Crypto
Home›Crypto›Market Structure›Google develops Frozen v2 chip to run Gemini faster an…
Google develops Frozen v2 chip to run Gemini faster and cheaper
The company is targeting a 2028 deployment, with engineers projecting a six to 10 times efficiency boost versus current TPUs, as Alphabet shares rose about 3% on the news.
Google is reportedly developing a server chip called Frozen v2 that hardwires part of Gemini's architecture directly into silicon, aiming to improve efficiency for running the model.
According to Decrypt, engineers project Frozen v2 could deliver six to 10 times the efficiency of current TPUs and target a 2028 deployment, with the design meant to cut redundant calculations and reduce data shuttling during queries.
The move comes as Google faces capacity constraints serving AI demand it has already generated, including a March episode where it told Meta it could not fill the volume of Gemini compute Meta wanted, leading Meta employees to ration usage.
Decrypt also links the effort to Alphabet's AI infrastructure spending, reporting Google is investing up to $190 billion on AI infrastructure this year, while the company’s shares climbed about 3% during Monday trading ahead of Q2 2026 earnings due Wednesday, July 22.