According to TechCrunch, citing The Information, Alphabet is developing a chip internally called Frozen v2 for release in 2028. The new AI chip could deliver 6–10 times better efficiency than Google's current processors when measured by tokens per watt. Google declined to confirm the project directly, offering only a statement about ongoing hardware and software co-design.

A 2028 target signals a long-term bet, not a quick fix. Google will likely lean harder on its existing TPU lineup while pushing Gemini into customer products before this new chip ships.

After the demo phase, the AI race looks different. Better models still matter, but the companies that win the next stretch will be the ones that can afford to run them at giant scale. That puts boring words—efficiency, power, utilization—at the center of the fight. Google has always wanted the full stack. This report says it still thinks hardware is where margin gets defended.