Google is building an AI chip that locks Gemini's architecture directly into silicon, trading hardware flexibility for efficiency gains of 6 to 10 times over its latest TPUs. The project, internally called "Frozen v2," was reported by The Information and later confirmed by Reuters and Bloomberg Law. Alphabet shares rose 3.7% on the news.
Instead of loading model weights onto general-purpose AI accelerators, Frozen v2 embeds Gemini's neural-network blueprint into the chip's circuitry. Engineers can refresh the model by loading new weights, but the underlying structure stays fixed, or "frozen." How much of the model gets hardwired is reportedly still under debate.
The efficiency target is striking. The Information reports the chip could deliver 6 to 10 times more tokens per unit of power than Google's latest TPU generation, the TPU 8t for training and TPU 8i for inference announced earlier this year. Deployment is targeted for as early as 2028.
Frozen v2 is a response to a capacity crunch inside Google. The Information says the squeeze has been severe enough that Google Cloud has turned away outside customers and stirred internal tensions.
Running AI models at scale is expensive, and every watt saved at data-center scale compounds into real money. The chip would also cut latency. Because the design is fixed, it can respond with very little delay, which suits real-time applications like voice assistants.
But the approach carries risk. Hardwiring model architecture into silicon means the chip can only support future Gemini versions if Google keeps its foundational architecture intact.
AI advances fast, and a chip designed around today's Gemini could look dated by 2028. Production volumes are expected to fall well short of TPU levels, and the effort is treated internally as an exploratory exercise rather than a full-scale rollout.
Google has not confirmed the project. A spokesperson said only that its teams experiment with high-efficiency ideas and that not every lab project reaches production. The broader context: Google has been designing custom silicon since the mid-2010s, with its first TPU revealed in 2018.
Amazon, Microsoft, and Meta all have their own AI chip efforts. But embedding model-specific logic into silicon goes further than any competitor has publicly announced. It signals that Google views Gemini as a long-term franchise worth defending with substantial hardware investment.
If Frozen v2 ships, rivals will have to answer.
