Chinese AI Assistant Shocks Big Tech with Low-Cost, Efficacy

A Chinese AI assistant called DeepSeek has roared to the top of the iPhone App Store charts, sending Big Tech stocks into a tailspin and raising new questions about the investments these firms have made. How is it possible that this thing is so good?

“Experience seamless interaction with DeepSeek’s official AI assistant for free!” the app’s listing on Apple’s App Store reads. “Powered by the groundbreaking DeepSeek-V3 model with over 600B parameters, this state-of-the-art AI leads global standards and matches top-tier international models across multiple benchmarks. Enjoy faster speeds and comprehensive features designed to answer your questions and enhance your life efficiently.”

DeepSeek is made by Hangzhou DeepSeek Artificial Intelligence, or, as it’s written in the App Store, 杭州深度求索人工智能基础技术研究有限公司. So that will be confidence inspiring to any western eyes. But it appears that this is a year-old AI startup that has somehow delivered a large language model (LLM), DeepSeek-R1, that performs similar to OpenAI’s o1 model. DeepSeek-R1 is open source and licensed with the MIT License for commercial use. And you can try it out for yourself on the web right now.

I ran a few quick tests of my own, mostly C# and WPF programming topics, and I have to say, it’s quite impressive. Not only does it appear to be accurate, but the formatting of the information is excellent and it provided multiple code samples in each case.

But you shouldn’t take my word for it. Venture capitalist and former Netscape co-founder Marc Andreessen raved about DeepSeek on Twitter/X, calling it “one of the most amazing and impressive breakthroughs he’s ever seen — and as open source, a profound gift to the world.” And Jim Fan, a senior researcher at Nvidia, declared that “makes no sense … DeepSeek-R1 not only open-sources a barrage of models but also spills all the training secrets. They are perhaps the first OSS project that shows major, sustained growth of an RL [reinforcement learning] flywheel.”

What’s most troubling about DeepSeek is that it’s able to rival the best AI models provided by Big Tech and its AI partners, but at a fraction of the cost. The Wall Street Journal reports that DeepSeek’s owners spent just $4 million training its latest models, compared to the hundreds of millions or billions of dollars that companies like OpenAI and Anthropic spend on similar work. And Microsoft, notably, is spending $80 billion in its current fiscal year just to build out the infrastructure required to run these models.

DeepSeek isn’t without problems–it won’t answer sensitive questions about China, for starters–but its instant success triggered a Big Tech/AI stock sell-off on Friday. As I write this, Nvidia’s stock price is down 11.4 percent, Microsoft is down 17 percent, and Google is off 6.3 percent. I’m curious if this will have a long-term impact on these companies, which have thus far dominated this new AI era.