Google is developing a new server chip codenamed Frozen v2, designed to improve the operational efficiency of its Gemini AI models, with a targeted release in 2028.
The chip could process 6 to 10 times more tokens per unit of power than Google's existing tensor processing units, potentially offering a significant leap in energy efficiency for large-scale AI workloads. Alphabet's stock rose roughly 3% following the disclosure.
The move comes as major AI firms push to reduce reliance on third-party silicon. OpenAI announced its first custom chip, dubbed Jalapeño, in June, while Anthropic is in talks with Samsung over a chipmaking partnership. Custom hardware has become a strategic priority across the sector as companies race to control the infrastructure underpinning generative AI.