How Civilian Tech Is Powering China’s Army to Rival the US.

Georgetown’s CSET analyzed 2,857 PLA AI-related award notices (Jan 2023–Dec 2024), identifying 1,560 suppliers and a core of 338 repeat awardees. The findings show China’s military is systematically harnessing civilian firms and universities for AI, tightening competition with the U.S. Army and complicating export controls and due-diligence screens.

The Center for Security and Emerging Technology (CSET) published an issue brief in September 2025 detailing how the PLA mobilizes civilian AI using a dataset of 2,857 award notices from January 2023–December 2024. The study identifies 1,560 organizations that won at least one award, with focused analysis on 338 repeat awardees, and categorizes suppliers as SOEs, research institutions, and nontraditional vendors (NTVs), with NTVs winning 764 contracts. This diversified supplier base expands the PLA’s access to dual-use AI and raises the bar for export controls, research funding safeguards, and outbound-investment screening.
Follow Army Recognition on Google News at this link

The GJ-11 stealth attack drone made its public debut in Beijing during the October 1, 2019, National Day parade. (Picture source: China MoD)

Beijing treats military-civil fusion as a full-scale armament policy. The objective is to convert civilian building blocks into deployable military effects at short notice, to keep pace with the US Army. CSET shows that state owned conglomerates and their affiliated institutes still dominate contract value, with CETC, CASC, NORINCO, CSSC, and AVIC at the forefront. The Seven Sons of National Defense and entities within the Chinese Academy of Sciences are among the frequent recipients, anchoring modernization in organizations experienced with complex programs. At the same time, a cohort of nontraditional vendors, younger and dual-use oriented, wins many awards by count, especially for subsystems, software, and data services. This mix of heavy incumbents and agile suppliers gives procurement flexibility and accelerates the diffusion of AI at the unit level.

In operational terms, the first effect is more sensors and more automation near frontline units. Awards to CETC and its institutes emphasize C4ISR functions, data fusion, and simulation, which sit at the core of the detect, identify engage cycle. CASC delivers payloads and simulators for UAVs, providing tools to prepare, validate, and then employ swarms or ISR patrols without constant dependence on higher echelons. In practice, a brigade can equip more drones with tailored payloads, route these feeds into a normalized data architecture, and obtain fire corrections in tens of seconds rather than minutes. The terrain picture sharpens, and massing forces becomes costlier for an opponent.

Nontraditional vendors fill gaps and add redundancy. They supply embedded vision modules, speech processing, multiplatform simulation tools, and data integration services. Their number and geographic spread allow units to gain quick incremental improvements, a more robust model to recognize a silhouette on video, a more realistic simulator for FPV team training, and a mapping engine that stitches orthomosaics within minutes after an overflight. Tactically, this raises the density of “eyes,” increases the frequency of firing opportunities, and expands terminal guidance options. Logistically, smaller and more numerous contracts sustain a short iteration rhythm that fits the high wear of drones and sensors in contested zones.

This procurement model shapes maneuver. At company to battalion level, the PLA can standardize very short cycles to launch a micro UAS, detect, classify locally on an edge board, transmit coordinates, fire, correct by video feedback, then displace. The adversary must disperse assets, vary routes, tighten emissions discipline, and accept shorter movement windows. At brigade and theater level, C4ISR normalization driven by large groups improves coherence across heterogeneous sensors. Fire planning and combined arms coordination gain responsiveness even under jamming, because more processing happens at the edge.

The political economy of this approach matters as much as the technology. AI related awards rise over the observed period, with an increase in published notices and declared values from May to December year over year. Supplier distribution is concentrated in Beijing, with strong clusters in Jiangsu, Sichuan, Shaanxi, and Guangdong, where many nontraditional vendors are located. This network allows the PLA to source near existing industrial bases, shorten testing and integration timelines, and draw on civilian AI talent pools. For Western control policies, the supplier dispersion and the frequent absence of sanctions on these entities make screening technology flows more complex.

On the ground for the PLA, this means more drones and specialized payloads down to small units; onboard AI processing that reduces dependence on high bandwidth; simulators and test benches that speed training and validation; a more uniform C4ISR backbone that spreads the tactical picture quickly; and steady use of civilian vendors to keep upgrade cadence. For an opponent, the response must combine layered counter-UAS, electromagnetic protection, visual and radio deception, and logistics that are able to absorb attrition under near-continuous observation.

Reference points from CSET support these conclusions. The fifteen most awarded entities include a majority of state-owned enterprises and affiliated institutes, with CETC prominent. Nontraditional vendors account for nearly three-quarters of the 338 entities studied and receive the highest number of awards, although the average award value is lower. The number and value of published notices increase markedly between May and December from one year to the next. More than a quarter of entities are based in Beijing, with dense concentrations in Jiangsu, Sichuan, Shaanxi, and Guangdong, illustrating the overlap between legacy defense basins and civilian AI hubs.