Loading...
Back to Insights
DefenseJanuary 2, 2026·7 min read

China's AI Drone Swarms Are Performing Well in War Games

Autonomous Systems Capable of Coordinated Operations Without Human Oversight Are Being Tested

China's AI Drone Swarms Are Performing Well in War Games
BR
Brian Rutherford
FAA Part 107 Pilot USMC Reconnaissance Veteran C-UAS Consultant

The New York Times Editorial Board doesn't typically issue warnings about military technology. When they do, the assessment merits attention.

December 2025. The Times publishes "Overmatched"—a series analyzing how cheap, autonomous drone swarms are reshaping warfare faster than the Pentagon can adapt. The central finding: the United States is falling behind China and Russia in this capability area.

In war games simulating a Taiwan conflict, the side with superior swarm capability wins. Consistently. The Pentagon has run these simulations. The results raise questions about current defense posture.


Summer 2025 Demonstrations

Summer 2025. The People's Liberation Army conducted drone swarm tests that merit examination.

June 2025. China flight-tested the Jiu Tian SS-UAV—a high-altitude, long-endurance drone mothership with a 25-meter wingspan. It carries 100 to 150 smaller loitering munitions in dual internal bays. One aircraft. Over a hundred weapons. Deployed with minimal direct operator control.

This isn't a concept. It flew.

The PLA isn't building individual drones. They're building systems that deploy swarms as a single weapon. Release, disperse, coordinate, engage. Limited human input. Coordinated lethality.

A PLA Air Force senior colonel described recent tests as "a very important step toward the PLA having true swarm capabilities."

The distinction matters: not remote-controlled groups, but autonomous systems that coordinate, adapt, and engage without operators managing each platform individually.


How Swarms Change Calculations

Traditional air defense assumes discrete targets. Identify a threat. Track it. Engage it. Move to the next.

Swarms challenge that model.

When 100 platforms approach simultaneously, sharing data, adjusting formations, selecting targets collectively—point defense systems face saturation. Radar tracks become difficult to manage. Engagement timelines compress beyond comfortable human reaction capacity.

A $2 million surface-to-air missile engaging a $500 drone isn't efficient defense. It's arithmetic working against the defender. Expend your magazine on the first wave. The second wave arrives with reduced opposition.

RAND simulations show this pattern. Large numbers of relatively cheap unmanned aircraft operating as fully autonomous swarms prove "absolutely essential for coming out on top" in Taiwan scenarios.

The side with swarm capability prevails. The side without it loses expensive platforms to cheap, distributed attackers.


The Autonomy Question

The Times editorial addressed directly: fully autonomous swarms capable of "hunting and killing without any human oversight" remain in advanced testing.

That gap is narrowing.

Andrew C. Weber, who oversaw nuclear, chemical, and biological defense programs under President Obama, observed: "The speed of warfare will soon outpace human ability to control it."

This isn't speculation. Human cognition processes information in hundreds of milliseconds. AI systems operate in microseconds. When swarms engage each other, the engagement cycle compresses to speeds where human decision-making becomes a bottleneck rather than a safeguard.

The United States officially maintains "human-on-the-loop" requirements for lethal decisions. A human must authorize engagement.

China may not apply similar constraints. Beijing appears to prioritize speed and efficiency, building systems designed to act quickly with minimal human intervention.

If one side requires human authorization and the other doesn't, the autonomous side operates faster. In combat, speed advantages matter.


The DeepSeek Development

January 2025. Chinese AI startup DeepSeek matched American competitors' performance at a fraction of the cost. That breakthrough had immediate military applications.

The PLA integrated DeepSeek into weapons systems—autonomous combat vehicles, robot dogs, and drone swarms.

Research from Beihang University shows DeepSeek enabling drone swarms to fly in formation, identify targets, and adjust flight paths without human guidance. Xi'an Technological University papers claim DeepSeek can evaluate 10,000 battle scenarios in 48 seconds.

Two dozen military tenders and patents document Chinese efforts to integrate AI into drones for autonomous target recognition, tracking, and formation flying with minimal human intervention.

This is the convergence point: efficient AI that works, affordable drones that scale, and doctrine that doesn't require human oversight for every engagement.


The Taiwan Scenario

The PLA views drone swarms as potentially solving their most difficult challenge: a potential operation against Taiwan.

Taiwan sits 100 miles from mainland China. The Taiwan Strait represents one of the most defended bodies of water on Earth. A conventional amphibious assault faces layered anti-ship missiles, air defense systems, and minefields.

Swarms could change the calculus.

A pre-operation strike using thousands of autonomous drones could saturate Taiwan's air defenses. Loitering munitions could search for mobile missile launchers. Coordinated swarms could stress naval point-defense systems.

The 2020 Air Force Wargaming Institute simulation showed exactly this. An autonomous drone swarm using distributed mesh networking was cited as a key contributor in that scenario.

But that simulation assumed U.S. swarm capability superiority. Four years later, that assumption may not hold.

CNA analysis states directly: "Without deep magazines of substantially enhanced counter-drone capabilities, the United States risks having its distributed warfighting strategies overwhelmed by massed Chinese drone attacks, and the United States could lose a war over Taiwan."


The Production Reality

China isn't just testing swarm technology. They're producing it.

The Jiu Tian mothership concept puts 100+ drones on a single launch platform. Multiple motherships create swarms in the thousands. The technology has been demonstrated. Manufacturing follows from there.

Meanwhile, the Pentagon's Replicator program delivered hundreds of drones instead of thousands. The software to coordinate multi-manufacturer swarms doesn't exist in inventory. The response to China's swarm capability involves reorganizing programs and extending timelines.

This is the mismatch the New York Times identified: China advances while we adapt. They field capability while we run simulations. The gap may widen with each budget cycle.


The Ethics Dimension

American doctrine embeds human oversight in lethal decision-making. Someone authorizes engagement. The machine executes, but a human decides.

That requirement has operational costs. Human decision loops add latency. Oversight requirements add complexity. Ethical constraints shape system design.

China has publicly committed to maintaining human control over weapons systems. Whether that commitment persists in actual operations is uncertain.

Their research, patents, and testing point toward systems with "minimal human intervention." They're building for speed. The United States builds for oversight.

When those systems meet in contested airspace, one approach will prove more effective. The outcome has implications beyond Taiwan.


Recommended Responses

Accept asymmetric timelines. China is fielding capability while we study it. Matching their production rates requires examining acquisition processes.

Address the software challenge. Swarm capability requires coordination across platforms from different manufacturers. That software needs development. Without it, drones are expensive individuals, not coordinated capabilities.

Invest in counter-swarm. If offensive swarms are difficult to field quickly, defensive capability becomes more important. Electronic warfare, kinetic options, directed energy—whatever proves effective. Current counter-UAS posture assumes dozens of threats, not thousands.

Engage the autonomy question. The ethical constraints we apply may be appropriate. They're also operational considerations when facing adversaries without equivalent constraints. Doctrine should acknowledge this tension.

Maintain program continuity. Replicator evolved. DAWG replaced elements. Drone Dominance was launched. Each reorganization consumes time that could be used for production. Stability enables execution.


Assessment

China is testing AI-enabled drone swarms that coordinate autonomously, select targets without direct human input, and deploy at scales that challenge conventional defenses.

The Pentagon knows this. The war games demonstrate it. The production comparisons confirm it.

The New York Times warned of a capability mismatch. The simulations show challenging scenarios. Intelligence assessments identify potential action timelines.

Autonomous swarms capable of coordinated operations without individual human oversight aren't theoretical. They're in advanced testing across the Taiwan Strait. Whether they're employed in combat is a question of Chinese decision-making, not American preparation.

Readiness gaps exist. Acknowledging them is the first step toward addressing them.


Brian Rutherford covers military technology with attention to capability gaps that assessments acknowledge but programs struggle to address.


Sources

#AI#DroneSwarms#China#Pentagon#Taiwan#AutonomousWeapons#MilitaryAI#PLA#WarGames#NationalSecurity
Share this article

Stay Informed on Drone Threats

Weekly intelligence on C-UAS developments, regulatory changes, and security insights from a Part 107 pilot and defense veteran.

Join 500+ security professionals. Unsubscribe anytime.