It’s hard to overstate just how impactful DeepSeek has been. In a couple of days, it rattled the entire AI industry, shattering the aura of invincibility that OpenAI (and American tech companies in general) had built around themselves. DeepSeek’s new AI is the number one most downloaded free app on the Apple Store, and it’s not hard to see why: it rivals or exceeds OpenAI’s paid options. But that’s not even the best part: DeepSeek claims it only took two months and cost under $6 million to build. DeepSeek’s R1 model costs 1/20th of the money to run ChatGPT’s flagship o1 model. It’s also open source.

They proved Sam Altman very wrong

In 2023, Altman was asked whether a smart team with a budget of around $10 million could try to create a product rivaling those of OpenAI. Altman’s answer was (as it often is) cocky:

“Look, the way this works is we’re going to tell you it’s totally hopeless to compete with us on training foundation models you shouldn’t try, and it’s your job to, like, try anyway.” Well, they tried, and Altman is not very cocky now.

He did praise DeepSeek R1 and vowed OpenAI will produce better models, but the market and expert pundits seem less confident.

It’s not just OpenAI, everyone’s freaking out

It’s not the first impactful open-source AI model out there (the source code is freely available for anyone to view, modify, and distribute under a permissive license). Facebook’s parent company, Meta, has also opted for a similar strategy with its Llama models. But Llama releases have been relatively slow and not nearly as cheap and efficient as DeepSeek. Meta is reportedly scrambling ‘war rooms’ of engineers to figure out how DeepSeek is doing it.

To make things even more striking, DeepSeek managed this performance while using Nvidia’s less-advanced H800 chips. AI chips are hundreds or even thousands of times faster and more efficient than CPUs for training AIs, and Nvidia is by far the leader in this field. These chips were considered so important that President Biden issued the CHIPS Act, which prevented state-of-the-art US-made chips from being exported to China. Well, DeepSeek did it without these chips.

The tech-heavy Nasdaq stock index took a dive on the news, as did stock market sweetheart Nvidia. Nvidia, which had grown by almost 2000% in the past 5 years, went down 16% in one day. Overall, DeepSeek poked a $1 trillion hole in global markets, largely as it can be used, downloaded, and refined by anyone for free.

Vaibhav Srivastav, an engineer at Hugging Face (a platform for building, sharing, and deploying machine learning models, with an open-source ethos) said this must be a “humbling” moment for America’s AI giants. He added that if OpenAI had shared its resources with other players in America, technology would have advanced further. He added that the real question is why Meta’s open-source approach didn’t yield as much benefits. “What the hell is Meta doing? This was theirs to lose,” the researcher said.

Last week, Meta said it would spend over $60 billion on AI development this year alone, while Altman said the industry would need trillions of dollars in investment to support the data centers that run the complex models.

What this really means for the industry

It’s unlikely that everyone will shift to the Chinese-based product overnight. Market sell-offs often overreact, and time will tell whether DeepSeek is a true competitor for AI leadership. As Srivastav notes, it’s often not the ability of the model but how well it is integrated into other applications.

Besides, the world of AI is very much an active field of research with constant shifts and changes. Yet considering that the project was considered a “side project”, and that it’s so cheap, the results are indeed disruptive.

The AI industry was getting complacent. President Trump embraced Big Tech and Big Tech embraced him back so everything seemed to be smooth sailing — but competitive industries are rarely smooth. This will come as a wake-up call for the industry and show that the AI innovation race is far from won.

The general approach (and what kept Nvidia’s stock rally going) was the idea that you need sheer computational superiority to build better models. This was a barrier for other entrants. If you need a lot of processing power to build models, you need a lot of money for it. This is what prompted Altman’s “hopeless” comments. DeepSeek’s results suggest that’s not really the case and the door may be open for other competitors as well. The fact that the model is open-source likely means we’re going to see many spinoffs soon enough.

If DeepSeek’s approach is confirmed, Big Tech companies like Amazon, Alphabet, Meta, and Microsoft may need to pivot from current plans. They were already heavily invested in AI data centers (which involve a lot of chips and a lot of energy). If that’s not necessary, this would be good news for the planet as power-hungry data centers were already making an impact on global energy usage. It could free up some resources from Big Tech as well and enable them to focus on other types of innovation other than the “brute force” approach. But it also suggests that the fortress around Big Tech’s AI efforts may not be as impenetrable as we thought. Simply put, we’ll probably be seeing way more competitive models soon — because DeepSeek (unlike OpenAI’s products) can be downloaded and run on personal computers, even without an internet connection.

What you should know about DeepSeek

The company was established in 2023 by Liang Wenfeng — not a household name in the world of AI until now. Wenfeng is an electronic engineer who graduated from his Master’s studies in 2010. Since then, he has worked in quantitative trading and tried to integrate mathematics and AI in trading. By 2019, he was managing around $13 billion in assets. He has repeatedly spoken about how Chinese companies are too used to following and should instead try to lead markets.

DeepSeek was founded as Liang acquired 10,000 Nvidia A100 GPUs before the US government imposed AI chip restrictions on China. In China, it quickly became known for its cheap alternative to AI, prompting a price war with other rival Chinese AI companies like Bytedance. Even though DeepSeek was charging lower prices than its competitors, it was profitable, while its competitors were losing money — a sign that their models were already very efficient.

However, Liang has stated that DeepSeek is focused solely on research and has no detailed plans for commercialization, which allowed it to avoid the stringent provisions imposed by the Chinese government. This also means it has to comply with the censorship imposed by China. For instance, it will refuse to talk about the Tiananmen massacre or other topics China deems as “sensitive,” such as China’s persecution of Uyghurs or human rights in China.

Simply speaking, DeepSeek is an AI-powered chatbot, like ChatGPT. Its app description vaguely says it is designed “to answer your questions and enhance your life efficiently.” It works on the same general approach as Open AI’s O1 model, using multistep “reasoning,” currently considered the most powerful approach.

China’s Communist Party hasn’t yet commented on the achievements, but Chinese state media were jubilant, noting that Silicon Valley is “losing sleep” over DeepSeek.

“In China, DeepSeek’s advances are being celebrated as a testament to the country’s growing technological prowess and self-reliance,” says Marina Zhang, an associate professor at the University of Technology Sydney, for the BBC.

“The company’s success is seen as a validation of China’s Innovation 2.0, a new era of homegrown technological leadership driven by a younger generation of entrepreneurs.”

Consumers looking to use this product should be aware of potential privacy concerns. While censorship is confirmed on DeepSeek, there are also fears that the AI system could be used for foreign influence, disinformation, surveillance and the development of cyberweapons for the Chinese secret service. Much like TikTok, experts warn that the app “explicitly” transfers personal data to China.