On Monday, the Chinese startup for artificial intelligence Deepseek took over the coveted place of rival Openai as the most downloaded free app in the USA AppleApp Store, dethroning chatt for Deepseek's AI assistant. Global tech shares sold and were up to date to extinguish billions of market capitalization.
Later on Monday Deepseek said that the registrations of the users would temporarily restrict “due to malicious attacks” to his services, although existing users can register as usual.
Tech executives, analysts, investors and developers say that the hype and the resulting fear of being able to be in the constantly changing AI hype cycle could be retrieved. Especially in the time of the generative AI desert arms, in which Tech giants and startups run equally to ensure that they do not predict sales of 1 trillion US dollar in a market within a decade.
What is Deepseek?
Deepseek was founded in 2023 by Liang Wenfeng, co-founder of High-Flyer, a quantitative hedge fund that focuses on AI. The KI startup emerged in April 2023 from the KI research unit of the hedge fund and its rivals say that they quickly pursue. According to Jefferies analysts, Deepseek is still completely owned and is financed by high-flyers.
The enthusiasm for Deepseek began at the beginning of this month when the startup R1 released. It is open source, which means that every AI developer can use it and has put himself at the top of the app stores and industry and industry grabs, with users praising their performance and argumentation functions.
Like other Chinese chatbots, it has its limits when asked about certain topics: If he is asked about some of the guidelines of the Chinese leader Xi Jinping, Deepseek, for example, controls the user of similar questions of the survey.
Another important part of the discussion: Deepseeks R1 was built, although the US chip exports curb three years to China in three years. Estimates differ exactly how much Deepseeks R1 costs or how many GPUs have come into it. Jefferie's analyst estimated that a recently carried out version “had training costs of only $ 5.6 million (assuming 2 USD/H800 hours -lessons). That is less than 10% of the costs of Meta'S Lama. ” Google and other.
As a result, the AI sector is flooded with questions, including the question of whether the increasing number of astronomical financing rounds and billion dollar ratings is required-and whether a bubble is about to outbreak.
Read more CNBC reporting about KI
Shares of Nvidia With chipmaker, almost 17% of the market fell in the closure on Monday ASML Almost 6%. The Nasdaq fell by more than 3%. Four tech giants – MetaPresent MicrosoftPresent Apple And ASML will report profits this week.
Analysts from Raymond James have described some of the questions that the AI industry plagues this month and wrote: “What are the implications in the investments? What does it say about Open -Source models and proprietary models? Is it really a panacea For the export restrictions to throw money on GPUs? [DeepSeek]? Well, you could be bad or a non-event, but be sure that the industry is incredulous and speculated. “
Bernstein analysts wrote in a note on Monday that “after the many (occasionally hysterical) hot parts we have seen [over the weekend,] The implications range from “This is really interesting” until 'This is the death bone of the AI infrastructure complex as we know it.' “
How US companies react
Some American tech CEOs climb to react before customers switch to potentially cheaper offers from Deepseek. According to reports, Meta allegedly started four “war rooms” in Deepseek in its generative AI department.
Microsoft CEO SATYA NADELLA wrote on X that the deepseek phenomenon was only an example of the Jevons paradoxical and wrote: “When AI becomes more efficient and more accessible, we will see that the use in sprays and turn it into a goods that we simply turn Not enough. several of her children can give him a direction through DIN von Siegen. “
Yann Lecun, the chief Ki scientist of Meta, wrote about LinkedIn that Deepseek's success is a success on the change in the tides in the AI sector for the preference of open source technology.
Lecun wrote that Deepseek has benefited from some of Meta's own technology, i.e. its LAMA models, and that the startup “developed new ideas and built them up. Because their work is published and everyone can be able to benefit, everyone can benefit Them from it.
Alexandr Wang, CEO of Scale Ai, said CNBC last week that Deepseek's last AI model was “shaking earth” and that his R1 publication is even more powerful.
“We found that Deepseek … the top performance is or corresponds to the best American models,” said Wang, adding that the AI race between the USA and China is a “AI war”. The Wang company offers training data to important AI players, including Openai, Google and Meta.
At the beginning of this week, President Donald Trump announced a joint venture with Openai, Oracle and Softbank to invest billions from dollars in the USKI infrastructure. The Stargate project was unveiled in the White House of Trump, the CEO of Softbank, Masayoshi Sohn, Oracle co-founder Larry Ellison and the CEO of Openai, Sam Altman. The most important first technology partners include Microsoft, Nvidia and Oracle as well as the semiconductor company ARM. They said they would invest 100 billion US dollars for the start and up to 500 billion US dollars over the next four years.
AI develop
The news about Deepseek's skills also come in the middle of the growing hype about AI agents -models that go beyond chatbots to do multiple complex tasks for a user -who pursue tech giants and startups alike. Meta, Google, Amazon, Microsoft, Openai and Anthropic have expressed their goal to build Agentic AI.
Anthropic, the AI startup founded by ex-Openai research managers, increased its technology development last year, and in October the startup said that its AI agents could use computers like humans to do complex tasks. The use of anthropics computer use function enables its technology to interpret what is located on a computer screen, select buttons, enter text, carry out websites and perform tasks via any software and real-time internet browser, according to the Startup.
The tool can basically use computers as well as we do, “said Jared Kaplan, Chief Science Officer from Anthropic, CNBC in an interview at that time. He said it could do tasks with “ten or even hundreds of steps”.
Openai released a similar tool last week and introduced a function called Operator, with which tasks such as planning holidays, filling out forms, ordering food and food are automated.
The Microsoft-Backed startup describes it as “an agent who can go on the web to perform tasks for you”, and added that it is trained to interact with “the buttons, menus and text fields, use people every day” on the web . It can also ask follow-up questions to further personalize the tasks you do, e.g. B. Registration information for other websites. Users can take control of the screen at any time.
– Michael Bloom from CNBC contributed the reporting.
Comments are closed.