Google's "Frozen v2" chip reportedly bakes Gemini's architecture directly into silicon for efficiency gains
Quick Answer
Google's new 'Frozen v2' chip integrates the Gemini AI model architecture into silicon, promising 6 to 10 times the efficiency of current TPUs.
Quick Take
Set for deployment in 2028, it aims to enhance internal AI compute capacity while potentially offering a competitive edge against OpenAI and Anthropic.
Key Points
- Frozen v2 could be 6 to 10 times more efficient than existing TPU chips.
- The chip embeds parts of the Gemini model architecture directly into hardware.
- Deployment is planned for 2028, with a smaller production volume than TPUs.
- Frozen v2 aims to optimize Google's internal AI compute capacity.
- If successful, it could help Google capture market share from competitors.
DeepSignal Analysis
What happened
Google is developing a new chip named 'Frozen v2' that integrates the Gemini AI model architecture into its silicon. This chip is expected to be 6 to 10 times more efficient than current TPUs and is set for deployment in 2028. Unlike TPUs, Frozen v2 is designed specifically for the Gemini architecture, potentially enhancing Google's internal AI compute capacity.
Key evidence
- The 'Frozen v2' chip is projected to be 6 to 10 times more efficient than Google's current TPU chips, according to sources from The Information.
- Frozen v2 will embed parts of the Gemini model structure directly into the hardware, differing from TPUs that support multiple models.
- Google plans to deploy Frozen v2 starting in 2028, viewing it as a test for specialized chips with a smaller production volume than its TPU line.
Why it matters
The development of Frozen v2 could provide Google with a significant advantage in the AI market by optimizing inference costs, which are crucial for profit margins. If successful, this chip may allow Google to run advanced AI models more economically, potentially increasing its market share against competitors like OpenAI and Anthropic. However, its utility may be limited to Google's internal operations due to its architecture specificity.
Source Excerpt
Google is developing "Frozen v2," a server chip that bakes the Gemini architecture directly into hardware. According to internal sources, it could be 6 to 10 times more efficient than current TPUs. Scheduled for 2028, the chip would drastically cut Google's AI inference costs and could give the company a price advantage over OpenAI and Anthropic.
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from The Decoder
See more →
An AI model programmed nonstop for 19 days on a single MirrorCode task that cost $2,600 to run
Epoch AI's MirrorCode benchmark reveals Claude Opus 4.7 as the leader with a 56% solve rate, reconstructing a 16,000-line toolkit in 14 hours. Despite this, all models tested struggle with the most complex tasks, highlighting limitations in current AI capabilities. The single task consumed $2,600 over 19 days, raising questions about cost-effectiveness in AI development.
