BEIJING, China, 1 ​April 2025 – A storm broke out in ​the world arena of artificial intelligence (AI), and in his eyes is a Chinese start-up that most Westerners had not even heard of ​six months ago. DeepSeek, an AI laboratory left High-Flyer Capital Management, ​quickly climbed from darkness to international ​visibility, topping graphics of application stores and brilliant government anxieties, corporate consulting rooms and academies. But it’s not just another striking AI ​model making waves. DeepSeek’s ​approach, architecture and support represent a tectonic change in AI development that disrupts market ​expectations and geopolitical strategies.

What is DeepSeek, ​and why is he doing the headlines now?

Founded in 2023 as an IA laboratory dedicated by the Chinese hedging fund High-Flyer Capital, DeepSeek was born out of the ambitions of Liang Wenfeng, a graduate of Zhejiang University with a knife for data-based commerce. Since its creation, DeepSeek has taken ​a different path from OpenAI and Meta tastes, choosing to build its own groups of data centers and focusing on computational efficiency. According to TechCrunch, its app ​quickly turned to the ​top of the Apple App Store and Google Play, capturing the ​interest of the user with its powerful – and surprisingly profitable – ​models. DeepSeek’s rise became impossible to ignore when its iconic models began to overcome the most established offers of open ​source leaders such as Meta Flama and only API platforms such as OpenAI GPT-4.

The launch of DeepSeek-V2 ​in the spring of 2024 ​and of V3 in December resulted in the exchange of benchmarks, particularly in the processing of natural language and image recognition. ​But perhaps the most critical moment came with the release of DeepSeek R1 in January 2025, a model based ​on reasoning designed not only ​to process information, but ​to think about it. The self-control capabilities of R1 make it very attractive for scientific, mathematical and technical tasks. However, according to the New York Times, his censorship of sensitive issues has also raised concerns about his ​alignment with Chinese ideologies.

Is DeepSeek a ​threat ​to American dominance in AI?

This ​issue is at the centre of global discussions, especially in Washington. ​The United States has ​long been leading AI’s career through companies like OpenAI, Meta, Google and Nvidia. But the rapid growth ​of DeepSeek, along with an efficient and cost-effective infrastructure, requires a reassessment. ​As stated in the investor’s Daily Business, DeepSeek models have achieved results close to parity or performance at several industry benchmarks, although they have access to ​less powerful chips such as the ​Nvidia H800 (due to US export restrictions).

CTO Palantir Shyam Sankar ​did not measure the words ​by going to Wall Street: “We are at war with China. We are in an AI ​arms race.”

This sentiment echoes throughout the U.S. defense ​and intelligence ​communities, especially as DeepSeek’s tools begin appearing in commercial and government ​ecosystems worldwide.

What makes ​DeepSeek technology different?

DeepSeek’s advantage is not only in performance, but also in ​accessibility. ​Unlike many other Chinese startups, DeepSeek offers its models under ​license. This semi-open approach, although not entirely open to strict definitions, has encouraged widespread ​adoption and experimentation. On Hugging Face, developers have already created more than 500 ​R1 derivatives, resulting in more than 2.5 million downloads, according to CEO ​Clem ​Delangue. These models are used ​in everything, from finance to medical ​diagnosis.

But the ​real differencer is the efficiency of DeepSeek. Unlike most US companies, ​which sell billions of infrastructure, DeepSeek claims to have made significant progress in ​computer rationalization. While some experts question these claims, the results, particularly the affordability of R1 and ​V3, have forced competitors to rethink their pricing strategies. ByteDance and Alibaba, two giants in ​China’s technological space, responded by reducing the cost of their own models, some even being ​made freely accessible.

How does the West react to DeepSeek’s rise?

With admiration and ​concern. Microsoft added DeepSeek to its Azure AI foundation, indicating the companies’ interest in implementing the models. Meta and OpenAI have ​made efforts to reassure investors who are still at the ​forefront of AI. But behind the scenes, concerns about data confidentiality, censorship and foreign influence lead to concrete actions. Reuters confirmed that several U.S. government ​agencies, including NASA and the Navy, have ​banned DeepSeek from federal aircraft. New York, Texas and ​Virginia have imposed similar ​state-level prohibitions.

At the political ​level, ​there is a ​growing refrain that calls for stricter regulation of AI foreign platforms. At a press conference in March, President Donald Trump described DeepSeek ​as ​”a ​wake-up call” and stressed the need to invest in US infrastructure. His administration’s Stargate project aims to ensure that the United ​States retains its technological superiority. ​However, critics like James Ooi of Tiger Brokers warn that DeepSeek’s low-cost model can make these investments ineffective in the long term.

Is it “Sputnik Moment” from China to AI?

Many think so. With reference to the launch in 1957 ​of the Soviet satellite Sputnik, which led the United States to the space race, DeepSeek is considered a period of calculation. As Ooi said, “IA’s career is not just a matter of efficiency, ​it’s an area.”

Comparisons to the space race are not ​unfounded. The impact on Nvidia’s stock ​alone ​— a staggering $600 billion drop in January ​— speaks to the disruptive potential of ​DeepSeek’s entry.

And it ​is ​not only ​America that ​takes note. Russia’s largest bank, Sberbank, has ​partnered with Chinese researchers ​to develop DeepSeek-inspired AI tools. This collaboration, reported ​by Reuters, can give ​a ​new life ​to ​the Russian-sanctioned technology sector by extending the AI from China to the global South. According ​to the Foreign Affairs Council, this partnership offers mutual ​benefits: Russia has access to low-cost innovation, and China has an ​influence on emerging markets.

What does DeepSeek mean for IA ​Free Governance?

Beyond ​trade disruption, DeepSeek is reshaping ​the discussions ​on governance. According to the Atlantic Council, the draft Model Law of China AI and its AI security governance framework indicate a ​change in acceptance of the principles of free source, at ​least superficial. Section 19 of the Bill encourages open source ecosystems and reduces the liability ​of proponents ​who meet government standards. However, this is accompanied by caverns: models ​still have to be aligned with basic ​socialist values, and ​censorship remains ​deeply rooted.

This dual system, open to developers and users, poses a threat to ​democratic ​standards. ​The United States and the European Union, which have defended ​digital rights, ​are now facing technological but also ideological challenges. While companies like Meta have launched open weight models like Llama, the real ​legal frameworks that support open source AI remain fragmented. On the other hand, China codifies its ambitions, making its opening version the world standard for export.

Can the West ​compete freely?

Not yet, but you have ​to. Although EU ​rules, such ​as the Artificial Intelligence ​Act, ​recognize open source models, they do not ​encourage their proliferation. ​Similarly, the United States has ​not enacted any legislation specifically to support open access AI. The previous ​executive order of the ​Biden administration had made gestures to open up models, but under President Trump, this order has ​been revoked, ​and no new directive has ​yet been implemented.

Atlantic Council experts recommend a transatlantic initiative to establish governance standards around open source AI. This would involve mobilizing institutions such ​as ​the Global Partnership for AI (GPAI) and funding models that respond to democratic values. If ​these efforts were successful, the global AI ecosystem ​could remain open, secure ​and consistent with the principles of free expression and human rights.

Bets are high. Without decisive action, the Chinese model, powerful, cheap and ideologically filtered, can become the standard ​of fact. As the West hastened to catch up, DeepSeek has ​already moved on to the next iteration. It remains to ​be seen whether the US and the EU ​can respond in ​time.