Google plans AI chip for Gemini models

• Google is developing a new server chip designed to integrate parts of its Gemini AI models directly into hardware.
• Alphabet shares rose 3% in early trading following reports about the planned chip.
• The chip, known internally as Frozen v2, is being developed to improve AI model serving efficiency and address computing capacity constraints.

Alphabet (NASDAQ:GOOGL) is developing a new server chip that would integrate elements of its Gemini AI models directly into hardware to improve AI processing efficiency.

The chip, internally referred to as “Frozen v2,” is expected to be deployed as early as 2028, although engineers are still finalising its design.

“Frozen v2” could be six to 10 times more efficient than Google’s latest custom AI chips based on AI tokens served per unit of power, according to reports from The Information.

Google is developing the chip as a separate project from its existing Tensor Processing Units and does not intend it to replace TPUs.

Following the announcement, Alphabet's share price was up 3% in early trading.

The company is developing the chip as demand for AI computing capacity increases, with reports saying Google Cloud has faced limits on available computing resources.

Alphabet has continued expanding its artificial intelligence infrastructure through Gemini models, custom chips and cloud computing services.