Jordan Schneider uncovers a startling paradox in the global artificial intelligence race: a Chinese firm that gives its most powerful technology away for free, yet is on track to generate hundreds of millions in revenue. This piece is essential listening because it moves beyond the usual geopolitical finger-pointing to reveal a unique management philosophy and a business model that challenges the very definition of "open source" in the AI era.
The Economics of Openness
Schneider begins by dismantling the assumption that releasing model weights for free is a charitable act or a loss-leader strategy. Instead, he presents evidence that this is a calculated commercial play. He writes, "When R1 came out in 2025, there was a wave of Chinese businesses and government entities connecting DeepSeek to their internal systems... DeepSeek does not earn revenue from on-premise deployments by corporate entities, so those actual corporate customers must be paying for API tokens."
This reframing is crucial. It suggests that the "open" model is actually a funnel to a high-margin service layer. The data supports this aggressive pivot: the team published an analysis showing that on an average day, the API earned over half a million dollars at a profit margin exceeding 500%. As Schneider notes, "The hypothetical figure Liang apparently gave in the investor meeting for DeepSeek's enterprise-end revenue this year is in the hundreds of millions of US dollars."
The strategy relies on the idea that once the technology is ubiquitous, the value shifts to the infrastructure that supports it. This is a sophisticated take on how to monetize a commodity. However, critics might argue that relying on API sales in a market where competitors are also open-sourcing their models creates a fragile long-term moat, especially if hardware costs continue to rise.
The "Main Quest" for General Intelligence
The core of Schneider's analysis lies in the worldview of Liang Wenfeng, the CEO. Unlike many tech leaders driven by quarterly earnings or product launches, Liang appears driven by a singular, almost academic obsession. Schneider observes that Liang "sincerely holds a specific and highly personalized worldview, if not an entire ideology."
This ideology posits that the path to Artificial General Intelligence (AGI) is not through building better consumer apps, but through solving the fundamental mechanics of machine learning. Liang is quoted as saying, "The path to AGI, in his eyes, runs through mechanisms that allow models to keep acquiring knowledge." He explicitly dismisses other popular research trajectories, such as "world models," as "irrelevant" for the immediate pursuit of higher intelligence.
"More capable models easily pull the rug out from under competitors, especially when the latter have fallen into path dependencies based on older technology."
Schneider's reporting highlights a fascinating divergence in strategy. While the West often focuses on immediate application and integration, DeepSeek is betting everything on the "main quest" of AGI. This approach is bold, but it carries significant risk. If the "learning" mechanisms Liang bets on do not materialize as predicted, the entire company's value proposition could collapse. Yet, the author notes that Liang seems to view the world "as an unfolding kaleidoscope of puzzles," suggesting a mindset that prioritizes solving the problem over selling the solution.
China's Role as the "Token Factory"
The piece also offers a nuanced look at the geopolitical landscape, reframing China's position not just as a challenger, but as a specific type of disruptor. Schneider explains that Liang envisions China playing the role of a "token factory at global scale, pushing the price of intelligence down as it did for countless other industries during its manufacturing boom."
This is a pragmatic acceptance of hardware constraints. Liang acknowledges that while the US maintains a capability advantage, the sheer scale of Chinese manufacturing and software optimization can erode those advantages over time. He reportedly told investors that "Nvidia's CUDA moat to erode," and expressed cautious optimism about training on domestic chips like those from Huawei.
"Liang expects Nvidia's CUDA moat to erode, and — with limited specifics — expressed cautious optimism about training on Huawei chips."
Schneider points out a critical tension here: Liang wants to support the domestic ecosystem but is wary of being seen as a mouthpiece for the state. He claims the government "won't give [them] a cent" if the company fails, a statement Schneider rightly identifies as likely hyperbolic given the coercive power of the state. This balancing act is delicate. If the administration or the executive branch decides to nationalize the AI effort more aggressively, Liang's "regular people" narrative could be quickly dismantled.
A New Theory of Management
Perhaps the most surprising element of Schneider's coverage is the description of DeepSeek's internal culture. In an industry known for grueling hours, Liang has built a culture where "mandatory" tasks take up no more than half of an employee's time, leaving the rest for self-directed research.
Schneider writes, "Good research ideas come from exploring idle curiosity, and there would be no room for such curiosity if researchers were constantly preoccupied with pressure." This stands in stark contrast to the "996" work culture prevalent in China's internet sector and the increasingly workaholic environment in Silicon Valley.
"Rather than jumping onto every vertical and rushing to be first, as is the norm across much of China's internet industry, Liang is betting that focus and discipline will win the day in the AI chapter."
This management theory is compelling because it treats research as a creative process rather than a manufacturing line. However, it assumes a level of intrinsic motivation that may be difficult to sustain as the company scales or faces external pressure. The comparison to Demis Hassabis of DeepMind is particularly apt here. Schneider notes that while Hassabis tried to balance mission and corporate structure, he ultimately became "just another employee" in Google's labyrinth and resigned. Liang, by contrast, seems determined to avoid this fate, even pausing funding rounds to protect his vision.
"The pursuit of it seems inherently worthy to Liang. Is Liang Wenfeng the Demis of China?"
The danger, as Schneider implies, is that the "siren call of the mission" can blind leaders to the dangers of the technology they are building. Throughout the leaked minutes, the potential risks of AGI are notably absent. The focus is entirely on the capability, not the consequence.
Bottom Line
Schneider's analysis provides a rare, granular look at a player who is successfully decoupling AI progress from the traditional constraints of hardware and capital. The strongest part of the argument is the revelation that "open source" can be a highly profitable, high-margin enterprise if the business model is built around API access rather than software licensing. The biggest vulnerability, however, remains the geopolitical tightrope: Liang's vision of a "regular people" company operating outside state control may be impossible to maintain as the US-China tech race intensifies. Readers should watch closely to see if DeepSeek's "token factory" strategy can indeed erode the Western hardware moat, or if the political winds will force a change in direction.