Google's Frozen v2 Chip to Integrate Gemini Architecture for 6-10x Efficiency

Google’s upcoming Frozen v2 chip is projected to achieve an efficiency improvement of 6 to 10 times over current TPUs, according to internal sources. Scheduled for release in 2028, this chip will integrate the Gemini architecture directly into silicon. This integration is expected to drastically reduce Google’s AI inference costs, potentially providing a competitive price advantage over rivals such as OpenAI and Anthropic. The development of Frozen v2 follows Google’s ongoing efforts to enhance AI infrastructure, as seen in previous advancements like the SensorFM trained on extensive wearable data. For AI engineers and product managers, this chip could significantly alter the cost structure of deploying AI models at scale, making it a pivotal development in the industry. The Decoder reported.