On Monday, the beginning of the Chinese artificial intelligence began with the beginning of the rival Openi’s prestigious space as the most downloaded free app in the United States. appleDrewing Chattagpat for The App Store, DPSC’s AI Assistant. Global tech stock was sold and the market was at the pace of wiping billions.
Tech leaders, analysts, investors and developers say the hype-and always changing AI hype cycle can be guaranteed fear. Especially during the generating AI arms race, where tech companies and startups are racing the same to ensure that they will not be lagging in any market. 1 trillion dollars predicted In the tax within a decade.
What is the Deep Soccer?
The DPCAC was founded in 2023 by Lang Wenfering, a co -founder of the High Flyer, a quarterly hedge fund AI. AI startup allegedly focusing on large language models more than the AI ​​research unit of the hedge fund in April 2023 and a branch of Artificial General Intelligence, or AGI -AI that equivalent to human intellect on widespread tasks Or goes beyond that, focuses on focusing on, which the openings and its rivals say they are pursuing fast. According to Jeffrez analysts, DiPsic still provides fully ownership and financing of high flight.
The echoes around Deep Sak began to steam earlier this month, when Startup released R1, the reasoning model, which rivals Openi’s O1. This is open source, which means that any AI developer can use it, and has come to the peak of app stores and Industry Leader BoardsWith, appreciating consumer performance and reasoning capabilities.
Like other Chinese chat boats, there are limits when asked about some topics: When some Chinese leader Shi Jinping’s policies were asked, for example, DiPsic allegedly Removes the user Similar lines of questioning.
Another important part of this debate: Depsek’s R1 was built despite the United States Stop the chip exports China three times in three years. The estimation is quite different about how many Dipic’s R1 costs, or how many GPUs have gone into it. Jefferiz analysts estimates that the current version “costs only US $ 5.6 million ($ 2/h800 hours of rent costing). This is less than 10 % of the cost Method‘Lama. “But regardless of the specific numbers, reports agree that the model was developed in a part of the cost of rival models through Openi, Entropic,, Google And the other.
As a result, the AI ​​sector is roaming about questions, including whether the industry’s astronomical funding round and a rising number of billion dollars are essential-and whether a bubble is about to explode. –
Shares of nvidia 11 % fell on Monday with a chipmaker asml More than 6 % below. Nes Deck more than 2 %, and four tech companies- MethodFor, for, for,. MicrosoftFor, for, for,. apple And ASML is ready to report income this week.
Raymond James analysts described some questions affecting the AI ​​industry this month, “What are the investment implications? What do you say about open source vs. proprietary models? I am throwing money really a laughter? [DeepSeek]? Well, they may be serious, or non -event, but assured, the industry is suffering from disbelief and speculation. “
Burnstein’s analysts wrote in a note on Monday that “Many (occasionally Hydrical) we saw [over the weekend,] Its implications are due to the death of the AI ​​infrastructure complex, from ‘really interesting’ to ‘really interesting’ as we know. ‘
How are US companies responding
Some US -Tech CEOs are quarreling to respond before client Deep Sek go to potentially affordable offers, allegedly started with Meta’s four -Deep See. “The fight room“Within its productive AI department.
Microsoft CEO Satya Nadella Is written On X that the Depsy trend was just one example of Jevenus Paradox, writing, “As soon as AI becomes more efficient and accessible, we will see its use Sky cricket, and turn it into an item. Can’t get enough. ” Open CEO Sam Altimon tweeted a quote attributed to Napoleon, “A revolution can neither be created nor stopped.
Yan Licon, Chief AI scientist of Meta, Is written LinkedIn, the success of Dippec is an indication of changing the tide in the AI ​​sector in favor of open source technology.
Likone writes that DiPsic has benefited from some of his Meta’s technology, namely his lama models, and that the startup has emerged with new ideas and made them the top in other people’s work. Because their work is published and open source, everyone can do it.
Scale AIC CEO Alexander Wang told CNBC last week that the last AI model of DPCAC was “Arthur Shatteration” and that the release of her R1 is even more powerful.
Wang said, “What we’ve got is that Deep Sik … is performing the top, or is equal to the best American model.” Wang’s company provides training data to key AI players, including Openi, Google and Meta.
Earlier this week, President Donald Trump announced a joint venture with Openi, Oracle and Softbank to invest billions of dollars in US AI infrastructure. The project was unveiled by Star Gate, Trump, Soft Bank CEO Masashi Son, Oracle co -founder Larry Allison and Open CEO Sam Altman at the White House. The main partners of early technology will include Microsoft, NVIDIA and Oracle as well as the arm of the semiconductor company. He said he would invest 100 billion and $ 500 billion to start in the next four years.
Appetite
News of the capacity of the dedicic also comes between AI agents growing hype-a model that moves beyond chat boats to complete multi-faceted complex tasks for a user-which giants and startups a Like Meta, Google, Amazon, Microsoft, Openi and Anthropic have all expressed their purpose in building Agent AI.
This device “can mainly use computers as we do,” Jared Kapilin, chief science officer of Anthropic, told CNBC in an interview at that time. He said it could work with “tens or even hundreds of steps”.
Openai released a similar device last week, introducing a feature called operator that automatically automate works such as planning holidays, filling forms, restaurant reservations and gourming order.
MicrosoftBaked Startup describes it as “an agent who can go to the web to perform work for you”, and added that he training to communicate with “buttons, menos and text fields Given people who use daily “use on the web daily. It can also ask its completed tasks more personal follow -up questions, such as login information for other websites. Users can control the screen at any time.
– CNBC’s Michael Bloom supported reporting.
