Sharp Sword: China on leading edge of AI-powered air war

China’s People’s Liberation Army Air Force (PLAAF) has released its first footage of a J-20 stealth fighter flying in formation with a GJ-11 stealth attack drone, signaling the country’s entry into pilot-drone integration.

The video, aired to mark the PLAAF’s 76th anniversary, was reported by the South China Morning Post (SCMP), which said it marked China’s public debut of manned-unmanned teaming (MUM-T) technology.

The footage shows the GJ-11—codenamed Xuanlong 08, also known as Sharp Sword—operating with the J-20 Weilong and the J-16D electronic-warfare fighter. Radio exchanges between the aircraft indicate coordination via a secure datalink, suggesting the J-20 could direct the GJ-11 to extend its range and strike capability inside contested airspace.

Military analyst Song Zhongping told SCMP that the three aircraft could form a coordinated combat triad combining stealth penetration and electromagnetic suppression. The GJ-11, first revealed at the 2019 National Day parade and seen again in 2025 with redesigned tail fins, is believed to function as both a stealth strike drone and a “loyal wingman.”

While state media claimed the drone could coordinate autonomously through a digital battlefield network, Song cautioned that the technology remains in a training phase. Even so, the display underscored China’s determination to integrate AI, networking and stealth into a coherent combat web for future wars.

Across the Pacific, the US is racing to operationalize its own version of pilot-drone integration. In a July 2025 report, The War Zone (TWZ) revealed that the US Air Force plans to make the F-22 Raptor the first operational controller of Collaborative Combat Aircraft (CCA) drones under its Crewed Platform Integration (CPI) program.

According to TWZ, 143 combat-capable F-22s will be equipped with tablet-based interfaces and supporting hardware costing about USD 86,000 each, allowing pilots to direct autonomous loyal wingmen via the F-22’s secure Inter-Flight Data Link.

The initiative aligns with broader CCA development efforts involving General Atomics and Anduril, with plans to field 100 to 150 Increment 1 drones and eventually expand to 1,000 across future phases. Anduril’s YFQ-44A and General Atomics’ YFQ-42A have both completed initial flights, with weapons testing and prototype production set for 2026.

At roughly US$25–30 million each—far below the $81 million cost of an F-35—CCAs are intended to augment crewed fighters by extending range, absorbing attrition, and saturating enemy defenses.

The PLAAF’s concept is similarly ambitious but less mature. In an August 2025 report for the RAND Corporation, Shanshan Mei writes that the PLA views MUM-T as a force-multiplying concept for a Taiwan conflict, particularly during an “island-landing” campaign.

According to Mei, PLA researchers simulated missions in which a J-20 or J-16 leads formations of up to sixteen drones to jam, strike, or reconnoiter enemy air defenses, with the drones relaying targeting data and acting as decoys to saturate Taiwan’s radar network.

She adds that these simulations remain theoretical, focused on doctrinal modeling rather than fielded practice, illustrating that the PLA’s ambition to overwhelm enemy decision-making still exists primarily on paper.

A September 2025 Hudson Institute report by Bryan Clark and Dan Pratt outlines how the US may employ MUM-T in a Taiwan scenario by pairing crewed fighters with attritable CCAs in a phased “edge-pulse-core” sequence to reduce risk and dilute PLA targeting.

Clark and Pratt argue that CCAs acting as “weapon trucks” carrying sensors, electronic-warfare gear, or munitions would operate from the same airfields as fighters, augmenting rather than replacing them. This layered use of attritable assets would, they write, expand sensor coverage and strike capacity while allowing pilots to hold back until conditions are set.

In an April 2025 report for the Center for Strategic and Budgetary Assessments (CSBA), analyst Travis Sharp adds that 36 CCAs per site in Japan and the Philippines could balance sortie generation and attrition risk.

In the Journal of Indo-Pacific Affairs that same month, Nicholas Stockdale echoes this logic, stating that five CCAs per manned fighter would serve as force multipliers, complicating China’s targeting matrix and raising its costs of engagement.

Yet neither side is without constraints. An April 2025 report for the China Aerospace Studies Institute (CASI) by John Chen and Emilie Stuart notes that China’s MUM-T concept remains rudimentary and unproven in practice.

They cite PLA analysts who warn that large-scale operations risk command-and-control confusion and depend on vulnerable datalinks, while centralized command and limited training restrict autonomy. They add that high costs, weak doctrinal guidance, and an overly scripted command structure underscore how the current PLA experimentation still relies on basic, non-autonomous coordination between manned fighters and drones.

Sharp, in his April 2025 CSBA assessment, also warns that the US program faces its own bottlenecks. He writes that CCA-enabled MUM-T will stall unless development, deployment, and employment advance together—three steps that currently lag behind earlier planning and resource phases.

Sharp points to unsettled designs and cost targets ranging from $10 to $30 million per unit, and to “gold-plating”—the tendency to inflate costs by adding excessive features—as a potential threat to the logic of affordable mass.

He further notes unaccounted bills for software, basing, manning, and sustainment, warning that technology alone cannot deliver reliable fighter-drone teaming without organizational and logistical adaptation.

China’s J-20–GJ-11 pairing shows tangible progress in linking stealth aircraft and drones, but limited autonomy and top-down control still cap combat payoff. The US is moving to operationalize control via retrofitting F-22s and early CCA prototypes, yet cost, design, basing, sustainment and employment shortfalls remain unresolved.

In a Taiwan fight, the edge goes to whichever side can convert these trials into a resilient, scalable, and attritable human-machine strike network.