In October 2022, Unitree Robotics joined Boston Dynamics, Agility Robotics, and three other firms in signing an open letter pledging not to weaponize their machines and to review customers’ intended applications. This Chinese startup based in Hangzhou had earned its place in that group. Founded in 2016 by a 26-year-old engineer named Wang Xingxing, who quit his job at the drone maker DJI during his probationary period, Unitree set out to build affordable quadruped robots. It worked. By 2023, the company held over 60 percent of the global quadruped robot market by unit sales. Its investors included Sequoia China, Meituan, and Shenzhen Capital Group. Its consumer-grade robot dogs are sold on Amazon.
Two years after the pledge, China’s state broadcaster aired footage of Unitree’s B1 quadrupeds carrying assault rifles in joint military exercises between the Chinese and Cambodian armed forces. By September 2025, robot dogs appeared in China’s largest-ever military parade on Chang’an Avenue. In July of the same year, a military training exercise at a Chongqing vocational college — jointly developed by students, faculty, and a military training team — featured Unitree machines fitted with rifles and rocket launchers. In an August 2025 statement, Unitree affirmed that it “has always been a civilian robotics company” and that all militarized modifications were made by third parties.
The interesting thing is, Unitree isn’t really lying. And that is precisely the policy problem. Unitree did not seek out the Chinese military. The Chinese Communist Party brought the military to Unitree through channels that required no direct order, no classified contract, and no corporate consent. Understanding how that system works, and why current U.S. policy does not reach it, is the most urgent analytical gap in the American response to Chinese military-civil fusion.
What Congress Heard — and What It Missed
When four expert witnesses testified before the House Homeland Security Committee on March 17 about China’s growing dominance in robotics and artificial intelligence, they produced an impressive threat diagnosis. Former National Security Council China director Rush Doshi cited International Federation of Robotics data showing that China installed roughly 300,000 industrial robots last year — 10 times the U.S. figure. Scale AI estimated that China now controls approximately 90 percent of the commercially available robotics AI data market. Boston Dynamics warned that China already has dozens of companies like Unitree, backed by a national strategy aimed at embedding artificial intelligence in virtually every major industry.
The witnesses converged on a policy response: expand the information and communications technology supply chain review framework to cover Chinese robots, direct the Cybersecurity and Infrastructure Security Agency to audit robotic deployments, and prohibit federal procurement of Chinese systems — systems that, as the subcommittee’s chairman noted, already operate in American police departments, universities, and parts of the federal government. Each proposal targets either the supply side or the finished product. None addresses the structural middle layer: the system of state designations, tax incentives, procurement channels, and standards bodies that shapes firm behavior long before any product reaches a military exercise field.
Three days after that hearing, on March 20, the Shanghai Stock Exchange accepted Unitree’s application for a $610 million initial public offering on its Science and Technology Innovation Board and published the company’s 364-page prospectus. When I read through it, I found detailed disclosures of tax incentives, revenue structure, and state-linked investors. I did not find a single reference to defense, the military, or the People’s Liberation Army. The absence is itself the evidence. This is a system that does not require its participants to acknowledge what it produces.
How the System Converts Commercial Success into Military Capability
The conversion follows a sequence. China’s Ministry of Industry and Information Technology designated Unitree a national-level “Little Giant” enterprise under the specialized, refined, differentiated, and innovative (zhuanjingtexin) program — a status that unlocks preferential tax treatment, subsidized financing, and procurement access. In January 2025, Unitree’s home city of Hangzhou, which has a population of over 12 million people, designated Unitree one of the city’s “six little dragons” alongside DeepSeek, allocating 15.72 percent of its industrial policy budget to foster such firms. The company’s own prospectus discloses tax incentives totaling 75.9 million yuan ($10.56 million) in the first nine months of 2025 alone — including a reduced 15 percent corporate income tax rate as a certified high-tech enterprise, research and development expense super-deductions, and value-added tax refunds. Unitree operates from the Hangzhou High-Tech Zone, which the House Select Committee on the Chinese Communist Party has described as a designated military-civil fusion hub, with the zone’s development and reform bureau holding an institutional mandate to coordinate military-civil fusion planning and ‘civilian participation in military’ projects across the district.
Then came the university channel. Nearly 30 Chinese universities purchased Unitree’s products over the past five years, and the company’s prospectus confirms that 73.6 percent of humanoid robot revenue in the first nine months of 2025 came from the “research and education” channel. Technology firms worldwide sell to universities. But the institutional environment differs decisively.
As investigative firm Kharon documented, Unitree’s university customers include at least five institutions on U.S. or Canadian export control lists for military-related activities, among them two of China’s “Seven Sons of National Defense.” Kharon’s review of procurement records found that North University of China — historically known as the “First School for People’s Ordnance” — described its Unitree purchase in December 2022 documents as intended for “weapon science and technology” applications. A 2017 State Council directive explicitly instructed universities and research institutes to deepen participation in defense research and weapons development under the military-civil fusion framework. By 2017, over 150 universities had obtained classified weapons research credentials, according to China’s Secrecy Work journal. Commercial equipment entering the university channel enters an environment where end-use visibility systematically diminishes. This is not a leak in the system. It is the system working as designed.
The appointment that completed the sequence came in November 2025, when the Ministry of Industry and Information Technology named Wang Xingxing vice chair of the newly created National Humanoid Robot Standardization Technical Committee. The ministry is a civilian organ under the State Council, not a military body. But that is precisely the point. The committee’s members, as Jamestown Foundation analysis has detailed, include the 43rd Research Institute of China Electronics Technology Group Corporation — a core defense conglomerate — along with multiple “Seven Sons of National Defense” universities and U.S.-sanctioned firms SenseTime and Huawei. When a civilian ministry’s standards-setting process draws its membership from this composition, military requirements embedding themselves in technical standards becomes a structural probability — regardless of what the committee’s charter says about its mission.
A founder who signed an anti-weaponization pledge now sits on a standards body alongside sanctioned defense firms and military-linked universities, shaping the procurement architecture his own products will be evaluated against. No one forced this. The appointment is, by every visible metric, a recognition of commercial achievement and technical leadership. It is also structural integration into the defense ecosystem — achieved through the ordinary operation of incentives.
The Layer That U.S. Policy Does Not Reach
The pattern is not unique to Unitree. It reveals a more fundamental structure: China’s system does not require a firm’s active willingness to participate. It requires only that firms respond rationally to the environment they inhabit — accept state designations and their accompanying benefits, operate where the subsidized infrastructure is, sell to buyers who arrive through legitimate channels, and join the standards body when appointed. The cumulative outcome is military integration — without requiring a single directive.
This is the layer that current U.S. policy does not reach. Chip export controls target the top of the technology stack. Information and communications technology supply chain reviews and entity lists target finished products at the point of sale or deployment. But the conversion from commercial success to defense capability happens in the structural middle — the designations, tax incentives, university procurement pipelines, and standards committees that shape firm trajectories years before any product is weaponized. Unitree’s prospectus confirms it has vertically integrated its core components, from laser-ranging sensors to joint motors, through full-stack in-house research and development. No entity list retroactively disassembles a vertically integrated supply chain or reverses a decade of state-subsidized capability accumulation.
In February 2026, the Department of Defense added Unitree to its Chinese Military Companies list, citing precisely the state support and military-civil fusion linkages described above. The designation is warranted. It is also structurally insufficient — it addresses one firm while the architecture that produced it remains intact. As Jessica Liao and Joshua Arostegui argued on War on the Rocks in January, the deeper contradictions in China’s defense procurement system are “fundamentally entrenched in the institutional fabric of the broader party-state system.” The problem is not that Beijing occasionally succeeds in militarizing a commercial firm. The problem is that the incentive architecture makes militarization the default trajectory for every commercially successful firm in a strategic sector.
The more consequential policy direction, as Doshi suggested in his testimony, is to build allied manufacturing scale that offers an alternative ecosystem. But the case for doing so rests on understanding why scale matters: not because China has more factories, but because China has built an incentive architecture that makes every commercially successful firm in a strategic sector a potential — and eventually probable — contributor to military modernization.
The United States is not without tools of its own. The research and development tax credit, Small Business Innovation Research grants, Defense Advanced Research Projects Agency programs, and non-traditional defense companies demonstrate real capacity. But the scale and structure tell different stories. According to the most recent annual report to Congress, the Small Business and Innovation Research program alone obligated $4.4 billion in fiscal year 2022, across all federal agencies and all sectors — not just robotics. The Defense Advanced Research Projects Agency’s budget of $4.4 billion funds the full spectrum of defense research. Meanwhile, a single Chinese city, Hangzhou, has earmarked 15.72 percent of its industrial policy budget for frontier industries, including robotics and artificial intelligence, drawn from a total municipal fiscal package of 50.2 billion yuan. Unitree’s own prospectus discloses $10.5 million in tax incentives in just nine months — benefits that arrived automatically through state designations, not through competitive grant applications. The American model is project-driven and contract-driven: The government enters as a purchaser after a firm matures. China has built something different: a front-loaded, automatically triggered incentive chain that runs from state designation to tax preference to procurement channel to standards body, enrolling firms into the defense ecosystem during their commercial ascent. Whether Washington can replicate that front-loading effect within a democratic system remains an open and urgent design problem: conditioning research and development tax incentives on dual-use readiness or building allied procurement frameworks that reward early-stage defense interoperability would be starting points rather than solutions.
The witnesses on March 17 were right that the United States is falling behind in robotics and artificial intelligence. What they did not say — and what policymakers need to hear — is that the gap is not primarily technological. It is architectural. China has built a system that converts commercial success into military capability through the routine operation of incentives. Replicating that system is neither possible nor desirable — it works precisely because it operates within an authoritarian party-state that can blur the line between commercial and military ecosystems without democratic accountability. But understanding the architecture is the precondition for any serious response. Until Washington grasps what it is competing against, it will keep sanctioning firms one at a time while the system produces new ones faster than any list can grow.
Charles Sun is a China-focused policy analyst and a Yale Sinovation fellow at the Yale School of Management. He was previously a global affiliate visiting fellow at the Shorenstein Asia-Pacific Research Center, Stanford University. His Stanford research introduced the proactive elite alignment theory framework for analyzing anticipatory compliance in China’s tech sector. His research draws on Chinese-language government records, corporate filings, and procurement documents, informed by a decade of professional experience in Beijing’s financial sector. His analysis has appeared in Jamestown Foundation’s China Brief and The Diplomat.
Image: Lance Cpl. Justin J. Marty via the United States Marine Corps.
