META PLATFORMS Inc. CEO Mark Zuckerberg arrived at Meta Connect in Menlo Park, California on September 25, 2024.
David Paul Morris | Bloomberg | Getty image
Yuan CEO Mark Zuckerberg said on Wednesday that it is too early to say that Deepseek’s breakthrough AI model will have on his company and the wider scientific and technological industry.
Nevertheless, Zuckerberg poured cold water on the concept of the overall artificial intelligence expenditure of Meta, because Deepseek was said to prove that the funds and computing resources required by the tip of the cutting -edge AI than before I imagined before the imagination. It is much less.
Zuckerberg said: “Now the trajectory of this infrastructure and capital expenditure means that it may be too early.” “There are many trends in one here.”
this Technical market The AI ​​model created by Deepseek this week is rolled up, this is a AI laboratory based on Hangzhou It is related to the orgasm of the China Hedie Fund. Deepseek claims that it took only two months to develop its development, less than $ 6 million R1 big language modelEssence The message causes NVIDIA shares tank 17 % on Monday is that the company does not need many chip manufacturer’s graphics processing unit for its artificial intelligence projects on Monday.
Zuckerberg discussed Deepseek on Wednesday with analysts on Wednesday as part of Meta Revenue in the fourth quarterEssence The company reports that sales have increased by 21 % year -on -year to $ 48.39 billion, which has advanced Wall Street’s expectations.
Meta’s stock rose nearly 2 % on Monday, indicating that some investors believe that the cost and benefits of Deepseek may mean that social media companies may be able to create a strong AI system with less money. Yuan announced the plan last week Investment 60 billion to 65 billion US dollars This year as part of AI promotion. The company said on Wednesday that its total expenses in 2025 will be between $ 114 billion and $ 119 billion.
Zuckerberg said that Meta was still digesting some of the feats of DeepSeek, and his team hoped that some of the progress was implemented for his AI project.
“This is a part of this way of operation, whether it is a Chinese competitor or not.”
Zuckerberg said that although the company may eventually need less computing resources to train these powerful models, Zuckerberg said that when the software is actual operation and execution, a large number of server power service quality may be required. “
Zuckerberg said: “We serve all these people because we provide services to many people, which will be expensive.”
Zuckerberg said Meta will continue to learn from technologies such as DeepSeek and release their own open source and free LLAMA AI models, which can also promote the industry.
He said: “I continue to think that over time, a large amount of investment on Capex and Infra will become a strategic advantage.” “We may learn at some time, but I just think that it is too early to call it too early. In this regard, I dare to bet on and to build this infrastructure will be a main advantage. “
Zuckerberg also said that the emergence of Deepseek confirmed Meta’s commitment to AI open source methods.
He said: “There will be an open source standard in the world.” “For our own national advantages, it is important to be the standard of the United States.”
watchThe Meta beating on the top and bottom lineEssence
