NewsBite analysis · AI, Cloud, Hardware
Toms Hardware
Google reportedly developing 'Frozen v2' chip with Gemini's architecture etched into the silicon — engineers project 6 to 10 times more tokens per watt than latest TPUs Published 21 July 26
Google is reportedly developing a new chip called 'Frozen v2', which will incorporate Gemini's architecture. Early projections suggest this innovation could enable processing efficiency to increase six to tenfold in terms of tokens per watt compared to the latest Tensor Processing Units (TPUs). This could have significant implications for machine learning and AI capabilities, potentially setting a new standard in the industry for power efficiency while processing extensive datasets.
MonitorConfidence: MediumMid term horizon
Published 21 Jul 2026, 10:45 UTC1 min read
Visit original article ↗Lead image — Toms Hardware
Summary
TL;DR
- Google is developing 'Frozen v2' chip designed for improved efficiency.
- Chip will incorporate Gemini's architecture etched into the silicon.
- Projected efficiency gains of 6 to 10 times more tokens per watt than current TPUs.
- Represents a significant advancement in AI hardware technology.
- Challenges existing power efficiency benchmarks in machine learning.
Interesting facts
- If successful, this chip could reshape training paradigms for large AI models.
- Gemini's architecture could provide competitive advantages in processing power.