Google plans to launch the Frozen v2 chip, integrating Gemini architecture to enhance processing efficiency by 2028.
Google is set to develop a new chip named Frozen v2, which will incorporate elements of its Gemini architecture directly into the silicon, as reported by The Information.
This innovative processor aims to streamline data processing by reducing computational requirements and minimizing data transfer for user requests. It is designed to complement, rather than replace, Google's existing tensor processing units (TPUs).
The initiative is expected to help address the ongoing computing power shortage that has led Google Cloud to decline some external client contracts. Engineers at Google project that Frozen v2 will be capable of processing six to ten times more tokens per energy unit compared to the latest AI accelerators.
However, the chip will only be compatible with future iterations of the Gemini models if Google maintains its core architecture. Currently, the company regards Frozen v2 as an experimental effort and does not plan to manufacture it at the same scale as its general-purpose TPUs, with a tentative deployment target set for 2028.
Following the announcement, Alphabet's stock experienced a 1.5% increase on July 20, according to TradingView.
In recent times, Google's AI sector has encountered significant hurdles, including delays in the rollout of the upcoming Gemini 3.5 Pro and the departure of four leading researchers to rival firms. Furthermore, Chinese AI models are increasingly popular among U.S. companies, reportedly accounting for as much as 46% of the tokens they utilize.
In July, the head of DeepMind emphasized the importance of testing advanced AI models prior to their release.
