Nvidia is set to boost AI server demand with its new Vera Rubin platform, expected to ramp up in the second half of 2026. The platform, succeeding the Blackwell architecture, promises significant improvements, including 10x lower inference token costs and 4x fewer GPUs needed for training mixture-of-experts models. Performance-per-watt is projected to improve by up to 50x over Blackwell. The Rubin platform is currently in production at TSMC, with six new chips slated for mass production in late 2026. Major cloud providers like AWS, Google Cloud, and Microsoft Azure are preparing to integrate Rubin-based instances, with Microsoft planning extensive deployments. This development could impact TSMC's capacity and intensify competition among chip designers, including AMD and custom silicon from Google and Amazon.