The rise of the Chinese AI company DeepSeek is causing panicsome of which may be unfounded, experts say.Jaap Arriens/NurPhotoGetty ImagesBy Andrew R. Chow and Billy PerrigoJanuary 30, 2025 2:56 PM ESTThis week, leaders across Silicon Valley, Washington D.C., Wall Street, and beyond have been thrown into disarray due to the unexpected rise of the Chinese AI company DeepSeek. DeepSeek recently released AI models that rivaled OpenAIs, seemingly for a fraction of the price, and despite American policy designed to slow Chinas progress. As a result, many analysts concluded that DeepSeeks success undermined the core beliefs driving the American AI industryand that the companies leading this charge, like Nvidia and Microsoft, were not as valuable or technologically ahead as previously believed. Tech stocks dropped hundreds of billions of dollars in days.But AI scientists have pushed back, arguing that many of those fears are exaggerated. They say that while DeepSeek does represent a genuine advancement in AI efficiency, it is not a massive technological breakthroughand that the American AI industry still has key advantages over Chinas.Its not a leap forward on AI frontier capabilities, says Lennart Heim, an AI researcher at RAND. I think the market just got it wrong.Here are several claims being widely circulated about DeepSeeks implications, and why scientists say theyre incomplete or outright wrong.Claim: DeepSeek is much cheaper than other models.In December, DeepSeek reported that its V3 model cost just $6 million to train. This figure seemed startlingly low compared to the more than $100 million that OpenAI said it spent training GPT-4, or the few tens of millions that Anthropic spent training a recent version of its Claude model.DeepSeeks lower price tag was thanks to some big efficiency gains that the companys researchers described in a paper accompanying their models release. But were those gains so large as to be unexpected? Heim argues no: that machine learning algorithms have always gotten cheaper over time. Dario Amodei, the CEO of AI company Anthropic, made the same point in an essay published Jan. 28, writing that while the efficiency gains by DeepSeeks researchers were impressive, they were not a unique breakthrough or something that fundamentally changes the economics of LLMs. Its an expected point on an ongoing cost reduction curve, he wrote. Whats different this time is that the company that was first to demonstrate the expected cost reductions was Chinese.To further obscure the picture, DeepSeek may also not be being entirely honest about its expenses. In the wake of claims about the low cost of training its models, tech CEOs cited reports that DeepSeek actually had a stash of 50,000 Nvidia chips, which it could not talk about due to U.S. export controls. Those chips would cost somewhere in the region of $1 billion. It is, however, true that DeepSeeks new R1 model is far cheaper for users to access than its competitor model OpenAI o1, with its model access fees around 30 times lower ($2.19 per million tokens, or segments of words outputted, versus $60). That sparked worries among some investors of a looming price war in the American AI industry, which could reduce expected returns on investment and make it more difficult for U.S. companies to raise funds required to build new data centers to fuel their AI models.Oliver Stephenson, associate director of AI and emerging tech policy at the Federation of American Scientists, says that people shouldnt draw conclusions from this price point. While DeepSeek has made genuine efficiency gains, their pricing could be an attention-grabbing strategy, he says. They could be making a loss on inference. (Inference is the running of an already-formed AI system.)On Monday, Jan. 27, DeepSeek said that it was targeted by a cyberattack and was limiting new registrations for users outside of China.Claim: DeepSeek shows that export controls arent working.When the AI arms race heated up in 2022, the Biden Administration moved to cut off Chinas access to cutting edge chips, most notably Nvidias H100s. As a result, Nvidia created an inferior chip, the H800, to legally sell to Chinese companies. The Biden Administration later opted to ban the sale of those chips to China, too. But by the time those extra controls went into effect a year later, Chinese companies had stockpiled thousands of H800s, generating a massive windfall for Nvidia.DeepSeek said its V3 model was built using the H800, which performs adequately for the type of model that the company is creating. But despite this success, experts argue that the chip controls may have stopped China from progressing even further. In an environment where China had access to more compute, we would expect even more breakthroughs, says Scott Singer, a visiting scholar in the Technology and International Affairs Program at the Carnegie Endowment for International Peace. The export controls might be working, but that does not mean that China will not still be able to build more and more powerful models.And going forward, it may become increasingly challenging for DeepSeek and other Chinese companies to keep pace with frontier models given their chip constraints. While OpenAIs GP4 trained on the order of 10,000 H100s, the next generation of models will likely require ten times or a hundred times that amount. Even if China is able to build formidable models thanks to efficiency gains, export controls will likely bottleneck their ability to deploy their models to a wide userbase. If we think in the future that an AI agent will do somebodys job, then how many digital workers you have is a function of how much compute you have, Heim says. If an AI model cant be used that much, this limits its impact on the world.Claim: Deepseek shows that high-end chips arent as valuable as people thought.As DeepSeek hype mounted this week, many investors concluded that its accomplishments threatened Nvidias AI dominanceand sold off shares of a company that was, in January, the most valuable in the world. As a result, Nvidias stock price dropped 17% and lost nearly $600 billion in value on Monday, based on the idea that their chips would be less valuable under this new paradigm.But many AI experts argued that this drop in Nvidias stock price was the market acting irrationally. Many of them rushed to buy the dip, resulting in the stock recapturing some of its lost value. Advances in the efficiency of computing power, they noted, have historically led to more demand for chips, not less. As tech stocks fell, Satya Nadella, the CEO of Microsoft, posted a link on X to the Wikipedia page of the Jevons Paradox, first observed in the 19th century, named after an economist who noted that as coal burning became more efficient, people actually used more coal, because it had become cheaper and more widely available.Experts believe that a similar dynamic will play out in the race to create advanced AI. What we're seeing is an impressive technical breakthrough built on top of Nvidia's product that gets better as you use more of Nvidia's product, Stephenson says. That does not seem like a situation in which you're going to see less demand for Nvidia's product.Two days after his inauguration, President Donald Trump announced a $500 billion joint public-private venture to build out AI data centers, driven by the idea that scale is essential to build the most powerful AI systems. DeepSeeks rise, however, led many to argue that this approach was misguided or wasteful.But some AI scientists disagree. DeepSeek shows AI is getting better, and its not stopping, Heim says. It has massive implications for economic impact if AI is getting used, and therefore such investments make sense.American leadership has signaled that DeepSeek has made them even more ravenous to build out AI infrastructure in order to maintain the countrys lead. Trump, in a press conference on Monday, said that DeepSeek should be a wake-up call for our industries that we need to be laser-focused on competing to win.However, Stephenson cautions that this data center buildout will come with a huge number of negative externalities. Data centers often use a vast amount of power, coincide with massive hikes in local electricity bills, and threaten water supply, he says, adding: We're going to face a lot of problems in doing these infrastructure buildups.More Must-Reads from TIMEL.A. Fires Show Reality of 1.5C of WarmingBehind the Scenes of The White Lotus Season ThreeHow Trump 2.0 Is Already Sowing ConfusionElizabeth Warrens Plan for How Musk Can Cut $2 TrillionWhy, Exactly, Is Alcohol So Bad for You?How Emilia Prez Became a Divisive Oscar FrontrunnerThe Motivational Trick That Makes You Exercise HarderZelenskys Former Spokesperson: Ukraine Needs a Cease-Fire NowWrite to Billy Perrigo at billy.perrigo@time.com