Sources: Google is developing a specialized server chip, informally dubbed "Frozen v2", that integrates Gemini AI model blueprints in the silicon, set for 2028 (The Information)
Google is aiming for a 2028 server chip that would hardwire parts of Gemini to cut inference costs.
Techmeme’s note says The Information reports the chip is informally called “Frozen v2” and is designed around Gemini’s model blueprint. The linked summaries say the goal is higher efficiency than Google’s newest TPUs, with some posts citing a 6x to 10x tokens-per-watt target. The item frames it as a Google effort to ease compute constraints for Gemini rather than a general-purpose TPU successor.
Techmeme's note
Techmeme’s note says The Information reports the chip is informally called “Frozen v2” and is designed around Gemini’s model blueprint. The linked summaries say the goal is higher efficiency than Google’s newest TPUs, with some posts citing a 6x to 10x tokens-per-watt target. The item frames it as a Google effort to ease compute constraints for Gemini rather than a general-purpose TPU successor.
Techmeme's note
score 6