Over 50 billion dollars — and no shares for investors

According to tech publication The Information, DeepSeek's valuation reached 50 billion dollars. That's a figure that places it alongside the most valuable AI companies outside the United States. For comparison: France's Mistral AI was last valued at roughly 6 billion dollars, Germany's Aleph Alpha at a fraction of that amount. But the key detail is the mechanism by which the money flows into the company. Standard venture capital works by an investor buying a stake in the company and acquiring corresponding voting rights. DeepSeek went the opposite way: investors are putting their money into a partnership entity directly controlled by founder and CEO Liang Wenfeng. For their billions, they receive no voting rights, and their capital is locked up for five years — they cannot sell or transfer their stake.

Why would anyone agree to that?

The answer is more complex than it might seem. Throughout 2025, DeepSeek has shown it can develop models comparable to the best American competitors at a fraction of the cost. Its V3 model cost roughly 6 million dollars to train — versus roughly 100 million dollars for OpenAI's GPT-4. The R1 model from January 2025 then shook Silicon Valley so profoundly with its reasoning capabilities that it wiped 600 billion dollars off Nvidia's market value in a single day. This makes DeepSeek a strategic asset that even those willing to forgo voting rights want access to. This isn't an ordinary investment — it's a bet on the geopolitical future of AI.

Who the key players are

The sole exception to the rules is the China National AI Industry Investment Fund, which was the only entity to make a direct investment in the company and retained both voting rights and the ability to exit at any time. In short, the state negotiated a privileged position for itself. Among other investors are names that show just how seriously China is taking its AI sector. Tech giant Tencent is considering a 10 billion yuan investment (about 1.5 billion dollars), and battery manufacturer CATL — the world's largest producer of EV batteries — is contributing 5 billion yuan. Founder Liang Wenfeng himself, according to Reuters, put in 20 billion yuan from his personal fortune.

From a single model to a 50-billion-dollar empire

DeepSeek was founded in July 2023 as a spin-off of Chinese hedge fund High-Flyer. The company employs only about 160 people — for comparison, OpenAI has over 5,000 employees. Yet DeepSeek became the most downloaded app on the US App Store in January 2025, when it launched its chatbot based on the R1 model. In April 2026, the company previewed the V4 model in two variants: V4-Pro with 1.6 trillion parameters and the lightweight V4-Flash with 284 billion parameters. Both models offer a million-token context window and significantly improved capabilities in so-called agentic AI — the ability to independently plan and execute complex tasks. V4-Pro goes head-to-head in independent benchmarks with the latest models from OpenAI (GPT-5.5) and Google (Gemini).

What this means for European users

DeepSeek is freely available to Czech users — the mobile app and web interface (deepseek.com) work without restrictions in the Czech Republic. The chatbot understands Czech and can reply in Czech, though the quality of the Czech occasionally fluctuates. For the average user, this means another free alternative alongside ChatGPT, Gemini, or Claude. For companies and developers, there's an interesting aspect: DeepSeek publishes its models under the MIT open-source license, so anyone can download them and run them on their own infrastructure. This is a fundamental difference from OpenAI's closed models, which are accessed only via API. For European companies that must comply with strict GDPR rules and the emerging EU AI Act, running a Chinese open-source model locally can be, paradoxically, easier for compliance than using cloud APIs from American companies.

Is this about Chinese AI, or a global phenomenon?

DeepSeek's funding comes at a time when the global AI race is intensifying. China is investing massively in AI infrastructure — by the latest estimates, Chinese tech companies invested over 80 billion dollars in AI in 2025. The United States is responding with the Stargate project (500 billion dollars over four years) and accelerating investments from Microsoft, Google, and Meta. Europe, for now, is falling behind in this contest. While the European Commission is preparing an investment plan in the hundreds of billions of euros, concrete results are yet to materialize. The DeepSeek story shows that regardless of regulations and chip export restrictions, capable teams can break through — and raise unprecedented capital to do so.

A unique model that could reshape venture capital

The structure DeepSeek used — a founder-controlled partnership, with no investor voting rights and a five-year lock-up — is virtually unprecedented in the tech world. It more closely resembles structures used by sovereign wealth funds and state investments. The question is: will this be a one-off experiment, or the start of a trend? If DeepSeek succeeds, it will send a signal to other AI startups — especially those with unique technology and strong negotiating positions — that they can demand capital from investors on similarly asymmetric terms. On the other hand, few startups have a position as strong as a company whose models were able to rattle the US stock market in January 2025.