Olympus cores and energy efficiency aimed at large AI data centers.
- Industry leaders signaled support, with Microsoft outlining Fairwater AI facilities using hundreds of thousands of Vera Rubin superchips and major cloud providers preparing deployments in the second half of 2026.
- Nvidia also introduced Alpamayo for ‘physical AI’ in autonomous vehicles, with the first car expected to appear on U.S. roads in the first quarter, and highlighted recent record revenue and a $5 billion Intel stake.