Google Reportedly Developing 'Frozen v2' Chip That Hardcodes Gemini Architecture into Silicon

Google is developing a next-generation AI inference chip codenamed Frozen v2 that hardcodes the Gemini large model architecture directly into silicon, according to a report from The Information published July 20. Unlike general-purpose processors that run models through traditional compute units, Frozen v2 is designed with dedicated circuits optimized specifically for Gemini's architecture.

Sources cited in the report suggest the chip achieves approximately 10x improvement in processing per unit of power consumption compared to current solutions. This efficiency gain could significantly reduce Google's infrastructure costs for running AI inference at scale across its products, from Search to Cloud services.

Alphabet shares rose in after-hours trading following the report's publication. The news was picked up by CNBC, Bloomberg, Reuters, and Yahoo Finance, reflecting strong market interest in Google's evolving AI hardware strategy. Google already has a robust custom silicon program through its TPU series, which has been deployed across its data centers for years.

The Frozen v2 approach represents a deeper level of hardware-software co-design. By embedding model architecture directly into dedicated circuits — sometimes described as "baking the model into the chip" — Google could bypass the overhead of running model weights on general-purpose compute. This is distinct from traditional ASIC approaches, as the chip's design is driven by the specific neural network architecture of Gemini itself.

If the report is accurate, Google's strategy would position it on a different AI chip trajectory from NVIDIA, whose general-purpose GPUs dominate AI training and inference today. While NVIDIA's hardware excels at training, purpose-built inference chips optimized for specific architectures can offer compelling total cost of ownership advantages at massive deployment scale.

Google has not officially confirmed the Frozen v2 development. Key signals to watch include whether the company discloses the chip at its next I/O or hardware event, and what timeline it sets for production and deployment across its data centers.