Google's Frozen v2 chip targets 10x efficiency to break Nvidia's grip
google is developing a chip named frozen v2, targeting 2028 release with up to 10x better token-per-watt efficiency, aiming to reduce reliance on nvidia hardware.
google is developing 'frozen v2', a custom server chip scheduled for 2028 that aims to deliver tokens 6-10 times more efficiently per watt of power than its existing silicon. this chip targets inference workloads for google's gemini models and is part of a broader strategy to wean the company off nvidia hardware, which has dominated the market but created supply dependencies. the design addresses investor concerns regarding google's planned $180-$190 billion capital expenditure by promising better unit economics. news of the chip boosted google stock by 3%, reflecting market relief over the company's efforts to maximize ai infrastructure roi.