Google develops secret chip Frozen v2 to boost Gemini AI efficiency

Google is developing a new server chip, codenamed Frozen v2, designed to run its Gemini AI models with six to ten times greater efficiency per unit of power compared to its current TPUs Source: lavanguardia. The chip integrates parts of Gemini's logic directly into silicon, reducing data movement and mathematical operations, according to The Information Source: lavanguardia.

What is Frozen v2

Frozen v2 is a specialized chip separate from Google's TPU line, aiming to alleviate computing capacity shortages that have caused internal tensions Source: elespanol. The chip could be deployed as early as 2028, but it remains a pilot project and may not reach production Source: larepublicaco.

Market reaction

Following the news, Alphabet's shares rose up to 3.7%, reaching $359.68 in New York Source: larepublicaco. The development reflects a broader trend among AI leaders like OpenAI and Anthropic to create custom chips for competitive advantage Source: larepublicaco.

Challenges ahead

Frozen v2's extreme specialization poses a risk: if Gemini's architecture changes, the chip could become obsolete Source: lavanguardia. Google has not committed to mass production, emphasizing that not all explorations become products Source: elespanol.