Google is developing a new server chip designed to run its Gemini AI models far more efficiently, according to a report from The Information that was picked up widely across the tech and financial press.
The chip, informally dubbed "Frozen v2," would take an unusual approach: rather than running Gemini purely as software on general-purpose hardware, it would embed parts of the model's architecture — what several outlets, including CNBC, described as Gemini's "blueprint" — directly into the silicon. Reuters, relaying The Information's July 20 report, said the goal is to serve Google's AI models to users more efficiently. The Information cited two sources.
Some outlets went further on the potential payoff. WION and TradingKey reported that the chip could make Gemini up to 10 times more efficient, framed as processing per unit of power consumption. According to Business Standard, Google expects Frozen v2 to help ease an AI computing capacity crunch — one that has reportedly fueled internal tensions and led Google Cloud to turn away some outside customers.
This is a longer-term project, not an imminent product. Multiple reports, including Crypto Briefing and Techmeme's summary of The Information, indicated the chip is slated for deployment around 2028.
Investors responded quickly. CNBC, Yahoo Finance, Seeking Alpha and others reported that Alphabet's stock rose on the news, reflecting optimism that more efficient in-house silicon could strengthen Google's position in the costly AI hardware race.
Why it matters: the enormous electricity and computing costs of running AI are one of the industry's biggest constraints, so a chip that could squeeze far more performance from each watt would help Google control costs, expand capacity, and lean less on outside chipmakers — even if the promised gains and 2028 timeline remain unproven.