Google's Frozen v2 chip embeds Gemini model architecture into silicon with targeting 6-10x more tokens per watt than current TPUs
Google is working on a new chip called Frozen v2 and is expected to launch sometime in 2028. Instead of a generic AI accelerator, it embeds parts of Gemini's actual architecture into the silicon, cutting the calculations and data movement needed to generate a response. Projected gain is expected to be around six to ten times more efficient than Google current AI chips, based on tokens per unit of power
but it'll only work with future Gemini models if Google keeps the same underlying architecture, & Google is currently treating it as more of a trial run than a TPU-scale replacement