Google embeds Gemini architecture directly into silicon: Frozen v2 chip announced
Google has officially confirmed plans to release a specialized chip called Frozen v2. This is a processor that integrates key elements of the Gemini language model architecture directly into the silicon substrate. This is a fundamentally new approach: instead of relying solely on universal tensor processing units (TPUs), the company is creating hardware tailored to a specific model.
From an engineering perspective, this solution allows for a radical reduction in the volume of data moved and the number of computational operations when processing requests. According to Google developers' estimates, the new chip will be able to process six to ten times more tokens per unit of energy consumed compared to the company's current AI accelerators. This is a colossal leap in efficiency.
Complement, Not Replacement
It is important to emphasize: Frozen v2 is not intended as a replacement for the TPU line, but as a complement to it. The project is partly experimental — Google does not plan to produce it in the same volumes as universal accelerators. However, this particular chip could be the key to solving the acute problem of computing power shortages. As reported, due to a lack of resources, Google Cloud has previously been forced to turn down some deals with external clients.
Limitations and Timeline
There is also an important limitation: the chip will only work effectively with future versions of Gemini models if Google maintains the basic architecture. This means Frozen v2 is not a universal solution, but a deeply specialized tool. The chip is scheduled to be put into operation in 2028.
Market Reaction and Context
On July 20, following the emergence of information about the project, Alphabet (GOOG) shares rose by 1.5% on the Nasdaq. However, the positive market sentiment does not negate the serious problems facing Google's AI division. The launch of the next version, Gemini 3.5 Pro, is delayed, and the company has lost four leading researchers who left for competitors (Anthropic and OpenAI). Meanwhile, Chinese models continue to strengthen their positions: they already account for up to 46% of all tokens processed by American companies.
My analysis. The Frozen v2 project is a strategic move by Google in the race for efficiency. In a context where scaling models is hitting energy and computational limits, integrating architecture into silicon is the only path to sustainable growth. However, delays in the release of Gemini 3.5 Pro and the outflow of key personnel are alarming. If Google cannot synchronize the development of software and hardware, the market could change beyond recognition by 2028.