Google is developing Frozen v2, a server chip that integrates Gemini architecture directly into silicon. Internal sources claim the hardware will be 6 to 10 times more efficient than current TPUs. Scheduled for 2028, the chip aims to slash inference costs. This hardware advantage would allow Google to undercut the pricing of OpenAI and Anthropic.