Alphabet Targets Efficiency With 'Frozen v2' Silicon
AI-generated image · US National Wire
Google's push for proprietary AI chips aims to curb massive AI spending and reduce reliance on Nvidia.
Alphabet is developing a new server chip, internally referred to as “Frozen v2,” to increase the operational efficiency of its Gemini AI models, as TechCrunch reported, citing The Information. The chip is expected to launch in 2028 and could potentially be six to 10 times more efficient than Google's current AI hardware when measuring tokens generated per unit of power.
While Google did not explicitly confirm the project to TechCrunch, the company stated that its “full stack approach” involves co-designing hardware and software to optimize real-world workloads. This pivot comes as Alphabet faces investor scrutiny over its planned AI expenditures, which Google stated earlier this year would range between $180 billion and $190 billion.
From a deals perspective, the move is part of a broader industry trend to mitigate dependency on Nvidia, which has historically dominated the AI chip market. Other major players are pursuing similar paths: OpenAI in June announced a custom inference processor called Jalapeño, and earlier this month reports indicated Anthropic has been in discussions with Samsung regarding a chipmaking partnership.
Market reaction to the Frozen v2 news was positive, with Alphabet's stock rising approximately 3% on Monday morning.