# Large language models

The Token Wars Begin: Nvidia’s Vera Rubin vs China’s Low‑Cost Inference Push

At GTC 2026 Nvidia declared the AI era has shifted from training models to continuously generating tokens and presented Vera Rubin, a full‑stack platform it says can cut token costs dramatically. At the same time, Chinese large‑model providers are already undercutting foreign counterparts on token prices and capturing high API volumes, creating a global contest over who will set token pricing and infrastructure standards.