What happened
Alphabet is designing "Frozen v2", a new server chip slated for 2028, to boost the efficiency of its in-house Gemini models. The chip could be six to ten times more efficient than Google's current AI chips, measured by tokens generated per unit of power, according to The Information. This development follows a trend of AI companies, including OpenAI and Anthropic, developing custom silicon to reduce reliance on external vendors and optimise model operations. Google stated it is "constantly researching and experimenting" with hardware and software co-design.
Why it matters
The potential for a 6-10x efficiency gain in Gemini model operations will directly impact infrastructure architects and procurement teams, reducing the operational cost per inference. This internal hardware development, following OpenAI's "Jalapeño" chip announcement, signals a broader industry shift towards custom silicon to manage escalating AI compute expenditures. For founders and investors, this move mitigates the financial risk associated with Alphabet's $180-190 billion AI investment, demonstrating a clear path to cost optimisation and reduced dependency on third-party chip suppliers.
