Alphabet’s Google is designing a new server chip codenamed Frozen v2. This hardware embeds parts of the Gemini AI architecture directly into the silicon to reduce computation and data movement.
The design aims to be six to ten times more power-efficient than current Google custom hardware. This project addresses an internal shortage of AI computing capacity that has forced Google Cloud to decline some external customer deals.
Frozen v2 will serve as a specialized line of hardware to complement existing Tensor Processing Units. The project remains in the design phase with a potential deployment target of 2028.