AI Briefing
KO

Google's New Chip for Gemini

·2026.07.21 09:00

Key point

Google is developing a 'Frozen v2' server chip that could be up to 10 times more efficient than existing AI chips, targeting a 2028 release.

Details

Google is internally developing a new server chip called Frozen v2. It is targeting a 2028 release, and has been reported to be 6-10 times more efficient than existing AI chips in terms of tokens generated per unit of power.

AI companies are accelerating their moves to develop their own chips to improve model operation efficiency and reduce reliance on Nvidia. OpenAI unveiled its first custom inference chip, 'Jalapeño', in June, and Anthropic is reportedly discussing a chip development partnership with Samsung.

Google's parent company Alphabet announced plans to spend $180-190 billion on AI infrastructure this year, making proof of investment efficiency a challenge. Following the Frozen v2 report, Alphabet's stock rose about 3% on Monday morning.

nvidiaopenaigoogleanthropicgeminiai-chipalphabetcustom-silicon
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

Google AI BlogGoogle Releases Gemini 3.8 Flash TTS: Voice Customization via Natural Language PromptsGoogle AI BlogGoogle Integrates New Connected Apps Like Adobe and Peloton into GeminiHacker NewsGoogle Launches First Suncatcher Orbital AI Data Center Test on October 1TLDR AIAnthropic Pacing to $100B ARR; Rumors of OpenAI Solving Navier-Stokes Spark ControversyGeekNewsDiffusion Language Model (dLLM) Research: Overcoming AR Model Speed Limits with Parallel Generation and Commercialization Trends like Mercury 2TLDR AIOpenAI and Anthropic's RSI Strategy and the Uncertainty of AI Profitability