Google Developing New “Frozen v2” Chip to Make Gemini More Efficient
Alphabet shares rose after a report revealed Google is building a server chip that embeds Gemini's architecture directly into silicon, targeting up to 10 times greater efficiency.
Google is trying to squeeze more performance out of every watt it spends on AI. The company is developing a new server chip that would build elements of its Gemini models directly into the hardware, The Information reports.
The chip, informally called “Frozen v2,” is designed specifically to make Gemini run more efficiently rather than replace Google’s existing custom processors.
Google’s parent Alphabet shares climbed on the news, just as the company’s second-quarter earnings report looms.
How Frozen v2 Would Work
The chip’s core idea is unusual: rather than treating hardware and software as separate layers, Frozen v2 would incorporate parts of the Gemini model’s architecture directly into the silicon itself, according to Reuters.
CNBC reports that permanently embedding elements of Gemini into the chip would reduce the computation and data movement needed to generate AI responses, driving the projected efficiency gains.
Reuters notes Google’s engineers are still finalizing the chip’s design and how much of the model will ultimately be hardwired, with deployment targeted as soon as 2028.
Once complete, the chip could serve six to 10 times more AI tokens per unit of power than Google’s current custom processors, according to the report.
Both Reuters and CNBC note the project complements, rather than replaces, Google’s existing Tensor Processing Units, giving the company a second homegrown chip family built specifically around Gemini’s architecture.
Addressing an AI Computing Crunch
The timing reflects a real capacity problem inside Google.
Reuters reports the Frozen v2 project is intended to help address an AI computing capacity crunch that has fueled internal tensions, forced the company to buy additional compute power from SpaceX, and prompted Google Cloud to turn down deals with outside customers.
The push mirrors a broader shift across the AI industry, where companies are increasingly building custom chips to run AI models more efficiently and reduce reliance on Nvidia, the AI chip market leader and is about to become the world’s most valuable company.
OpenAI announced its first custom inference chip, Jalapeño, in June, while Claude maker Anthropic was reported earlier this month to be discussing a new chipmaking partnership with Samsung.
Asked about the Frozen v2 report, Google neither confirmed nor denied it, telling TechCrunch only that its teams are constantly researching and experimenting with new innovations, and that co-designing hardware and software is vital to optimizing its systems for real-world workloads.
Market Reaction Ahead of Earnings
Investors responded quickly. TechCrunch reports Alphabet’s stock climbed roughly 3% Monday morning following the report, while CNBC notes shares closed 1.51% higher for the day.
The reaction comes against a mixed backdrop for Google’s AI ambitions: last week, Google delayed its newest Gemini model after it fell short of internal performance goals, particularly in coding capability.
TechCrunch notes investors have also worried about the scale of Alphabet’s AI spending, with the company saying earlier this year it plans to invest between $180 billion and $190 billion in its AI buildout.
Against that backdrop, news of a chip promising significantly greater efficiency appears to have reassured shareholders ahead of Alphabet’s second-quarter earnings report, expected Wednesday, July 22.
Source: Google Plans New ‘Frozen’ Chip to Run Its AI Models Much More Efficiently
