Chinese Dibsic recently published a search paper that shocked Wall Street. The startup company claims that it has spent less than 6 million dollars to train artificial intelligence (AI) The model that matches their performance or nearly a match with US models. Relatively, Openai has spent more than $ 100 million in GPT-4.
Nafidia (Nasdaq: nvda) The shares decreased sharply on the news. Its market value decreased nearly $ 600 billion in one day, which is the largest daily loss by any company listed in history. The logic behind the accident is clear and direct: If Deepseek built a good model for Amnesty International for less money, American companies can use the same training methods to achieve similar efficiency.
Consequently, investors are concerned that American companies spend less than expected on NVIDIA graphics processing units, which are usually the most expensive element in Amnesty International’s infrastructure budgets. But the market may be exaggerated. NVIDIA shareholders got good news from Definition platforms (Nasdaq: Meta) and Microsoft (Nasdaq: msft)Two of its largest customers.
Read more to learn more.
Susa Li Financial Director Meta platforms told analysts on The fourth quarter profit call Capital expenses will increase to 66 % to $ 65 billion in 2025 to support artificial intelligence efforts and basic business. This is a material acceleration of an increase of 39 % in capital expenditures last year.
CEO Mark Zuckerberg also presented a context, as analysts told Meta’s ability to spend heavily on artificial intelligence “strategic advantage” over time. He also said that the most efficient training methods do not necessarily reduce the need for artificial intelligence chips. Instead, he believes that modern breakthroughs are an opportunity to apply more computing power to the burden of inference work “to generate a higher level of intelligence and the quality of service higher.”
Satia Nadilla, CEO of Microsoft, said not to be outperformed by Deepseek. Latest profit call: “We ourselves see great gains in efficiency in both training and reasoning for years until now. On inference, we usually saw more than 2x gain the basic performance of each devices generation and more than 10x per generation of models due to the improvement of programs.”
However, Nadila believes that the consequences will be favorable to NVIDIA. “When artificial intelligence becomes more efficient and accessible, we will see more demand,” he told analysts in the call. Nadella also posted on X: “Jevons Paradox Strikes again! Since artificial intelligence becomes more efficient and accessible, we will see its Skyroct.”
https://s.yimg.com/ny/api/res/1.2/FkBq5aLsHw1VurHjPUKInQ–/YXBwaWQ9aGlnaGxhbmRlcjt3PTEyMDA7aD04MDE-/https://media.zenfs.com/en/motleyfool.com/44fc1f4e28c6742cd450a834e5750d72
Source link