Google is embedding Gemini architecture directly into silicon: the Frozen v2 chip has been announced
Google is taking a decisive step toward hardware optimization of artificial intelligence. The corporation has announced the development of a specialized processor, Frozen v2, which for the first time integrates key architectural elements of its flagship Gemini model directly into the silicon substrate. This is a fundamentally new approach: instead of relying solely on universal tensor processing units (TPUs), Google is embedding the logic of AI operation directly into the hardware.
This solution drastically reduces the number of intermediate computations and the volume of data transferred when processing user requests. According to the company's engineers, the new chip will process six to ten times more tokens per unit of energy consumed compared to the latest generations of TPUs. This is not just an evolutionary improvement—it is a paradigm shift in the battle for data center efficiency.
It is important to emphasize: Frozen v2 is not intended as a replacement for the existing TPU lineup, but as a highly specialized supplement to it. The project is partly experimental, and mass production on a scale comparable to universal accelerators is not yet planned. The chip is scheduled to be deployed in 2028. However, behind this lies an acute necessity: a shortage of computing power, according to my data, has already forced Google Cloud to turn down several deals with external clients.
There is also a significant technical limitation: Frozen v2 will only be able to work effectively with future versions of Gemini if Google maintains the model's basic architecture. Any fundamental changes to the architecture may require redesigning the chip.
Against this backdrop, Alphabet (GOOG) shares rose 1.5% on the Nasdaq on July 20. However, behind the external optimism lie serious challenges. Google's AI division is going through a difficult period: the launch of Gemini 3.5 Pro is delayed, the company has lost four leading researchers who moved to competitors, and Chinese models have already captured up to 46% of tokens among American corporations.
My expert assessment: Frozen v2 is a strategic response to growing pressure from competitors and the company's own infrastructure problems. If Google manages to implement the project on time, it will give it a significant efficiency advantage, but betting on "freezing" the Gemini architecture carries the risk of technological inertia in the rapidly changing world of AI.