DeepSeek’s Restraint: The New AI War

In a closed room, before a group of potential investors, DeepSeek founder Liang Wenfeng uttered a word that could redefine the global race for artificial intelligence: restraint. It was neither a concession nor a sign of weakness. For nearly four hours, according to a leaked transcript that has gone viral on Chinese social media, Liang laid out a corporate philosophy that directly challenges the foundations of Silicon Valley’s tech capitalism: thin margins, open-source models, and a discipline that prioritizes long-term survival over immediate commercialization. In an ecosystem where an AI company’s worth is measured by its funding rounds and its ability to scale at any cost, Liang’s proposal is no cultural quirk—it is a strategic declaration of war. And the leaked data—which DeepSeek has not officially confirmed, as reported by Coco Feng from Guangdong for the South China Morning Post—offers the first real window into the mind running one of China’s most mysterious labs.

The Recluse Who Wants to Change the Game

Barely a couple of photos of Liang Wenfeng circulate online, according to the SCMP article itself. He is a reclusive figure in an industry that adores its celebrities: Sam Altman, Demis Hassabis, Elon Musk. But his low profile conceals a career blending the quantitative world of finance with the frontier of AI. Liang founded DeepSeek out of the hedge fund High-Flyer Quant, a quantitative fund that already operated machine learning algorithms for financial markets. That dual identity—hedge fund financier and AI scientist—explains much of his approach.

The leaked transcript, which the SCMP says spread widely on platforms like Weibo and WeChat, contains no technical revelations about DeepSeek’s models. No architectural secrets or breakthroughs in computational efficiency. What it contains is a statement of principles that, in the context of the U.S.-China tech war, proves more explosive than any algorithmic advance. Liang insists that restraint is not a limitation but a competitive advantage. In a market where U.S. giants burn billions on infrastructure and talent, DeepSeek’s proposal is to win not through resource superiority but through strategic discipline.

Three Pillars of an Uncomfortable Philosophy

Liang’s strategy rests on three axes that, according to the leaked transcript, define DeepSeek’s identity. The first is the low-margin business model. Instead of maximizing profit per user or per processed token, DeepSeek bets on prices that barely cover costs. This is not philanthropy: it is a way to erode competitors’ customer bases, especially those reliant on immediate profitability to justify astronomical valuations. If the long-term goal is market dominance, a low-price strategy can force rivals to choose between losing market share or cutting their own margins.

The second pillar is open-source models. While OpenAI and Anthropic jealously guard their architectures, DeepSeek releases its models for anyone to download, modify, and run. This not only accelerates global adoption of its technology but creates an ecosystem of developers dependent on its tools. In a move reminiscent of Android’s strategy against iOS, DeepSeek sacrifices direct control for ubiquitous presence. And in the context of U.S. sanctions on advanced chip exports to China, open-source allows Chinese AI to deploy on any infrastructure, free from reliance on U.S. clouds.

The third pillar is corporate discipline. Liang, according to the transcript, insists DeepSeek will not be swayed by investor pressure to scale prematurely. The company maintains a flat structure, avoids superfluous marketing spend, and focuses on fundamental research. This discipline, per the SCMP article, resonates deeply in a culture that values humility, discipline, and long-term ambition over quick commercialization. But it is also a pragmatic response to capital constraints: DeepSeek lacks access to the same funding volume as its U.S. rivals and must optimize every yuan invested.

The Context of the Chip War and State Funding

Liang’s restraint strategy cannot be understood outside the geopolitical framework in which it operates. Since the first restrictions on advanced chip exports to China were imposed, Chinese AI labs have had to innovate under pressure. DeepSeek, in particular, has stood out for achieving competitive results with fewer computational resources than its U.S. counterparts. But restraint is not just a response to sanctions: it is a philosophy aligning with the state capitalism model Beijing is promoting.

A companion SCMP article analyzes how the Chinese government channels funding into strategic tech companies through state-controlled investment vehicles. DeepSeek, though not publicly traded or disclosed its shareholding structure, operates in an ecosystem where state capital is omnipresent. Liang’s restraint can be interpreted as a way of aligning with Party priorities: stability, long-term planning, and technological sovereignty over private profit maximization.

This convergence between DeepSeek’s corporate philosophy and Beijing’s strategic interests raises an uncomfortable question for the West: what if the low-margin, open-source business model is not a weakness but a form of competition that venture capitalism cannot match? Silicon Valley depends on the promise of exponential returns to attract investment. If a Chinese rival offers comparable technology at cost and with no immediate profit motive, the financial equation breaks down.

The Mirage of Transparency and the Risks of the Leak

Prudence is necessary, however. DeepSeek has not officially confirmed the leaked transcript’s authenticity. As the SCMP article notes, the company did not respond to requests for comment, and independent verification of the document is impossible without access to the original. Leaks in China’s tech ecosystem are often tools of pressure or propaganda, and it cannot be ruled out that the document has been manipulated or taken out of context.

But even if the transcript is authentic, the restraint strategy carries its own risks. Low margins can work while DeepSeek remains a small player, but as it grows, the need for infrastructure investment—data centers, chips, talent—could force a course change. Open-source, meanwhile, can generate dependency but also allows competitors to copy and improve models royalty-free. And corporate discipline can turn into rigidity if the market demands a rapid response.

Moreover, restraint as a power strategy has a limit: AI is a field where scale matters. The most advanced models require clusters of tens of thousands of chips, and training a frontier model costs tens of millions of dollars. If DeepSeek cannot access that scale, its restraint could turn into irrelevance.

The Future of Competition: Chinese Patience vs. American Vertigo

The DeepSeek leak is not just a window into the mind of a reclusive founder. It is a symptom of a deeper shift in AI geopolitics. For years, the dominant narrative has been that China trails behind the U.S., constrained by sanctions and a lack of chips. But Liang Wenfeng’s strategy suggests that Beijing and its tech champions are playing a different game: not one of speed, but of endurance.

In a Chinese corporate culture that values patience and discipline, restraint can be a real advantage. While U.S. companies burn out in a race for the next funding round, DeepSeek can afford to wait, optimize, and grow organically. And if open-source makes its models the de facto standard for developers worldwide—especially in the Global South, where low prices are decisive—the first-mover advantage of scaling could vanish.

The final reflection, from a digital sovereignty perspective, is uncomfortable: AI will not be won solely with better hardware or more data. It will be won with business models that withstand the pressure of time. If Liang Wenfeng is right, restraint is not a defensive strategy: it is the smartest way to attack. And the West, trapped in its own logic of exponential growth, may not yet have begun to understand what it is up against.