At this year's World Robot Conference (WRC), some robot companies couldn't secure a booth even after waiting in line. While embodied intelligence narratives have reached a fever pitch, Keqiang Robotics founder and CEO Li Tong remains remarkably composed. With over a decade in the industry spanning餐饮, hotel, and elderly care scenarios, he has lived through two technological revolutions, watching peers get lifted by capital only to quietly disappear later.
In an interview, Li Tong admitted that he has experienced two rounds of AI technological revolutions in the past decade. The industry cycles keep shifting, yet the fate of robot companies keeps repeating — those that achieve large-scale deployment tend to thrive, while those that just build demos and put on shows vanish the moment capital markets cool down. "Surviving cycles doesn't depend on fundraising or demos, but on real commercialization — robot companies that don't make money are just being irresponsible." Li Tong's responses contain little sloganeering, just practical business wisdom about survival.
As a leader in China's embodied service robot sector, Li Tong's Keqiang Robotics has shipped over 100,000 robots cumulatively, with products sold in more than 600 cities across over 70 countries and regions. His observations on the current robotics industry are worth heeding. Below is the edited conversation transcript:
Q: This year's robot conference is exceptionally hot, with people queuing up just for a booth. As a veteran attendee, what are your thoughts?
Li Tong: We've participated for many years. In earlier years, attending this expo was about proving to the industry that we were still alive — if you didn't show up, people would start asking whether the company still existed. This year, hearing that people can't even get booths despite queuing, I'm genuinely thrilled as an industry practitioner. Over the past decade, the robotics industry has gone through two technological revolutions: first, the rise of deep learning in 2015; second, the explosion of large models in 2023. The technology differs, but company fates keep repeating. In the 2015 wave, about eighty or ninety robot companies emerged nationwide, and after one full cycle they split into two groups — some grew bigger, others disappeared. The difference boils down to one thing: whether you achieved genuine commercialization and large-scale real-world deployment. Those with mass deployments are doing fine; those who just build demos and put on shows can survive when capital markets are favorable, but vanish the moment they cool off. Capital always runs in cycles — booming this year, cooling the next, then heating up again. To survive these cycles, you must achieve true commercialization. Companies stuck at the demo stage or doing "to VC" business will gradually fade away. Technology advances in leaps: 2015 to 2023 is one wave, and following this rhythm, another revolution might come around 2031, with each leap spawning a new batch of companies and business models.
Q: Capital is cyclical, and the industry is currently hot. How can robot companies survive these cycles?
Li Tong: AI companies must adapt to technology shifts — before 2023, very few people were building large models globally, and the PhDs we hired had only one or two years of experience, but after a year or two of learning, everything changes completely. No one is born good at something; when technology arrives, you need to catch up quickly and use it quickly. But at the end of the day, you're running a business — you must pursue commercialization and make money. Robot companies that don't make money are just being irresponsible. To truly commercialize technology, you must understand what customers actually need and what they're willing to pay for, rather than dreaming in the lab. What Keqiang has consistently focused on is robot commercialization — that's where we differ from many other companies. Our approach is to identify relatively well-defined job roles. For example, in our collaboration with Haidilao, from ordering to serving, three people are involved: one prepares dishes from the ticket, one transports trays, and one serves tea and water. We use robots to take over the repetitive "pick-up and transport" work in the middle, leaving both ends to humans. This role doesn't require infinite generalization — just limited generalization with some tolerance for positional deviation. Once we master this role, it's not just hotpot restaurants — McDonald's and KFC meal preparation and sorting can all use it. One role can sell hundreds of thousands or even millions of units. More importantly, when robots are deployed on-site and run in operations, real data flows back, the data flywheel starts spinning, and the embodied intelligence paradox of "no robots means no data, no data means no robots" gets broken. One role, two roles, three roles — it gradually builds momentum.
Q: Keqiang started with hotel scenarios. When will robots actually enter households? And what direction will you focus on next?
Li Tong: Embodied deployment has three major directions: industry, services, and households. Households are every company's dream — I want it too, but honestly, it needs time. Think about what a hired housekeeper does: laundry, cooking, childcare — capable of everything, yet only costing 30-50 yuan per hour. For robots to enter homes, prices shouldn't be higher than that. This places extremely high demands on home robots. I believe it will definitely happen, but at least five years out. Some reporters ask me why all founders say five years or more — because honestly, I don't know the exact timeline. Say two years and you'll be exposed immediately; say ten years and people walk away. However, I believe finding relatively well-defined roles in industry and commercial services doesn't need five years. That's also our judgment: to enter households, becoming an "expert" in the service industry is the necessary path. First, master being a cleaning expert or bed-making expert in hotels, then a chef in restaurants — as these capabilities accumulate, robots naturally become "full-time housekeepers." Trying to achieve full versatility from the start is too difficult. Additionally, I hold another view: large-scale robot deployment may start in developed countries first. Robots are inherently expensive — with all those joints, cost reduction has limits. Overseas aging populations drive up labor costs, so they can afford prices several times higher. In China, labor is relatively abundant, so why should robots be expensive? Thus, this business might first become economically viable overseas.