Why is China's DeepSeek R1 AI model shaking up the tech world?

The new AI model, called DeepSeek R1, was developed by DeepSeek, a young Chinese startup founded just a year ago. Despite its infancy, DeepSeek has already made such a remarkable breakthrough that renowned tech investor Marc Andreessen has called it “AI’s Sputnik moment,” a milestone that marks a major shift in the AI ​​race.

What makes R1 so formidable is not only its powerful performance, but also its ability to compete directly with leading models like OpenAI's GPT-4, Meta's Llama, and Google's Gemini, but at a fraction of the cost. This success has shaken the tech world and posed a major challenge to the industry giants.

DeepSeek claims that it only cost $5.6 million to run its underlying AI model, a modest figure compared to the hundreds of millions, even billions, of dollars that US tech companies pour into AI development.

This is all the more surprising given that the US has imposed strict restrictions for years aimed at cutting off the supply of high-performance AI chips to China, citing national security reasons.

That means DeepSeek has figured out how to build a powerful AI model at low cost, even using relatively limited-power AI chips, an achievement that has left the tech world in awe.

What is DeepSeek?

DeepSeek was founded in late 2023 by Liang Wenfeng, a Chinese hedge fund manager, and quickly became one of the most prominent startups in the AI ​​race. Like dozens of other startups, DeepSeek emerged with the ambition of attracting large investments to take advantage of the AI ​​wave that is reshaping the global technology industry.

Liang Wenfeng is now known as the “Sam Altman of China,” an AI pioneer who not only promotes research but also invests heavily in breakthrough technologies. His hedge fund, High-Flyer, focuses entirely on AI development, reflecting his deep belief in the future of AI.

Over the past year, DeepSeek has released several cutting-edge AI models that compete directly with major players like Anthropic and Perplexity of the US. One of the most notable releases, DeepSeek V3, has helped raise the company's profile in the tech world.

However, according to The Wall Street Journal, DeepSeek still faces major challenges, including content restrictions related to sensitive topics about the Chinese government, as well as concerns about the company's long-term viability in a fiercely competitive AI market.

Like GPT-4, Claude, Llama, or any of the other popular generative AI models, DeepSeek is a generative AI model. However, it is more accurately a collection of different AI models, each variant designed to serve specific applications.

The biggest difference between DeepSeek and OpenAI's GPT models is that DeepSeek is released under the MIT open source license, which allows for commercial use without any barriers.

This means that the entire source code and internal workings of the model are public, allowing anyone to use, customize, or build on top of the existing platform, completely free of charge.

This makes DeepSeek a disruptor in the AI ​​market in the short term and has the potential to lead to more profound corrections in the long term.

However, all of these advantages would be meaningless if DeepSeek were not truly powerful. What makes this model the center of discussion is its performance.

By AI benchmarks, DeepSeek R1 is comparable to GPT-4 and other leading models, even outperforming them in certain scenarios. But what’s really shocking is not just its performance, but its development costs, which are staggering compared to the massive investments made by Western tech giants.

The DeepSeek app has made an impressive breakthrough on the app store rankings, surpassing ChatGPT on January 27 and reaching nearly 2 million downloads.

Why is DeepSeek R1 shaking up the tech world?

AI is a technology that consumes huge amounts of energy and is very expensive to develop, to the point that US tech giants have had to acquire nuclear power companies to meet the energy needs of their AI models.

Meta previously announced it would spend more than $65 billion this year on AI development, while OpenAI CEO Sam Altman warned last year that the AI ​​industry would need trillions of dollars to develop high-powered chips to meet the massive energy demands of the data centers needed to run complex models.

So when an AI model as powerful as the ones from the US giants can be built at a much lower cost and using lower-power chips, it represents a major shift in the industry's view of the level of investment needed for AI.

The technology still faces skepticism and opposition from many quarters, but proponents assert that AI will usher in a new era for the global economy, making work more efficient and creating new opportunities for everything from research to development.

Andreessen, founder of Andreessen Horowitz, one of the leading venture capital firms in Silicon Valley (USA), called DeepSeek "one of the most amazing breakthroughs I have ever witnessed" in a post on the social networking platform X.

If the potentially world-changing power of AI can be harnessed at a much lower cost, it opens up a whole new range of possibilities, but also poses unpredictable threats to the future of our planet.

What does this mean for America?

The US believes it can use sanctions to maintain dominance over a critical technology it says underpins national security. Just a week before leaving office, former President Joe Biden moved to tighten restrictions on AI chip exports, aiming to block rivals like China from gaining access to cutting-edge technology.

DeepSeek, however, challenges this notion and threatens to undermine the US tech industry’s invincibility. While the US may have bought itself some time with its chip export restrictions, its lead in AI has been significantly eroded despite those efforts.

DeepSeek shows that blocking access to a critical technology does not necessarily mean the US will win, a powerful message for President Donald Trump, especially as he pursues an “America First” policy.

Wall Street was already nervous about the developments, with U.S. stocks bracing for a sharp selloff on the morning of January 27. Nvidia, a leading AI chip supplier whose shares have doubled in the past two years, fell as much as 12% in premarket trading.

Meanwhile, Meta, Alphabet (Google's parent company), Marvell, Broadcom, Palantir, Oracle and many other tech giants also recorded sharp declines in stock value.

Still, it’s too early to dismiss American innovation and leadership in the field. One achievement, while impressive, may not be enough to counter the years of progress that American companies have made in AI. Moreover, the likelihood of customers switching to a Chinese startup remains slim.

Keith Lerner, an analyst at US financial firm Truist, said: "The emergence of the DeepSeek model is making investors question the leadership of US companies, how much money they are spending and whether this spending is profitable or whether it is overspending."

“However, we believe that the cost of data and related factors in AI will be huge and US companies will still maintain their leadership position,” added Mr. Keith Lerner.

While DeepSeek’s cost savings are significant, their R1 model, while competitive with ChatGPT in the consumer natural language space, has yet to prove itself capable of handling ambitious industrial AI applications, which still require a huge infrastructure investment.

Giuseppe Sette, President of US AI research firm Reflexivity, said: "With abundant human resources and capital, the US remains the ideal 'home ground' and is where we expect to see the first self-improving AI."