A Georgetown analysis of roughly 9,000 PLA procurement documents published between January 2023 and December 2024 names the US capabilities Beijing wants to close, including naval, space based, and signals intelligence systems, giving Washington a public blueprint of where the…
A dataset released in February 2026 by Georgetown's Center for Security and Emerging Technology (CSET) turns what was once rumor and reverse-engineering into a paper trail. Researchers analyzed roughly 9,000 Chinese-language procurement requests published by People's Liberation Army (PLA) units between January 2023 and December 2024, and the picture that emerges is not a single "AI war machine" but a long shopping list of specific American strengths Beijing is trying to neutralize.
The categories are explicit. CSET documents demand for AI that supports command-and-control, cyber operations, intelligence, surveillance and reconnaissance (ISR), targeting, facial and gait recognition, recovery of deleted data, deepfake detection, and what the report calls "cognitive operations." Space-based and undersea systems show up repeatedly. So does the unglamorous plumbing: data labeling, model training pipelines, and the chips required to run them.
This is not a forecast of battlefield performance. It is a record of what PLA units asked to buy. That distinction matters, because the Pentagon's 2025 Annual Report to Congress on Military and Security Developments Involving the People's Republic of China (PRC) frames PLA AI as part of a broader acceleration in hypersonics, biotech, and other military technologies, while stressing that announced defense spending has nearly doubled since Xi Jinping's first full year in power without resolving whether demand translates into fielded capability.
That gap between demand and fielding is where the procurement trail earns its keep. Most PLA AI capability discussions lean on inference from demos, leaked papers, or US intelligence assessments. CSET's open-source dataset instead names the vendors, the requesting units, and the dollar ranges for the categories Beijing wants to buy into. A reader can watch specific PLA offices over time, see which contracts go to which state-owned defense conglomerates, and compare those patterns against the US defense industrial base.
The US response, as documented, runs through the White House's "America's AI Action Plan", released in July 2025 with an official PDF and a stack of executive orders. The plan reorients federal AI procurement, leans on deregulation, and pushes infrastructure buildout. Legal briefings from Skadden, Orrick, and CRA GovAffairs describe how the new posture changes what the Department of Defense (DoD) and the broader defense industrial base (DIB) will be asked to deliver.
Whether that posture lines up with the demand signal in CSET's dataset is the open question. The PLA's publicly stated gap is in space, undersea, ISR, and cyber. The AI Action Plan is built around domestic buildout, export controls, and a deregulated federal market. DefenseScoop's December 2025 coverage of the Pentagon report flagged Beijing's progress on large language models as the most concrete AI-specific concern; USNI News noted the same report's broader framing of PLA modernization across hypersonics, biotech, and AI as a single acceleration curve.
Two things are worth watching next. First, whether future CSET updates extend the procurement window past December 2024, which would let readers see whether demand for space and undersea AI systems continued to climb or shifted toward the large-model work that DefenseScoop flagged. Second, whether the US AI Action Plan's procurement language is followed by named contracts in those same categories, or whether the gap CSET documents in Beijing's shopping list ends up being closed by a different part of the US budget entirely.
The dataset's quiet implication is that readers now have a public, falsifiable measure of PLA AI demand. Demand is not capability. But for the first time, the gap between what the PLA says it wants and what the PLA actually fields is a number a non-government analyst can try to count.