Zhou Yuxiang hired food delivery riders to sell AI to factories. Now he’s chasing China’s industrial AI breakthrough
So Zhou recruited his sales team from a different labor pool: food delivery riders. Couriers from Meituan and Alibaba’s Ele.me “have the best, you know, street-smart wisdom to meet the decision makers,” he explained. Drivers knew the back entrances, the factory layouts, and the best path to the boss of a mom-and-pop workshop.
Still, what these drivers had in chutzpah, they lacked in actual product knowledge. So Zhou put every new hire through what he called a “devil boot camp”: five days of 14-hour training sessions, a daily exam, and one contract closed over the phone by the final day. Only about half graduated.
Zhou, a Dartmouth grad and former investment banker, is now betting that the most interesting market for AI isn’t the consumer smartphone nor the corporate office, but rather the shop floor. Black Lake’s product is factory-management software: a cloud-based system that tracks orders, schedules, materials, and output—and, more recently, AI agents that can read an incoming order, generate a quote, and schedule production on their own. Black Lake says it has been profitable since 2025 and that revenue has grown more than 60% a year.
China is the testing ground for that idea. It is also, so far, an unproven one. Zhou’s larger wager isn’t just that AI can make existing work faster, but also that it can restructure how an industry operates.
He doesn’t believe it’s happened yet, anywhere. “I don’t think the Google moment has happened yet in the AI era, because we are only replacing [tasks]” he said. “I don’t think we have restructured the industry.”
The factory data desert
Black Lake is actually Zhou’s second attempt at building a startup. In 2015, he and two high school friends returned to China to found MadaData, an analytics startup Zhou describes as “a Palantir for China.” The idea was that, much like how Palantir sits on top of enterprise and government data, MadaData would sit on top of factory data.
It was a flop. “It turned out most of the factories that we visited back then had zero data,” Zhou says.
As the team liquidated MadaData, they were saved by a group of second-generation factory owners, “forced by their family to take over [the] family business.” Educated abroad like Zhou, they faced a market that was shifting towards smaller orders with faster turnarounds than the big, standardized runs their parents’ factories were built for.
This time, Zhou wanted hands-on experience. “We thought we should really go into the factory instead of locking ourselves up in the room and coding,” he said.
He took a job on a plastic-injection line at a factory that made cosmetics for a European giant, working alongside migrant laborers from less-developed provinces like Sichuan and Guizhou. Zhou timed what happened between batches, and found that, during a six-hour changeover, only one hour was useful. The rest was wasted due to “subjective decision-making” and information silos.
China’s industrial AI advantage
China has perhaps become the world’s second-most important source of AI innovation. After the U.S., open-weight models from Alibaba, DeepSeek, Z.ai, and Moonshot AI now deliver near-frontier performance at a fraction of the cost of closed Western rivals. Much of that energy is being applied to manufacturing. China installed 354,000 industrial robots in 2025, a record 59% of the world’s total, according to the International Federation of Robotics.
Despite the hype, China’s AI sector faces some significant constraints on its growth. U.S. export controls still deny China access to the frontier AI processors. China’s funding pool is also far shallower: OpenAI raised $122 billion in a single round this year and Anthropic $65 billion, totals which no Chinese company can match.
Yet China is still playing catch-up with the West. “For Western countries, that digitization … took place for 40-50 years, since … the founding of software companies in the 1980s or 1990s,” Zhou said. “But in China, I think really the concept of digitization became a really recognized concept to factory owners only 10 years ago.”
As frontier models catch up with one another within weeks, data is becoming the differentiating factor in AI. “How do we prepare data? How do we accumulate data?” he asked onstage. “That has become an industry question.”
Chasing AI’s Google moment
For all the agents Black Lake has deployed, Zhou doesn’t believe AI has yet reached its “Google moment,” and changed how industries fundamentally work; so far it has replaced tasks, not business models.
His attempt at an answer is a new platform called PopZao, a website and WeChat mini program that Black Lake has incubated for two years. The proposition is to connect China’s factories with its creator economy; Black Lake’s agents bridge the gap by converting ideas into designs, a 3D simulation, and industrial data sent to a factory.
Zhou tied the idea to the craze around Labubu, the Pop Mart-owned toy that went viral in 2025, helping to spur a 185% jump in revenue for the toy store chain last year. PopZao, in theory, allows anyone with a following—now just massive IP machines—to develop merchandise.
One early customer of PopZao was a Buddhist temple which wanted new merchandise for its younger visitors. After the temple’s master submitted a photo of its stone lions, Black Lake designed and manufactured a small batch of 500 units in just 30 days; they sold out within a week. “Now the temple is taking pictures of trees, the dogs, cats, different kinds of things as their souvenirs,” Zhou said.
