Google developing 'Frozen v2' chip to supercharge Gemini AI

Google is developing Frozen v2, a custom AI chip designed to optimise Gemini model inference and reduce computational costs. The chip represents Google's continued vertical integration strategy to compete with Nvidia's dominance in AI silicon. If you sell inference optimisation software or serve as a middleware layer between models and silicon, Google just built around you. The economics of AI deployment have permanently shifted—custom chips now beat commodity GPUs on cost, which means your margin is gone unless you're embedded inside that stack.