Google is working on a new AI chip designed to make Gemini more efficient
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Alphabet is designing a new server chip internally dubbed 'Frozen v2' to make its in-house Gemini models more efficient, targeting a 2028 release per a report from The Information citing anonymous sources
- Google's new chip could generate 6-10 times more tokens per unit of power than Google's existing AI chips, according to the report
- Google neither confirmed nor denied the report to TechCrunch, issuing only a generic statement about ongoing hardware-software co-design research
- Google plans to spend $180-190 billion on its AI strategy, and investors rewarded news of the efficiency gains with a roughly 3% stock climb on Monday morning ahead of the company's earnings report later this week
- OpenAI announced its first custom inference chip dubbed Jalapeño in June, while Anthropic is reportedly discussing a new chipmaking partnership with Samsung — both moves reflecting the industry's broader push to wean off Nvidia
- Nvidia's historical dominance of the AI chip market has left major AI makers dependent on its hardware, making in-house chip design both an efficiency play and a supply-chain hedge
Why it matters: Google committed $180-190 billion to AI infrastructure, and investors immediately rewarded news of a 6-10x efficiency gain with a 3% stock bump ahead of this week's earnings. The chip targets two pressure points simultaneously: Nvidia dependency and the efficiency narrative that has become crucial as AI-spend concerns have dampened market enthusiasm. A 2028 timeline is a long wait for validation, but the report alone was enough to move the stock.
