What is DeepSeek? China’s ChatGPT rival sends Nvidia and Microsoft stocks tumbling, rattles AI giants

America’s AI industry was left reeling over the weekend after a small Chinese company called DeepSeek released an updated version of its chatbot last week, which appears to outperform even the most recent version of ChatGPT.

But it’s not just DeepSeek’s performance that is rattling U.S. artificial intelligence giants. It’s the fact that DeepSeek built its model in just a few months, using inferior hardware, and at a cost so low it was previously nearly unthinkable. Here’s what you need to know about DeepSeek?

What is DeepSeek?

DeepSeek is a Chinese artificial intelligence lab. It was founded in 2023 and is based in Hangzhou, in China’s Zhejiang province. It has released an open-source AI model, also called DeepSeek. The latest version of DeepSeek, called DeepSeek-V3, appears to rival and, in many cases, outperform OpenAI’s ChatGPT—including its GPT-4o model and its latest o1 reasoning model.

However, the idea that the DeepSeek-V3 chatbot could outperform OpenAI’s ChatGPT, as well as Meta’s Llama 3.1, and Anthropic’s Claude Sonnet 3.5, isn’t the only thing that is unnerving America’s AI experts. It’s that fact that DeepSeek appears to have developed DeepSeek-V3 in just a few months, using AI hardware that is far from state of the art, and at a minute fraction of what other companies have spent developing their LLM chatbots.

How much did DeepSeek cost to develop?

Perhaps the most astounding thing about DeepSeek is the cost it took the company to develop. According to the company’s technical report on DeepSeek-V3, the total cost of developing the model was just $5.576 million USD.

Yes, that’s million.

For less than $6 million dollars, DeepSeek has managed to create an LLM model while other companies have spent billions on developing their own. (In training just GPT-4, OpenAI reportedly spent $100 million alone, Wired noted in 2023.)

This raises several existential questions for America’s tech giants, not the least of which is whether they have spent billions of dollars they didn’t need to in building their large language models.

The high research and development costs are why most LLMs haven’t broken even for the companies involved yet, and if America’s AI giants could have developed them for just a few million dollars instead, they wasted billions that they didn’t need to.

But the fact that DeepSeek may have created a superior LLM model for less than $6 million dollars also raises serious competition concerns. When LLMs were thought to require hundreds of millions or billions of dollars to build and develop, it gave America’s tech giants like Meta, Google, and OpenAI a financial advantage—few companies or startups have the funding once thought needed to create an LLM that could compete in the realm of ChatGPT.

But if DeepSeek could build its LLM for only $6 million, then American tech giants might find they will soon face a lot more competition from not just major players but even small startups in America—and across the globe—in the months ahead.

Wasn’t America supposed to prevent Chinese companies from getting a lead in the AI race?

Yes. The Biden administration placed a number of export controls on AI technologies in the hopes that they would do just that. Some of the export controls forbade American companies from selling their most advanced AI chips and other hardware to Chinese companies. Some of Nvidia’s most advanced AI hardware fell under these export controls.

That’s why DeepSeek’s success is all the more shocking. The model was developed using hardware that was far from being the most advanced. DeepSeek trained its LLM using Nvidia’s H800 chips—a mid-range AI chip.

Despite being consigned to using less advanced hardware, DeepSeek still created a superior LLM model than ChatGPT. It is also much more energy efficient than LLMS like ChatGPT, which means it is better for the environment.

In an interview with Perplexity CEO Aravind Srinivas about DeepSeek’s breakthroughs, Srinivas told CNBC, “Necessity is the mother of invention. Because they had to figure out work-arounds, they actually ended up building something a lot more efficient.”

How have America’s AI giants reacted to DeepSeek?

With shock and concern. At the World Economic Forum in Davos, Switzerland, on Wednesday, Microsoft CEO Satya Nadella said, “To see the DeepSeek new model, it’s super impressive in terms of both how they have really effectively done an open-source model that does this inference-time compute, and is super-compute efficient. We should take the developments out of China very, very seriously.”

Microsoft has spent billions investing in ChatGPT-maker OpenAI.

Meta’s chief AI scientist, Yann LeCun, has a slightly different take. On Threads he stated that DeepSeek’s success shows “Open source models are surpassing proprietary ones.”

“DeepSeek has profited from open research and open source (e.g. PyTorch and Llama from Meta),” LeCun wrote. “They came up with new ideas and built them on top of other people’s work. Because their work is published and open source, everyone can profit from it. That is the power of open research and open source.”

How have investors reacted to the DeepSeek news?

With some alarm. As of the time of this writing, major AI or AI-adjacent stocks are down in premarket trading.

NVIDIA Corporation shares (Nasdaq: NVDA) are currently down over 10%. Nvidia’s success in recent years, in which it has become the world’s most valuable company, is largely due to companies buying as many of its most advanced AI chips as they can. However, if companies can now build AI models superior to ChatGPT on inferior chipsets, what does that mean for Nvidia’s future earnings?

Shares of ASML Holding N.V. (Nasdaq: ASML) were also down 9% in premarket. ASML makes the equipment needed to produce advanced AI chips.

Shares in Microsoft Corporation (Nasdaq: MSFT), OpenAI’s biggest investor, were down over 6% in pre-market.

Can I use DeepSeek?

Yep. DeepSeek can be used for free—there’s no cost to use the most advanced DeepSeek-V3, which in most tests beats ChatGPT’s o1 model. The latter costs $200 a month to use.

DeepSeek can be used for free on the web. As you can see, its interface looks no different than the interfaces of other LLMS.

You can also use DeepSeek for free on your smartphone via the dedicated DeepSeek app for iOS and Android.

And in a sign of how DeepSeek has gained so much mindshare in the AI market over the past several days, the app is now the No. 1 app in Apple’s App Store.