In February 2026, a discover on the Individuals’s Liberation Military’s procurement platform revealed that Shanxi 100 Belief Data Know-how, a 266-person IT agency headquartered in China’s coal belt, had been banned from army procurement for one 12 months for submitting falsified bidding supplies. What made the penalty outstanding was not the violation itself, which is widespread sufficient in PLA procurement. It was the corporate.
100 Belief is what a Jamestown Basis evaluation recognized as “the one wholly privately-owned agency” in China’s xinchuang (信创) home IT innovation framework. But this non-public firm from Taiyuan holds classified-project clearance and has received a few of the PLA’s largest DeepSeek integration contracts.
The empirical image is putting, and constant throughout each main dataset accessible. In keeping with a Georgetown CSET research, of the 338 entities that received AI-related PLA contracts, near three-quarters had been nontraditional distributors with no self-reported state possession ties. The Jamestown Basis evaluation, which checked out tons of of DeepSeek-specific procurement tenders, discovered the identical sample: non-public firms, not state-owned enterprises (SOEs), received a majority of integration contracts. The development spans domains, from autonomous combat-support automobiles to battlefield situation evaluation instruments, as Reuters documented in a separate investigation.
Beijing is clearly critical about army AI. However non-public companies, not the state’s personal protection industrial base, are doing many of the delivering.
Why Non-public Companies Win
Two structural situations clarify the sample, predict which companies are most probably to maintain successful, and provide a framework for monitoring how this market evolves.
The primary is dependence on state-controlled assets. U.S. export controls on superior chips have created what may be known as an export management paradox: restrictions supposed to gradual China’s army AI growth have concurrently deepened non-public companies’ reliance on the Chinese language authorities for entry to computing energy. With Nvidia {hardware} more and more troublesome to acquire, companies in search of authorities and army contracts have converged on home options, primarily Huawei’s Kunpeng processors and Ascend AI chips, whose adoption could have turn out to be a de facto prerequisite for entry to state procurement.
Companies constructing on this stack have a structural incentive to reveal political reliability, and successful PLA contracts is among the many most seen methods to take action. 100 Belief’s personal product web page highlights “domestically produced core parts” as a key promoting level, and its major computing infrastructure runs on the Huawei stack. This isn’t nearly branding. It’s a market sign directed on the state.
The second situation is organizational capability for fast integration. DeepSeek’s open-source structure compressed what was as soon as a model-layer benefit right into a deployment-speed competitors. When the underlying mannequin is freely accessible, the differentiator turns into how briskly a agency can customise, combine and ship a working system to a army finish consumer.
DeepSeek-related procurement accelerated all through 2025, and compressed tender timelines structurally favor the companies that may transfer quickest. Non-public IT companies, lean and technically agile, maintain a structural edge in exactly this type of competitors. Researchers at Xi’an Technological College demonstrated the potential: a DeepSeek-powered system assessed 10,000 battlefield situations in 48 seconds, a activity they estimated would take standard army planners 48 hours.
The interplay between these two situations issues most. Companies with each excessive dependence on state-controlled compute and excessive capability for fast integration are those that win repeated PLA contracts. This can be a structurally predictable final result, and it’s exactly the profile of firms like 100 Belief.
The Potential for Corruption
On paper, 100 Belief is the best case of private-sector alignment with army priorities. Past the PLA contracts and classified-project clearance described above, the corporate builds its computing infrastructure totally on Huawei’s Kunpeng and Ascend structure, maintains vital contracts with the China Aerospace Science and Know-how Company, and says its merchandise had been reviewed by prime chief Xi Jinping on three separate events. By any measure, this can be a agency deeply embedded within the protection ecosystem, with each structural incentive to align.
But on February 8, the PLA’s army procurement platform revealed a penalty discover from the Our on-line world Power procurement division: 100 Belief had submitted falsified supplies in a bidding course of and was banned from all army procurement throughout all service branches for one 12 months. The penalty prolonged to the corporate’s authorized consultant and its licensed bidding agent.
The violation, sarcastically, stems from the very structural logic that explains non-public companies’ outsized function in PLA AI procurement. The identical agility that permits a small non-public firm to outcompete state-owned protection giants in fast integration cycles additionally creates the situations for opportunistic bid inflation, overstating capabilities to win contracts in a market the place demand outpaces verified provide. Non-public-sector pace is a double-edged sword: it accelerates each real integration and the temptation to faux it.
This dynamic is unfolding towards a backdrop of extraordinary institutional disruption. A CSIS ChinaPower database monitoring PLA purges has documented the elimination of over 100 senior officers on the common and lieutenant-general stage, many tied to procurement corruption within the Rocket Power and tools growth methods. The institutional equipment that historically mediated protection procurement – approval chains, oversight our bodies, relationship networks between consumers and suppliers – has been hollowed out.
But the central sign on army AI has by no means been clearer. Beijing’s prime management has made intelligentized warfare an express precedence, and DeepSeek procurement notices accelerated all through 2025, a interval throughout which the purge continued to deepen. The result’s a paradox: the order from the highest has by no means been clearer, however the institutional equipment meant to execute that order is much less and fewer succesful.
On this setting, companies that rely upon secure bureaucratic relationships to win contracts are deprived. Companies with excessive technical capability that may compete totally on deployment pace, slightly than on entrenched procurement relationships, profit disproportionately. The purge, counterintuitively, could also be reinforcing non-public companies’ structural benefit.
What Washington Will get Incorrect
The outsized function of personal companies in PLA AI procurement carries three underappreciated implications for U.S. coverage.
First, the U.S. Entity Listing, which governs export controls, has a structural blind spot. Present designation logic targets entities with documented or seen army ties, whether or not state-owned protection conglomerates, surveillance know-how companies, or universities with recognized PLA affiliations. However the companies successful the PLA’s main AI integration contracts are obscure non-public IT firms – companies that don’t seem on any present restricted entity checklist. The structure of U.S. export controls was not designed for a army procurement ecosystem by which the first integrators are commercially oriented non-public companies with no formal protection lineage.
Small non-public companies additionally possess a resilience that giant entities don’t: when a state-owned conglomerate or a significant company like Huawei is sanctioned, restrictions penetrate your complete company construction. When a small non-public firm is designated, the precise controller can dissolve the entity, reconstitute the identical group underneath a brand new registration, and resume operations. The U.S. Entity Listing was constructed to constrain establishments. It’s much less efficient towards people who deal with company shells as disposable.
Second, chip restrictions produce second-order results that minimize towards their supposed goal. Export controls impose real friction on China’s entry to superior computing {hardware}. However they concurrently speed up the militarization of the home chip ecosystem. The companies now successful PLA contracts are advertising themselves explicitly on Huawei Ascend and Kunpeng stacks, largely due to U.S. export controls. Restrictions that deepen non-public companies’ dependence on state-favored compute additionally deepen their incentive to reveal alignment by army work.
Third, procurement quantity is a poor proxy for functionality depth. The 100 Belief penalty is itself proof of a supply-side hole: when a agency with classified-project clearance whose merchandise have been introduced to Xi Jinping submits falsified bidding supplies, one thing is deeply fallacious with the PLA’s AI procurement. Washington wants sharper metrics to truly consider China’s progress in army AI – incorporating knowledge factors like repeat award patterns, defense-relevant hiring traits, and verified technical supply data – slightly than reacting to headline contract counts.
A 266-person IT firm from China’s coal belt received a few of the PLA’s largest AI integration contracts after which obtained caught faking its bids. The 100 Belief story is just not proof of Chinese language army AI success or failure. It’s a window right into a extra advanced actuality: Beijing’s strategic intent on army AI is unambiguous, however the market mechanism executing that intent is concurrently producing real functionality and inflated alerts.
Calibrating U.S. coverage to this differentiated actuality, distinguishing structural integration from procurement theater, issues greater than reacting to the subsequent headline about what number of contracts the PLA has signed. The 2 traits of Chinese language non-public firms recognized right here – dependence on state-controlled compute and capability for fast integration – aren’t distinctive to 100 Belief. They describe a structural sort, and Washington would profit from monitoring which companies match this profile slightly than reacting to particular person procurement headlines.
However the analytical window this text depends on could not stay open. Beijing has already restricted abroad entry to company registration databases, judicial data, and educational repositories. If army procurement knowledge follows the identical trajectory, the chance to tell apart structural integration from procurement theater will shut earlier than Washington has totally calibrated its response.
.png?trim=0,0,0,0&width=1200&height=800&crop=1200:800 "Ex-San Diego reporter shot at victims on roadside after asking ethnicity: Officials")
