Alphabet said to be building a ‘Frozen v2’ AI server chip to make Gemini far more efficient

Alphabet is developing a custom AI server chip, code-named “Frozen v2,” to run its Gemini models more efficiently, according to The Information. The silicon, targeted for 2028, could deliver six to ten times better efficiency, measured as tokens generated per unit of power, than Google’s current AI chips. Google neither confirmed nor denied the project, saying only that it routinely co-designs hardware and software to optimize real-world workloads. Tech giants are racing to design in-house chips to cut costs, ease supply constraints and reduce dependence on Nvidia’s dominant hardware. OpenAI has announced its first inference chip and Anthropic is reportedly weighing a chipmaking tie-up, underscoring the shift. Alphabet’s multiyear AI buildout—budgeted at $180 billion to $190 billion—has unnerved some investors; the Frozen v2 report sent shares up about 3% ahead of earnings this week.

Read more