Alphabet is reportedly developing a new in-house server chip, internally named "Frozen v2," intended to make its Gemini AI models run more efficiently. The chip, expected to launch in 2028, forms part of a wider industry push by AI firms to design their own hardware, both to boost efficiency and to reduce reliance on Nvidia, which has long dominated the AI chip market. Efficiency has become an increasingly important selling point as investor enthusiasm for heavy AI spending has cooled.

According to The Information, which first reported the news citing anonymous sources, Frozen v2 could be six to ten times more efficient than Google's current AI chips, measured by tokens generated per unit of power. Google did not confirm or deny the report directly, instead saying its teams are "constantly researching and experimenting with new innovations." The news follows similar moves by rivals, including OpenAI's custom "Jalapeño" inference chip and reported talks between Anthropic and Samsung on chip manufacturing. Alphabet shares rose around 3% on the news, offering a boost ahead of its earnings report this week, amid investor scrutiny of its planned $180-190 billion AI spending.