Google Develops Gemini-Linked Chip

News

1. CuspAI Launches Global AI Network to Discover New Materials

2. Google Develops Gemini-Linked Chip

3. Kanishka Narayan Takes UK AI Policy Into Cabinet

4. Hackers Exploit Critical WordPress Flaws Worldwide

5. SEBI Fines CDSL Over Login Server Left Outside Its Own Security Controls

6. Companies Double Down on AI Despite Elusive ROI: KPMG

Google is developing a server chip that would embed elements of its Gemini artificial intelligence models directly into the hardware, The Information reported, as the company seeks to ease computing constraints and serve AI products more efficiently.

The Alphabet-owned company could deploy the chip, informally known as “Frozen v2,” as early as 2028. Engineers are still finalizing its design and deciding how much model information to hardware into it, according to the report.

Unlike Google’s tensor processing units, or TPUs, which are used to train and operate AI models, the Frozen project is intended to create a separate family of internally designed chips that would work alongside the company’s existing processors rather than replace them.

The chip could deliver between six and 10 times greater efficiency than Google’s latest custom AI processors, measured by the number of AI tokens served for each unit of power consumed.

The project comes as Google faces pressure on its AI computing capacity. The shortage has created internal tensions and led Google Cloud to turn down some business from external customers, The Information reported.

Embedding parts of an AI model into a chip could allow Google to reduce the computing power and energy required to generate responses. The approach would also deepen the integration between Gemini software and Google’s hardware infrastructure.

Separately, Google delayed the release of Gemini 3.5 Pro after its latest model fell short of internal targets, particularly in coding, Bloomberg News reported last week.