Google targets 2028 rollout for Gemini server chip

TL;DR
Google is developing a server chip for its Gemini AI model, with The Information reporting the chip could begin deployment in 2028. Bloomberg said the chip, referred to as Frozen v2, is separate from Google's existing tensor processing units (TPUs) and is designed to reduce data movement across the system to improve responsiveness.

Google is developing a server chip for Gemini that could begin deployment in 2028, according to a report by The Information cited by Bloomberg. Bloomberg reported the chip is called Frozen v2 and is separate from Google's existing tensor processing units, or TPUs, which are chips designed for AI workloads.

Bloomberg reported the new chip is intended to reduce the amount of data moving through the system in order to improve response times for Gemini. A Google Cloud spokesperson told Bloomberg that researching new solutions is routine and that not every project reaches production, while also noting the benefits of developing hardware and software together.

Bloomberg had previously reported that Google CEO Sundar Pichai said the company is preparing to make its TPUs available to a select group of customers for their own data centres. The report placed Google's chip work alongside other recent AI infrastructure moves, including Meta Platforms using Amazon Web Services Graviton processors, OpenAI introducing its first in-house chip with Broadcom, Amazon using Trainium chips with Anthropic, and Anthropic holding early talks with Samsung on producing its own chip.

Related questions

• What is Google's Frozen v2 chip for Gemini?
• When could Google deploy its new Gemini server chip?
• How is Frozen v2 different from Google's existing TPUs?
• What other companies are developing custom AI chips?