Google plans new chip to run Gemini models more efficiently, the Information reports

The Information reports that Google is planning a new chip to run its Gemini models more efficiently. The report does not provide specifications, a launch timeline, expected cost savings, or details on whether the chip would be used only internally or offered more broadly.

The effort would touch Alphabet's AI infrastructure strategy and the economics of serving Gemini at scale. GOOGL generated $402.8 billion in fiscal 2025 revenue, up 15.1% year over year, with a reported 32.8% net margin and diluted EPS of $10.81.

A successful custom chip could improve inference efficiency, support Gemini deployment, and reduce dependence on third-party accelerators. The counterpoint is execution risk: custom silicon requires substantial development and deployment spending, while the report gives no evidence yet of realized savings or a material change to Alphabet's earnings outlook.

The next useful markers are confirmation from Alphabet, chip specifications, deployment timing, capital-spending implications, and evidence that Gemini usage or margins are improving. Until those details emerge, the story is strategically meaningful but difficult to translate into a precise near-term position.