Overview
- The Information reported Monday that Google is developing a server processor code‑named Frozen v2 that would bake elements of its Gemini model design directly into hardware to speed inference.
- Engineers cited in the coverage project the chip could deliver roughly six to ten times more AI tokens per unit of power than Google’s current custom chips if the design works as expected.
- Google has not confirmed the project and told reporters its teams routinely research ideas that do not always reach production, while the company reportedly views Frozen v2 as an experimental trial rather than a full TPU replacement.
- Designers plan to hardwire parts of Gemini’s architecture but keep model weights updatable, a choice meant to balance big efficiency gains with the risk that future model changes could reduce the chip’s usefulness.
- The effort responds to a reported internal compute shortage that has forced Google Cloud to limit outside deals and prompted temporary measures such as renting external GPU capacity, and investors pushed Alphabet shares higher after the reports.