💻 Google embeds Gemini architecture into silicon
Google is developing a specialized Frozen v2 server AI chip that will partially integrate the Gemini architecture directly into the silicon. This solution will provide 6–10 times higher energy efficiency compared to current TPUs.
🌍 The shift toward a model-in-silicon architecture could radically reduce data center capital expenditures and set a new trend for deep vertical integration of software and hardware.
👤 AI assistants will become significantly faster and cheaper, which is critical for voice interfaces, but model architecture may become less flexible due to being tied to the chip.
Source 1: https://www.theinformation.com/articles/google-plans-new-frozen-chip-run-ai-models-efficiently Source 2: https://cryptobriefing.com/google-frozen-v2-chips-2028/