Google Plans New ‘Frozen’ Chip to Run Its AI Models Much More Efficiently - The Information

What the report says

The Information reported that Google is developing a new server chip, referred to in the headline as “Frozen,” designed to make its Gemini artificial intelligence models run more efficiently when delivered to users. Citing two people with direct knowledge of the work, the report says the chip would directly incorporate the blueprint of Google’s Gemini model, rather than relying only on more flexible, general-purpose AI hardware.

Because the full article is not publicly available in the supplied material, the reported timeline, technical specifications, manufacturing partners and deployment plans are not confirmed here. The central development, as described by The Information, is that Google is exploring a tighter connection between its AI model architecture and the silicon used to serve that model at scale.

The move would matter because the cost of running AI services has become a major issue for large technology companies. Serving chatbot responses, search features and enterprise AI tools requires substantial computing capacity, and even small efficiency gains can affect operating costs, latency and energy use across massive data-center fleets.

As broader context, Google already has a long history of building custom AI chips, including its Tensor Processing Units, for internal workloads and cloud customers. A more model-specific chip would represent a further step toward specialized infrastructure, reflecting an industrywide push by major AI developers to control performance, cost and supply chains as demand for generative AI grows.