Alphabet Designs 2028 Frozen v2 Chip for 6-10x More Efficient Gemini AI
Updated
Updated · TechCrunch · Jul 20
Alphabet Designs 2028 Frozen v2 Chip for 6-10x More Efficient Gemini AI
3 articles · Updated · TechCrunch · Jul 20
Summary
Frozen v2, a server chip Alphabet is designing for a 2028 release, could make Gemini models six to 10 times more power-efficient by generating more tokens per unit of energy.
Google did not confirm the project directly, but said it is continually testing hardware-software designs as part of a full-stack strategy to optimize real-world AI workloads.
That efficiency push comes as AI companies face compute shortages, rising scrutiny of AI spending and pressure to reduce dependence on Nvidia's dominant chips.
Alphabet has said it plans to spend $180 billion to $190 billion this year on its AI buildout, and the report helped reassure investors ahead of earnings, lifting the stock about 3% Monday.