In February 2026, a notice on the People’s Liberation Army’s procurement platform revealed that Shanxi 100 Trust Information Technology, a 266-person IT firm headquartered in China’s coal belt, had been banned from military procurement for one year for submitting falsified bidding materials. What made the penalty remarkable was not the violation itself, which is common enough in PLA procurement. It was the company.
100 Trust is what a Jamestown Foundation analysis identified as “the only wholly privately-owned firm” in China’s xinchuang (信创) domestic IT innovation framework. Yet this private company from Taiyuan holds classified-project clearance and has won some of the PLA’s largest DeepSeek integration contracts.
The empirical picture is striking, and consistent across every major dataset available. According to a Georgetown CSET study, of the 338 entities that won AI-related PLA contracts, close to three-quarters were nontraditional vendors with no self-reported state ownership ties. The Jamestown Foundation analysis, which looked at hundreds of DeepSeek-specific procurement tenders, found the same pattern: private companies, not state-owned enterprises (SOEs), won a majority of integration contracts. The trend spans domains, from autonomous combat-support vehicles to battlefield scenario assessment tools, as Reuters documented in a separate investigation.
Beijing is clearly serious about military AI. But private firms, not the state’s own defense industrial base, are doing most of the delivering.
Why Private Firms Win
Two structural conditions explain the pattern, predict which firms are most likely to keep winning, and offer a framework for tracking how this market evolves.
The first is dependence on state-controlled resources. U.S. export controls on advanced chips have created what might be called an export control paradox: restrictions intended to slow China’s military AI development have simultaneously deepened private firms’ reliance on the Chinese government for access to computing power. With Nvidia hardware increasingly difficult to obtain, firms seeking government and military contracts have converged on domestic alternatives, primarily Huawei’s Kunpeng processors and Ascend AI chips, whose adoption may have become a de facto prerequisite for access to state procurement.
Firms building on this stack have a structural incentive to demonstrate political reliability, and winning PLA contracts is among the most visible ways to do so. 100 Trust’s own product page highlights “domestically produced core components” as a key selling point, and its primary computing infrastructure runs on the Huawei stack. This is not just about branding. It is a market signal directed at the state.
The second condition is organizational capacity for rapid integration. DeepSeek’s open-source architecture compressed what was once a model-layer advantage into a deployment-speed competition. When the underlying model is freely available, the differentiator becomes how fast a firm can customize, integrate and deliver a working system to a military end user.
DeepSeek-related procurement accelerated throughout 2025, and compressed tender timelines structurally favor the firms that can move fastest. Private IT firms, lean and technically agile, hold a structural edge in precisely this kind of competition. Researchers at Xi’an Technological University demonstrated the potential: a DeepSeek-powered system assessed 10,000 battlefield scenarios in 48 seconds, a task they estimated would take conventional military planners 48 hours.
The interaction between these two conditions matters most. Firms with both high dependence on state-controlled compute and high capacity for rapid integration are the ones that win repeated PLA contracts. This is a structurally predictable outcome, and it is precisely the profile of companies like 100 Trust.
The Potential for Corruption
On paper, 100 Trust is the ideal case of private-sector alignment with military priorities. Beyond the PLA contracts and classified-project clearance described above, the company builds its computing infrastructure primarily on Huawei’s Kunpeng and Ascend architecture, maintains significant contracts with the China Aerospace Science and Technology Corporation, and says its products were reviewed by top leader Xi Jinping on three separate occasions. By any measure, this is a firm deeply embedded in the defense ecosystem, with every structural incentive to align.
Yet on February 8, the PLA’s military procurement platform published a penalty notice from the Cyberspace Force procurement department: 100 Trust had submitted falsified materials in a bidding process and was banned from all military procurement across all service branches for one year. The penalty extended to the company’s legal representative and its authorized bidding agent.
The violation, ironically, stems from the very structural logic that explains private firms’ outsized role in PLA AI procurement. The same agility that allows a small private company to outcompete state-owned defense giants in rapid integration cycles also creates the conditions for opportunistic bid inflation, overstating capabilities to win contracts in a market where demand outpaces verified supply. Private-sector speed is a double-edged sword: it accelerates both genuine integration and the temptation to fake it.
This dynamic is unfolding against a backdrop of extraordinary institutional disruption. A CSIS ChinaPower database tracking PLA purges has documented the removal of over 100 senior officers at the general and lieutenant-general level, many tied to procurement corruption in the Rocket Force and equipment development systems. The institutional machinery that traditionally mediated defense procurement – approval chains, oversight bodies, relationship networks between buyers and suppliers – has been hollowed out.
Yet the central signal on military AI has never been clearer. Beijing’s top leadership has made intelligentized warfare an explicit priority, and DeepSeek procurement notices accelerated throughout 2025, a period during which the purge continued to deepen. The result is a paradox: the order from the top has never been clearer, but the institutional machinery meant to execute that order is less and less capable.
In this environment, firms that depend on stable bureaucratic relationships to win contracts are disadvantaged. Firms with high technical capacity that can compete primarily on deployment speed, rather than on entrenched procurement relationships, benefit disproportionately. The purge, counterintuitively, may be reinforcing private firms’ structural advantage.
What Washington Gets Wrong
The outsized role of private firms in PLA AI procurement carries three underappreciated implications for U.S. policy.
First, the U.S. Entity List, which governs export controls, has a structural blind spot. Current designation logic targets entities with documented or visible military ties, whether state-owned defense conglomerates, surveillance technology firms, or universities with known PLA affiliations. But the firms winning the PLA’s major AI integration contracts are obscure private IT companies – firms that do not appear on any existing restricted entity list. The architecture of U.S. export controls was not designed for a military procurement ecosystem in which the primary integrators are commercially oriented private firms with no formal defense lineage.
Small private firms also possess a resilience that large entities do not: when a state-owned conglomerate or a major corporation like Huawei is sanctioned, restrictions penetrate the entire corporate structure. When a small private company is designated, the actual controller can dissolve the entity, reconstitute the same team under a new registration, and resume operations. The U.S. Entity List was built to constrain institutions. It is less effective against individuals who treat corporate shells as disposable.
Second, chip restrictions produce second-order effects that cut against their intended purpose. Export controls impose genuine friction on China’s access to advanced computing hardware. But they simultaneously accelerate the militarization of the domestic chip ecosystem. The firms now winning PLA contracts are marketing themselves explicitly on Huawei Ascend and Kunpeng stacks, largely because of U.S. export controls. Restrictions that deepen private firms’ dependence on state-favored compute also deepen their incentive to demonstrate alignment through military work.
Third, procurement volume is a poor proxy for capability depth. The 100 Trust penalty is itself evidence of a supply-side gap: when a firm with classified-project clearance whose products have been presented to Xi Jinping submits falsified bidding materials, something is deeply wrong with the PLA’s AI procurement. Washington needs sharper metrics to actually evaluate China’s progress in military AI – incorporating data points like repeat award patterns, defense-relevant hiring trends, and verified technical delivery records – rather than reacting to headline contract counts.
A 266-person IT company from China’s coal belt won some of the PLA’s largest AI integration contracts and then got caught faking its bids. The 100 Trust story is not proof of Chinese military AI success or failure. It is a window into a more complex reality: Beijing’s strategic intent on military AI is unambiguous, but the market mechanism executing that intent is simultaneously producing genuine capability and inflated signals.
Calibrating U.S. policy to this differentiated reality, distinguishing structural integration from procurement theater, matters more than reacting to the next headline about how many contracts the PLA has signed. The two traits of Chinese private companies identified here – dependence on state-controlled compute and capacity for rapid integration – are not unique to 100 Trust. They describe a structural type, and Washington would benefit from tracking which firms fit this profile rather than reacting to individual procurement headlines.
But the analytical window this article relies on may not remain open. Beijing has already restricted overseas access to corporate registration databases, judicial records, and academic repositories. If military procurement data follows the same trajectory, the opportunity to distinguish structural integration from procurement theater will shut before Washington has fully calibrated its response.
