The Frozen v2 chip bakes Gemini architecture directly into silicon to boost efficiency by 6 to 10 times over current TPUs. Google targets a 2028 release for the hardware. This integration slashes inference costs. The move gives Google a pricing advantage over rivals like OpenAI and Anthropic in the competitive LLM market.