Google is developing a new server chip, code-named Frozen v2, to integrate elements of its Gemini AI model directly into hardware. The project aims to run flagship AI models more efficiently while addressing an internal shortage of computing capacity.
By hardwiring AI model components into silicon, the chip targets higher efficiency than Google’s existing custom Tensor Processing Units (TPUs). This initiative operates alongside current TPU projects rather than replacing them.
Engineers are currently finalizing the design. Google plans to deploy the new hardware as early as 2028.