Nvidia CEO Jensen Huang announced at CES 2026 that the company's Vera Rubin AI platform has entered full production. The platform, which combines new Rubin GPUs with Vera CPUs, represents Nvidia's architectural leap beyond Blackwell. Shipments to partners are planned for the second half of 2026.
A modular, cable-free tray design now allows technicians to assemble a Vera Rubin rack in about five minutes, compared to two hours previously-a 95% improvement. Microsoft Azure is already testing the first racks.
Nvidia has doubled its supply chain for Vera Rubin, now spanning over 350 factories in 30 countries, with 150 in Taiwan. The platform uses TSMC's 3nm process and next-generation HBM4 memory from Samsung and SK Hynix.
Nvidia claims up to 3.5 times better AI training performance and 5 times better inference versus Blackwell. The company now sells complete systems, not just GPUs.
For investors, the ability to scale production efficiently addresses past supply constraints, potentially reducing earnings volatility. However, heavy Taiwan concentration tempers geopolitical hedging.