Artificial intelligence (AI) stocks have been unbreakable for nearly two years. In fact, "Magnificent Seven" giants like AI king Nvidia (NVDA 8.93%) have gotten so big that they collectively consumed over one-third of the broader S&P 500 (^GSPC 0.92%).
However, on Monday morning, Nvidia and many other large AI stocks found themselves deeply in the red. The sell-off had nothing to do with an earnings blip despite the fact that several of the "Magnificent Seven" will report earnings this week. It had to do with a new AI chatbot developed in China called DeepSeek, which was supposedly developed in a far more cost-efficient manner than some of the dominant competitors on the market. The app climbed to No. 1 on Apple's U.S. app store seemingly overnight.
Did a small Chinese start-up really just upend the entire soon-to-be $1 trillion-plus AI market?
A much cheaper ChatGPT?
DeepSeek is another AI chatbot like OpenAI's ChatGPT, which uses AI and machine learning language models to replicate human-like responses, gather information, and complete tasks like no technology ever has before. Liang Wenfeng, who runs an AI quant hedge fund called High-Flyer, founded DeepSeek in 2023, and the start-up supposedly has fewer than 200 employees.
DeepSeek has gotten rave reviews so far, and the company claims that its math, coding, and reasoning abilities are just as strong as OpenAI. Andreessen Horowitz founder Marc Andreessen tweeted that "DeepSeek R1 is AI's Sputnik moment."
What's grabbing investors' attention is how Wenfeng and his team allegedly created the AI tool with such limited resources and at a fraction of the cost of competitors. Other chatbots like ChatGPT require vast sums of energy and high-end chips like those created by Nvidia. However, due to U.S. export restrictions and the U.S. government's concerns about how China will use AI, U.S. companies like Nvidia can't sell a lot of their advanced graphics processing units (GPUs) that many believed were necessary to build AI chatbots like ChatGPT to Chinese companies.
DeepSeek claims it costs them less than $6 million to train its latest AI models, while it costs ChatGPT $100 million. DeepSeek is able to charge only $0.14 per 1 million input tokens, compared to $15 at OpenAI. Token input pricing has to do with the amount of text that goes into training the AI language models.
Why is DeepSeek a problem?
Investors are likely concerned about whether multitrillion-dollar companies like Nvidia will now be vulnerable to smaller competitors. President Donald Trump recently announced an up to $500 billion private-sector investment from AI companies like OpenAI to build AI infrastructure in the U.S. called Stargate. Meta Platforms recently announced it will spend $60 billion or more to build an AI infrastructure data center that will consume a space nearly the size of Manhattan.
Investors are probably questioning if this level of capital expenditure will be worthwhile when a small competitor like DeepSeek can train its AI models for less than $6 million and put up similar performance. Nvidia has gross margins in the mid-70s, implying significant pricing power. But if it turns out DeepSeek can just do what it did with inferior chips, will companies need to spend crazy amounts of money on Nvidia chips? Will Nvidia's moat be at risk?
NVDA Gross Profit Margin data by YCharts
There's still a lot we don't know. By mid-morning Monday, many analysts had already called into question the $6 million training costs figure cited by DeepSeek. Bernstein analyst Stacy Rasgon said this figure "does not include all the other costs associated with prior research and experiments on architectures, algorithms, or data." Tesla CEO Elon Musk said he thinks DeepSeek likely had 50,000 of the more advanced Nvidia Hopper GPUs instead of the 10,000 A100s the company claims to have used to build DeepSeek.
It will take more time to iron out the details. However, I think today's AI sell-off shows how vulnerable these large, highly valued AI stocks can be. The margin for error is slimmer when a stock trades so high, so any sudden bad news is going to lead investors to shoot first and ask questions later. Investors should be mindful of this in today's market.
