Google developing Frozen chip to hardwire Gemini into silicon for AI efficiency
Google $GOOGL is developing a new chip, internally called Frozen v2, that hardwires parts of its Gemini AI model directly into silicon, The Information reports. It could run AI models 6 to 10 times more efficiently than Google's current custom chips. Deployment is planned as earl
Google $GOOGL is developing a new server chip, informally called Frozen v2, that hardwires elements of its Gemini AI model directly into the hardware, The Information reports. The chip is separate from Google's existing tensor processing units (TPUs) and is designed to reduce the amount of data that has to move around during inference, making it faster at responding to queries. According to the report, the chip could deliver six to ten times greater efficiency than Google's current custom AI chips, measured by tokens generated per unit of power. Google reportedly plans to deploy the chip as early as 2028, though engineers are still finalizing the design and how much of the Gemini model gets hardwired into silicon. The project is also aimed at addressing a shortage of AI computing capacity inside Google that has created internal tensions and reportedly led Google Cloud to decline some deals with outside customers. If it works as intended, hardwiring a model into silicon would mark a shift from general-purpose AI chips toward hardware built around a single, semi-fixed model architecture, a bet that Google's Gemini design has stabilized enough to justify locking it into custom silicon years in advance.