Google Is Building an AI Chip Just for Gemini—And Investors Already Moved On It
Google is reportedly building a new AI chip, codenamed Frozen v2, to run Gemini faster and much more efficiently. Unlike Google’s general-purpose TPUs, this chip would hardwire part of Gemini’s architecture into the silicon, reducing redundant computation and memory traffic. Engineers estimate a 6–10x gain in tokens per watt, which could sharply lower the cost of serving Gemini and improve Google’s margins. The project reflects a broader scramble among major AI companies to reduce dependence on Nvidia GPUs and build custom silicon for their own models. Google’s urgency is tied to capacity limits: it has reportedly turned away demand because it lacked enough compute to serve customers. Frozen v2 is still exploratory, not confirmed by Google, and would not be sold to Cloud customers. Deployment is reportedly aimed for 2028 or later, while Google uses rented Nvidia capacity as a temporary bridge.
