No one could have predicted the maelstrom that hit the U.S. tech industry when the markets opened on Monday. As if from nowhere, an unknown AI assistant developed in China, DeepSeek, became the most downloaded free app on Apple’s US App Store.
That in itself is a shocker, but the fallout was even more astonishing. Investors in the US and Europe, in an alarmed response, sent the value of leading tech firms tumbling, but those firms regained some of the losses in the following days.
NVIDIA initially dropped 16.9%, semiconductor firm Broadcom fell by 17.4%, Microsoft decreased by 2.14%, and Alphabet was down over 4%. On the other side of the pond, chip maker ASML declined by 7%, and Siemens Energy shares lost 20% of their value.
Why the Fallout?
The functionality of the DeepSeek AI Assistant is similar to well-known U.S. alternatives, such as OpenAI’s ChatGPT and Google’s Gemini. However, the new application seized attention following the release of a paper that claimed DeepSeek-V3, which powers the chatbot, cost only $5.6 million to train.
Comparably, OpenAI spent over $5 billion in 2024 alone. In a recent interview, Anthropic CEO Dario Amodei predicted that the cost of model training could reach as much as $100 billion in just three years. “There are models in training today that are more like a billion. I think if we go to ten or a hundred billion, I think that will happen in 2025, 2026, maybe 2027, and the algorithmic improvements continue apace, and the chip improvements continue apace, then I think there is in my mind a good chance that by that time we’ll be able to get models that are better than most humans at most things.”
That’s the issue at hand. The market was shaken because, on one hand, a Chinese developer claims that it can create a product comparable to ChatGPT for less than $6 million. Meanwhile, a leader in the U.S. industry is forecasting training costs to exceed $100 billion.
What It Means
There’s a lot to unpack here. First and foremost, the market destabilization caused by DeepSeek serves as a reminder that perceived dominance does not always align with reality. Some U.S. tech firms, bolstered by a close relationship with the federal government, have been particularly confident in their position, especially considering the success of ChatGPT. These companies are viewed as the perceived leaders in the race for AI dominance.
The U.S. government has further solidified this position by making it difficult for Chinese companies to compete, implementing export controls on advanced semiconductors and AI chips, and introducing strict licensing requirements for AI chip exports. Yet, what consumers want is the product that works the best.
AI Agent & Copilot Summit NA is an AI-first event to define opportunities, impact, and outcomes with Microsoft Copilot and agents for mid-market and enterprise companies. It takes place March 17-19 in San Diego, CA. Register now to attend.
AI can inject incredible force and power into underperforming economies. Just look at the UK and how desperately the recently elected government is trying to make AI investment its leading economic growth strategy.
Instead of squabbling, companies should learn from each other and provide viable alternatives; the market is big enough for them to do so. We might be a bit early to think about collaboration, especially given the current tensions between global superpowers like China and the USA. However, tech companies in different jurisdictions can still learn from their competitors and incorporate new ideas.
Of course, there are always political questions and angles to consider. Unsurprisingly, multiple testers have concluded that the DeepSeek application aligns with Chinese governmental censorship, as it refuses to answer questions about inflammatory events such as the Tiananmen Square massacre.
However, this censorship isn’t new. And with freedom of choice, aren’t consumers welcome to cherry-pick the technologies they need for different purposes?
As OpenAI CEO Sam Altman said on X in response to the DeepSeek advances and the shocking stock market plunge, “Deepseek’s r1 is an impressive model, particularly around what they’re able to deliver for the price. We will obviously deliver much better models and also it’s legit invigorating to have a new competitor!”
However, he also added that “mostly we are excited to continue to execute on our research roadmap and believe more compute is more important now than ever before to succeed at our mission.” These sound like the words of someone that may be spooked but who remains resolute in their position and mission. Perhaps, a little spooking was just what the U.S. AI industry needed, not only from a competitive perspective, but from a resource perspective as well.
Another indicator of how quickly US tech giants are responding: Microsoft officials said Thursday that the company has added the DeepSeek model to Azure AI Foundry and GitHub, adding to the 1,800+ models already supported.
If we are going to cultivate an AI industry at the scale that is anticipated, we need to think about resources and resource management. Cutting compute spend from billions to millions, as referenced by DeepSeek’s developers, seems like a good place to start.
