Google embeds Gemini architecture into silicon: Frozen v2 chip announced
Google has officially announced the development of a specialized processor, Frozen v2, which will be the next step in the evolution of its hardware for artificial intelligence. The key innovation of this chip is the integration of architectural elements of the Gemini language model directly at the silicon level. This approach fundamentally changes the logic of computation: instead of transferring huge volumes of data between memory and computing units, some operations will be performed directly on the chip, drastically reducing energy consumption and latency.
Frozen v2 does not replace the existing line of tensor processing units (TPUs), but complements it. The project aims to solve the acute problem of computing power shortages in Google Cloud. According to my data, due to a lack of resources, the company's cloud division has repeatedly denied large contracts to external clients. Google engineers estimate that the new chip will be able to process six to ten times more tokens per unit of energy consumed compared to current AI accelerators. This is a colossal leap in efficiency that could change the economics of cloud computing.
However, there are limitations: Frozen v2 will be optimized to work with future versions of Gemini, but only if the model's basic architecture remains unchanged. For now, the project is considered experimental, and mass production comparable to TPUs is not planned. The chip is expected to be deployed in 2028. Interestingly, following this news, Alphabet's shares rose by 1.5%, indicating a positive market reaction.
Meanwhile, Google's AI division is facing serious challenges. The launch of Gemini 3.5 Pro is delayed, and four leading researchers have left the company, moving to competitors. Additionally, Chinese models already account for up to 46% of tokens among American companies. Against this backdrop, Frozen v2 looks not just like a technological breakthrough, but a strategic necessity to maintain leadership.
Expert opinion: Integrating neural network architecture into hardware is a trend we are also seeing with other giants, but Google is taking a unique path here. If Frozen v2 indeed launches in 2028 with the stated specifications, it could create a serious barrier for competitors, especially in the energy-efficient computing segment. However, delays with Gemini and the talent drain raise questions: will the company manage to implement its ambitious plan before more agile players capture the market?