Google developing 'Frozen v2' chip to hardcode Gemini architecture with 10x efficiency

Google is planning for new chips production to run on its 'Gemini' models to boost AI.

As reported by The Information, Google is reportedly developing a new custom server chip internally codenamed "Frozen v2"—designed to incorporate elements of the Gemini model architecture directly into the hardware.

Unlike general-purpose AI chips that load models into memory and continuously shuffle data back and forth, Frozen v2 aims to hardcode portions of Gemini’s neural-network blueprint straight into the physical circuitry.

According to the reports, the chip could be 6 to 10 times more efficient than Google’s latest custom Tensor Processing Units (TPUs), measured by AI tokens served per unit of power.

The initiative is partly driven by an internal AI computing capacity constraint that has created friction and forced Google Cloud to turn away some outside customer deals.

Engineers are still finalizing the design and determining how much of the model will be permanently hardwired.

The 'Frozen' project is aimed at creating a new set of homegrown chips apart from Google's TPUs, rather than replacing them, the report said.

Bloomberg News reported last week that Google delayed the launch of its latest Gemini AI model after it fell short of internal goals, with the company working to improve its capabilities, particularly in coding.

A Google Cloud spokesperson said, "Our teams are constantly researching and experimenting with new innovations... By co-designing our hardware and software from the ground up, we ensure our systems are integrated and highly optimized."

The report informed engineers are still finalizing its design and the amount of model information that will be hardwired.

Notably, the chip is planned to complement—rather than replace—Google's existing TPU lineup, with a targeted deployment as early as 2028.