DeepSeek, a relatively unknown Chinese-developed AI model, is now the most popular app in the US Apple App Store based on hype associated with it releasing an an AI model that outpaces OpenAI’s and other companies’ models on some benchmarks while being trained on older chips at a much lower cost. It has caused Nvidia stock, which has become one of the most valuable companies in history thanks to the AI boom, to tumble and is causing more widespread panic in the U.S. market. Marc Andreessen has called this “AI’s Sputnik moment,” referring to the first Russian satellite which got ahead of and accelerated the US space program.
People in the AI space and those who follow it closely started freaking out when DeepSeek’s newest model, DeepSeek R1, was released last week, and that freakout has now seemingly captured the entire world, impacting the stock market, causing people to wonder if American companies like OpenAI and Nvidia can really dominate the AI industry, if the AI bubble is finally popping, and if this is a sign of imminent Chinese world domination and censorship. DeepSeek is particularly notable because it is free, modifiable, and less expensive to run, which has experts worried about the viability of OpenAI’s already unprofitable subscription products.
I’m going to be upfront with you here and say that 404 Media does not provide any financial advice and that if I had definitive answers to any of these questions I’d be playing the stock market instead of blogging, but in a day when the takes are going to come fast and furious my take is this: The AI industry continues to develop very fast, it’s hard to extrapolate how it’s all going to unfold based on single event, even if it’s monumental, and the fact that DeepSeek comes from China, a perceived adversary to the United States/the West is making hawks and xenophobes, and tankies foam at the mouth.
Let’s take a deep breath and start with the biggest headline, which is that Nvidia stock dropped over 12 percent early this morning, its worst performance since 2020. GPU maker Nvidia became a trillion dollar company because it is largely making the chips that power the generative AI boom. These are not only the chips that people need to generate text, images, audio etc locally on their machines, but the massive training clusters of thousands of chips that these foundational models are trained on. In July, for example, Elon Musk proudly announced that xAI started training “the most powerful AI training cluster in the world,” composed of 100,000 Nvidia H100s.
For the most part, AI companies in the US have competed on the general idea that more data and more compute creates more advanced and more “intelligent” AI models and tools. One of the general strategies, therefore, has been for companies like OpenAI, Google, and Meta to feed more and more data into their models and to compete to build incredibly expensive and resource intensive data centers. But last year, we started to see some doubts about the existing consensus on AI scaling laws, which up until recently showed that the performance of AI models improved as the size of the model, data, and compute increased, with some people, including Open AI co-founder Ilya Sutskever and Andreessen saying that AI companies are starting to see diminishing returns. Despite these doubts bubbling up, AI companies were still competing for compute, which largely means access to Nvidia’s chips. Musk wants to grow xAI’s cluster to one million GPUs eventually, and the CEO of Broadcom recently said he predicts other companies will attempt to build similarly gigantic clusters.
This demand for highly specialized and hard to produce hardware has made Nvidia incredibly valuable and critically important to building AI. Because the US government believes that the United States, not China, must be the world leader in AI, it’s also why it has introduced export restrictions that forbid Nvidia from selling its most advanced chips to Chinese companies. This is part of why you see OpenAI, Oracle, and SoftBank proposing to throw $500 billion into projects like Stargate, a massive AI data infrastructure project that they may or may not have the money for.
The main reason people are excited/scared/throwing up right now is that DeepSeek was developed and released under America’s export restrictions that prevent Chinese companies from getting the latest and most powerful Nvidia chips. As Wired explained, DeepSeek was spun out from High-Flyer, a Chinese hedge fund that originally acquired GPUs to analyze financial data, before it invested its money and resources in developing AI. That a new player in this space was able to build an AI model without access to the latest and greatest Nvidia chips (though people in China have found ways to obtain them despite restriction), using new, more efficient reinforcement learning strategies, has undermined the idea that companies like Nvidia or OpenAI have built a “moat” around their companies that will secure their lead in the AI race forever, and, by extension, has undermined the notion of American AI world supremacy. It also at least raises the possibility that a Chinese company has found a better, more efficient, and cheaper way to train AI models than any American company has discovered thus far.
As others have pointed out, it’s hard to say exactly what DeepSeek actually spent to make its model without trusting it blindly. The true cost may be hidden in ways we don’t understand, and is definitely benefiting by building on top of the very expensive research (primarily from American companies) that came before it. But if AI companies can build competitive models at a fraction of the cost on a comparatively tiny number of lesser GPUs, then much of Nvidia’s value and the billions of dollars AI companies are burning on training suddenly seems excessive and wasteful (even to AI boosters), hence the stock tumbling.
Does this mean Nvidia, OpenAI, and other AI companies are doomed? Again, this is not financial advice but the market appears to be spasming based on vibes, and definitely before we have a great understanding of DeepSeek’s impact. The most obvious rebuttal from Nvidia bag holders in this situation is that DeepSeek’s newfound efficiencies will only benefit AI incumbents. If these new methods give DeepSeek great results with limited compute, the same methods will give OpenAI and other, more well-resourced AI companies even greater results on their huge training clusters, and it is possible that American companies will adapt to these new methods very quickly. Even if scaling laws really have hit the ceiling and giant training clusters don’t need to be that giant, there’s no reason I can see why other companies can’t be competitive under this new paradigm. We should also probably hope that this is the case since it could lower the environmental impact of AI.
I don’t have a dog in this fight, but the argument I would add here is that this type of leapfrogging seems totally normal, and we have seem variations of it over the last couple of years. People love to prematurely dance on OpenAI’s grave whenever a new and shiny model is released. Meta’s Llama, France’s Mistral, and Anthropic’s Claude have all seemed like they’re getting ahead at one point or another and are favored by different users for different uses, only for another model to be released by OpenAI or another company that leapfrogs the hot new technology and makes them seem old.
The difference is that DeepSeek is from China and that a lot of people including the US government don’t like the idea of China being dominant in any arena, let alone one as supposedly consequential as AI. This is obvious given the hysteria on social media right now, the markets, and the way people are talking about DeepSeek’s censorship and the possibility that it could be tied to Chinese surveillance or the Chinese government in some way. Steven Heidel, who works at OpenAI, tweeted Sunday “americans sure love giving their data away to the CCP in exchange for free stuff,” which has gone viral and served as the basis for discussion about DeepSeek as possible surveillance software, the “new TikTok,” etc. What’s particularly notable here is that DeepSeek has been released in a way that can be run locally without an internet connection.
On various AI subreddits, where DeepSeek is all people have been talking about for days, some users are now suggesting that the conversation is being manipulated by “propaganda” from a few accounts. People have repeatedly shared screenshots on social media of DeepSeek refusing to engage with questions about Tiananmen Square and other topics subject to censorship in China, with the implication that this is the information ecosystem we’d live under if China was to dominate the AI race.
Fair enough, I suppose, but as the developers of “uncensored” AI models have been shouting from the rooftops since the beginning: any AI model that the user can’t control entirely is subject to censorship. OpenAI is a prude, and will refuse to engage users on a lot of topics, sometimes for reasons stated in OpenAI’s policy, and sometimes for reasons we’ll never understand because OpenAI is a black box.
“Why should the open-source AI running on my computer, get to decide for itself when it wants to answer my question? This is about ownership and control. If I ask my model a question, I want an answer, I do not want it arguing with me,” Eric Hartford, a developer of uncensored AI models, told me last year.
If anything, DeepSeek maps a better AI future for those concerned about censorship because it was released as an “open weights” model, meaning people could modify it to talk about Tiananmen Square and whatever else they want.
We do not know how this will all shake out, but the release of DeepSeek does seem to be a seismic moment for the AI industry. And it will certainly be used, rightly or wrongly, as a political cudgel to highlight the urgency of the competition for AI supremacy between the United States and China.
