'Frozen V2': Google eyes next-generation chip to boost Gemini performance: Report

Amid intensifying competition in the tech and AI race, Google is reportedly developing a new server chip that would integrate elements of its Gemini AI model directly into the hardware to deliver AI services more efficiently, according to The Information.

The chip, informally codenamed "Frozen V2," is designed to help ease Google's AI computing capacity crunch, which has reportedly fueled internal tensions and forced Google Cloud to turn away some external customer deals. Google is aiming to deploy the chip as early as 2028, although engineers are still finalising its design and determining which parts of Gemini will be hardwired into the hardware.

Google's proposed "Frozen V2" chip is designed to permanently embed key parts of Gemini's AI architecture directly into the silicon, reducing the amount of computation and data movement required to process AI queries. According to CNBC, engineers believe the chip could deliver six to ten times more AI tokens per unit of power than Google's latest Tensor Processing Units (TPUs), making it significantly more power-efficient. Rather than replacing TPUs, Frozen V2 is expected to serve as a specialised AI accelerator, complementing Google's existing chip lineup by optimising performance for real-world AI workloads.

More from Tech

According to the report, the chip could be up to 10 times more power-efficient than Google's latest custom AI chips, based on the number of AI tokens processed per unit of power. The initiative is intended to create a new family of in-house AI chips alongside Google's existing Tensor Processing Units (TPUs).

Google's long-term AI hardware ambitions come as the company recently delayed the launch of its latest Gemini model after it reportedly failed to meet internal performance goals ahead of its planned release.