Google is developing a 6-10x more efficient 'Frozen v2' AI chip for Gemini models
要点
- 1. Alphabet is developing the 'Frozen v2' chip, which is 6-10x more efficient than current chips, for 2028 for Gemini models.
- 2. After the news was published, Alphabet shares gained about 3% in value on Monday.
- 3. Rivals like OpenAI and Anthropic are also designing their own custom chips to reduce dependence on Nvidia.
- 4. Alphabet plans to invest 180-190 billion dollars in AI infrastructure in 2026.
数字で見る
Alphabet is developing a new server chip called "Frozen v2" to run its Gemini models much more efficiently. According to The Information sources, the chip planned for use in 2028 could be 6 to 10 times more efficient in terms of tokens produced per electricity compared to current AI chips. Following the news, Alphabet shares rose by about 3% on Monday.
The company neither confirmed nor denied the news, stating that they optimize their systems by designing hardware and software together from scratch. Alphabet's plan to invest 180 to 190 billion dollars in artificial intelligence infrastructure for 2026 encourages investors in terms of efficiency. Reducing the energy spent per token reduces the operational cost of each query, potentially saving billions of dollars.
The development of Frozen v2 is part of a trend among major AI companies to design their own custom chips independent of Nvidia to reduce reliance on their capabilities. OpenAI's "Jalapeño" chip produced with Broadcom and Anthropic's manufacturing agreements with Samsung also support this industry-wide strategic shift. The co-design of hardware and software provides much greater efficiency advantages compared to general-purpose chips.
リアクション
この記事について質問
回答はこの記事のみからAIが生成します。
よくある質問
- Q: What is Frozen v2 and when will it be released?
- A: Frozen v2 is a new server chip designed by Google to run its Gemini artificial intelligence models, planned for use in 2028.
- Q: How much more efficient will the Frozen v2 chip be?
- A: According to rumors, the chip will be 6 to 10 times more efficient than current Google AI chips in terms of tokens produced per spent electrical power.
- Q: What is the reason for Google developing this chip?
- A: Google wants to design its hardware and software (Gemini) in an integrated way to reduce operational costs and decrease reliance on Nvidia for hardware supply.