China’s DeepSeek AI Dethrones US’ Google, Meta, OpenAI

DeepSeek, a Chinese AI Startup, dethroned ChatGPT as the most-downloaded free software in the U.S. on Apple’s software Store. Global technology stocks dropped and were on track to lose billions of dollars in market capitalization.

Later on Monday, DeepSeek announced that it would temporarily suspend new registrations “due to large-scale malicious attacks” on its services, but existing users could continue to check in as usual.

According to tech leaders, analysts, investors, and developers, the hype — and the associated fear of falling behind in the ever-changing AI hype cycle — may be justified. Especially in the era of the generative AI arms race, when IT behemoths and startups alike are racing to avoid falling behind in a market expected to reach $1 trillion in sales within a decade.

What is the definition of DeepSeek?

DeepSeek was created in 2023 by Liang Wenfeng, a co-founder of High-Flyer, a quantitative hedge fund focused on artificial intelligence. The AI startup reportedly grew out of the hedge fund’s AI research unit in April 2023 to focus on large language models and achieving artificial general intelligence, or AGI — a branch of AI that equals or surpasses human intellect on a wide range of tasks, which OpenAI and its rivals say they are rapidly pursuing. According to Jefferies analysts, High-Flyer continues to own and fund DeepSeek in its entirety.

DeepSeek’s hype began to build earlier this month, when the startup published R1, a reasoning model that competes with OpenAI’s o1. It’s open-source, which means that any AI developer may use it, and it’s risen to the top of app store and industry leaderboards, with customers applauding its performance and reasoning abilities.

When queried about certain topics, it has limits, as do other Chinese chatbots: When queried about some of China’s leader Xi Jinping’s policies, DeepSeek apparently guides the user away from similar questions.

Another major point of discussion: DeepSeek’s R1 was built despite the United States’ three-year ban on chip exports to China. Estimates fluctuate as to how much DeepSeek’s R1 costs and how many GPUs were used. According to Jefferies analysts, a recent version had a “training cost of only US$5.6 million (assuming US$2/H800 hour rental cost).” This is less than 10% of the cost of Meta’s Llama.” Regardless of the particular numbers, publications concur that the model was constructed for a fraction of the cost of rival models from OpenAI, Anthropic, Google, and others.

As a result, the AI sector is filled with concerns, including whether the industry’s increasing number of astronomical fundraising rounds and billion-dollar valuations are necessary — and whether a bubble is set to burst.

Nvidia shares dropped about 17% at Monday’s market closing, while chipmaker ASML decreased approximately 6%. The Nasdaq fell more than 3%. This week, four tech heavyweights (Meta, Microsoft, Apple, and ASML) will report results.

Raymond James analysts addressed some of the problems affecting the AI market this month, writing, “What are the investment implications?” What does it indicate about open-source versus proprietary models? Is spending money at GPUs truly a panacea? Are US export limits effective? What are the broad implications of [DeepSeek]? They could be disastrous or insignificant, but the sector is filled with skepticism and speculation.”

The Bernstein analysts stated in a Monday note that “according to the many (occasionally hysterical) hot takes we saw [over the weekend,] the implications range anywhere from ‘That’s really interesting’ to ‘This is the death-knell of the AI infrastructure complex as we know it.”

How US Companies are Responding.

Some American IT CEOs are scrambling to respond before clients move to DeepSeek’s possibly cheaper products, with Meta apparently establishing four DeepSeek-related “war rooms” within its generative AI section.

Microsoft CEO Satya Nadella wrote on X that the DeepSeek phenomenon was just an example of the Jevons paradox, stating, “As AI becomes more efficient and accessible, its use will skyrocket, turning it into a commodity we can’t get enough of.” OpenAI CEO Sam Altman tweeted a quotation attributed to Napoleon, stating, “A revolution cannot be manufactured or stopped. The only thing that can be done is for one or more of its children to steer it through wins.”

Yann LeCun, Meta’s chief AI scientist, stated on LinkedIn that DeepSeek’s success is indicative of a shift in the AI market toward open-source technology.

According to LeCun, DeepSeek has benefited from aspects of Meta’s own technology, specifically its Llama models, and that the business “came up with new ideas and built them on top of other people’s work.” Because their work is public and open source, everyone can benefit from it. “That is the power of open research and open source.”

Alexandr Wang, CEO of Scale AI, told CNBC last week that DeepSeek’s previous AI model was “earth-shattering” and that its R1 release is even stronger.

“What we’ve found is that DeepSeek… is the top performing, or roughly on par with the best American models,” Wang said, adding that the AI competition between the U.S. and China is “AI war.” Wang’s startup offers training data to major AI players such as OpenAI, Google, and Meta.

Earlier this week, President Donald Trump announced a collaboration with OpenAI, Oracle, and SoftBank to invest billions of dollars in US AI infrastructure. At the White House, Trump, SoftBank CEO Masayoshi Son, Oracle co-founder Larry Ellison, and OpenAI CEO Sam Altman unveiled the project, Stargate. Microsoft, Nvidia, Oracle, and semiconductor manufacturer Arm will be key early technological partners. They stated that they will invest $100 billion initially, followed by up to $500 billion over the next four years.

AI Evolving

DeepSeek’s success comes amid increased interest in AI agents — models that go beyond chatbots to accomplish multistep difficult tasks for users — which tech titans and startups alike are pursuing. Meta, Google, Amazon, Microsoft, OpenAI, and Anthropic have all stated that they want to construct agentic AI.

Anthropic, an Amazon-backed AI business created by ex-OpenAI research executives, has increased its technological development over the past year, and in October, the company announced that its AI agents could use computers like people to execute hard jobs. Anthropic’s Computer Use feature enables its technology to read what appears on a computer screen, pick buttons, write text, access websites, and conduct tasks using any software and real-time internet browsing, according to the startup.

The tool can “use computers in basically the same way that we do,” Anthropic’s chief science officer, Jared Kaplan, told CNBC at the time. He claims it can complete jobs with “tens or even hundreds of steps.”

Last week, OpenAI unveiled a similar service called Operator, which automates chores such as vacation planning, form filling, restaurant bookings, and grocery shopping.

The Microsoft-backed firm advertises itself as “an agent that can go to the web to perform tasks for you,” trained to interact with common web elements such as buttons, menus, and text fields. It can also offer follow-up questions to tailor the tasks it completes, such as providing login information for other websites. Users can control the screen at any time.

Article Link: www.cnbc.com