Google creates a revolutionary chip with Gemini architecture: a breakthrough in AI efficiency
Google is preparing to launch a specialized chip, Frozen v2, which promises to be a turning point in AI model processing. This processor embeds key elements of the Gemini architecture directly into silicon for the first time, fundamentally changing the approach to computing.
Unlike traditional solutions, Frozen v2 minimizes the volume of data moved and the number of operations required to process requests. It is not a replacement for existing tensor processing units (TPUs), but a strategic complement to them. The main goal is to alleviate the acute shortage of computing power, which, according to my data, has forced Google Cloud to turn down deals with external clients.
Impressive Numbers
Google engineers estimate that the new chip will be able to process 6 to 10 times more tokens per unit of energy consumed compared to the company's latest AI accelerators. This is a colossal leap in energy efficiency that could change the economics of cloud computing.
However, there is a nuance: Frozen v2 will only be compatible with future versions of Gemini if Google maintains the basic architecture. For now, the project is considered experimental, and mass production on the scale of universal TPUs is not planned. The chip is scheduled to be operational by 2028.
Market Context and Challenges
Against the backdrop of the announcement, Alphabet's shares rose by 1.5% on July 20, reflecting investor optimism. However, the situation in Google's AI division remains tense. The launch of Gemini 3.5 Pro is delayed, and four leading researchers have left the company, moving to competitors — Anthropic and OpenAI. Meanwhile, Chinese models are gaining momentum: up to 46% of American companies' tokens are now processed on Chinese solutions.
My analysis: Frozen v2 is not just a technical upgrade, but a strategic response by Google to competitive pressure. If the project succeeds, it could give the company a decisive advantage in efficiency, but the delay until 2028 leaves room for other players to maneuver. As China actively captures market share, Google needs to accelerate, otherwise this chip risks becoming a belated response to an already changed landscape.
Recall that earlier, the head of DeepMind proposed checking advanced AI models before release, highlighting the growing attention to safety in the industry.