Google Develops Frozen v2 Chip to Boost Gemini Efficiency

The generative AI race has sharply increased demand for computing power and electricity, making chip efficiency critical to the cost and scalability of advanced models. Alphabet’s Google, which has long developed its own tensor processing units, is now pursuing a server chip tailored to Gemini. The strategy could reduce its reliance on Nvidia accelerators while lowering the energy and operating costs of running AI services.

Google is developing the chip under the code name “Frozen v2,” according to reports, with a launch targeted for 2028. The processor is expected to deliver six to 10 times the performance of current chips while improving Gemini’s computing and energy efficiency. Alphabet has not disclosed a development budget, production schedule or planned deployment scale, and the performance claims remain subject to validation in the finished product.