DeepSeek: A Shockwave for Global Artificial Intelligence

In just a few days, DeepSeek, the new Chinese generative artificial intelligence, has caused a seismic shift in this cutting-edge field. Developed at a fraction of the cost compared to its American counterparts and delivering performance that rivals—or even surpasses—that of OpenAI or Google DeepMind, this free, energy-efficient, and fully open-source chatbot is a wake-up call for the AI world.

A Revolution in Approach

$5.6 million—that’s the announced development cost of DeepSeek. A shockingly low figure considering that launching ChatGPT in 2023 exceeded $100 million and required 25,000 NVIDIA chips. In contrast, the Chinese startup claimed to have used only 2,000 less advanced chips from the same supplier, as the Biden administration imposed strict restrictions on exporting certain cutting-edge technologies to China.

Sam Altman, the creator of ChatGPT, has acknowledged it himself: “It’s an impressive model, especially considering what is offered for the price.” Other tech figures, such as Marc Andreessen—who is close to Donald Trump—even describe it as a “Sputnik Moment,” referencing the shock caused by the USSR’s first successful satellite launch in 1957.

The timing of China’s dramatic entry into the (highly exclusive) AI market is no coincidence. The launch of DeepSeek came just days after Donald Trump unveiled StarGate, a massive project set to inject $500 billion into U.S. artificial intelligence infrastructure, primarily aimed at building giant data centers.

Unlike American models that rely on sheer computational power and massive datasets, DeepSeek takes a radically different approach: adaptive “reinforcement.” Instead of absorbing astronomical amounts of information, its AI learns to search, ask questions, and refine its responses through a reward-based system.

The message is clear: AI does not necessarily require billions to finance ever more power-hungry data centers and increasingly sophisticated chips. China is proving that a more efficient and frugal approach can compete with American tech giants.

A Shock for the West

The shockwave was immediate. Within hours, the announcement of DeepSeek’s launch sent financial markets into turmoil. Shares of American tech giants, particularly NVIDIA, plummeted. The company, which holds a near-monopoly on AI-essential chips, saw its market valuation drop by 17%, translating to a loss of nearly $600 billion in just 24 hours.

As markets crashed on Wall Street, downloads of the application skyrocketed. DeepSeek accumulated millions of downloads within days on Apple and Google platforms.

In Europe, reactions were swift: Italy immediately banned the application, citing concerns over data protection. In France, the CNIL (National Commission on Informatics and Liberty) requested clarifications from China regarding how user data is handled.

Once again, the timing was anything but coincidental: this shockwave hit just before the Paris Summit on AI, where European leaders were set to debate digital sovereignty and technological independence from the United States. The irony is striking: while Europe struggles to define a more ethical and less US-dependent AI model, it is a Chinese company that has brought this vision to life.

In an interview with Le Monde, political scientist Asma Mallah summarized: “DeepSeek is inexpensive, energy-efficient, capable of functioning with smaller datasets compared to American models, and most importantly, it is open-source. Exactly the kind of AI Europe claims to advocate for!”

But does DeepSeek truly tick all the boxes of the AI model championed by Europe? Well—except for freedom of speech.

“Let’s talk about something else”: Open Data, Closed Ideas

Can we talk about everything? Not with just anyone. Unsurprisingly, an AI designed in China comes with strict safeguards. If you try to obtain information on Taiwan, the Uyghurs, the Tiananmen Square protests—or any political subject likely to contradict Chinese positions or interests—DeepSeek will simply suggest “Let’s talk about something else.”

In contrast, ChatGPT readily acknowledges that the invasion of Iraq was based on false pretenses or that the U.S. military employed torture in Afghanistan.

In the previously mentioned interview, political scientist Asma Mallah asserts that the contrast between these two models reinforces the concept of the “splinternet”: “the fracturing of cyberspace—and, correspondingly, the geopolitical landscape—into two massive blocs, each with radically different worldviews,” both deeply imbued with patriotism.

This patriotism is evident in both models: ask DeepSeek about ChatGPT’s superiority, and it deflects; pose the inverse question to ChatGPT, and it does the same. The machines have learned to defend their respective camps’ colors—”all’s fair in love and war.”