Google Developing Frozen v2 AI Server Chip

Google is working on Frozen v2, a specialized server chip designed to hardwire Gemini model architecture into silicon for efficiency.

Google is developing a specialized server chip codenamed Frozen v2 that etches Gemini model architecture directly into silicon, according to a report from The Information on July 20, 2026. Engineers project the chip could deliver six to ten times more tokens per unit of power than current Google TPUs. While the original Frozen design led by Jeff Dean proposed baking model weights into the die, Frozen v2 keeps weights updatable to avoid short life cycles. Google targets a 2028 deployment for the chip, which will not be produced at TPU volumes. This development follows Google Cloud turning down customers due to compute shortages and the split of eighth-generation TPUs into training and inference variants. Other industry moves include Toronto startup Taalas launching the HC1 chip with Llama 3.1 permanently wired and Nvidia licensing technology from Groq. A Google spokesperson noted that not every project reaches production.