Google announces breakthrough Frozen v2 chip: Gemini architecture in silicon
Google is betting on hardware evolution: the corporation has officially confirmed the development of a specialized chip under the codename Frozen v2. This processor will be the first to embed Gemini architecture elements directly into the silicon substrate, rather than emulating them on universal accelerators.
The key advantage of this approach is a radical reduction in the volume of data moved and the number of computational operations required to process user requests. According to Google engineers' estimates, token processing efficiency per unit of energy consumed will increase by 6–10 times compared to the company's current AI accelerators.
It is important to understand: Frozen v2 does not replace the tensor processing unit (TPU) line but complements it. The project is designed as a tool to mitigate the acute shortage of computing power, which, according to my data, has forced Google Cloud to turn down deals with external clients. The chip will only be able to work with future versions of Gemini if the basic architecture of the models is preserved — this makes it a niche but potentially very effective solution.
Currently, Frozen v2 is considered an experimental project, and mass production on the scale of universal TPUs is not planned. Commissioning is expected no earlier than 2028. Market reaction was positive: on July 20, Alphabet (GOOG) shares rose by 1.5% following the news about the chip.
However, the context should not be forgotten. In recent months, Google's AI division has been experiencing serious difficulties: the launch of Gemini 3.5 Pro is delayed, and the departure of four leading researchers to competitors (Anthropic, OpenAI) weakens the team. Meanwhile, Chinese models are strengthening their positions — they account for up to 46% of tokens processed by American companies.
My expertise: Frozen v2 is a strategic move by Google to cement hardware superiority, but the project's success directly depends on the stability of the Gemini software ecosystem. Without it, even the most efficient chip will remain an expensive experiment. The market awaits 2028, but competitors are not standing still.