The Rubin platform enters full production
NVIDIA announced at CES that the Vera Rubin platform is in full production, with 50 petaflops of NVFP4 inference and tokens at one-tenth the cost of Blackwell. Six open domain models were released alongside it.
Why it matters
Blackwell's successor moved from roadmap to production, and the token economics the whole 2025 build-out ran on were restated at a tenth of the cost.
NVIDIA describes Rubin as its first extreme-codesigned, six-chip AI platform: Rubin GPUs, Vera CPUs, NVLink 6 scale-up networking, Spectrum-X Ethernet Photonics scale-out networking, ConnectX-9 SuperNICs and BlueField-4 DPUs. The stated figures are 50 petaflops of NVFP4 inference and AI tokens at one-tenth the cost, with an AI-native KV-cache tier giving 5x higher tokens per second, 5x better performance per total-cost-of-ownership dollar and 5x better power efficiency for long-context inference. The same announcement released six open domain models: Clara for healthcare, Earth-2 for climate science, Nemotron for reasoning and multimodal AI, Cosmos for robotics and simulation, GR00T for embodied intelligence and Alpamayo for autonomous driving. Alpamayo is presented as an open portfolio of reasoning vision-language-action models enabling level 4-capable autonomy.