The C4ISR Architecture Behind China's Autonomous Airpower
The Datalinks, Hardware, and Algorithms Enabling the PLA's Autonomous Kill Web and How They Compare to U.S. Efforts.
This is the first in a two-part series. Part I deconstructs the technology that makes swarms and collaborative aircraft work. Part II will analyze how the PLA intends to use them in a future conflict.
From Ukraine’s “Operation Spider Web” with its hidden attack drones to Israel’s “Operation Rising Lion” using prepositioned loitering munitions, autonomous warfare and unmanned systems have become central to modern conflict discussions. Given this shift, I want to break down how Chinese autonomous air combat systems (自主系统)—particularly massed drone swarms (蜂群) and advanced Collaborative Combat Aircraft (CCAs)—actually function.
Before we begin, it is critical to define what a “drone swarm” truly is. A swarm is a system defined by decentralized intelligence. Unlike a centrally-commanded formation where a single operator directs each drone, a true swarm operates on the principle of emergent behavior. Individual drones follow a set of simple, local rules, and through their interactions with one another—without a leader—a complex and intelligent group behavior emerges. Think of a flock of birds, where no single bird is in charge, yet the flock moves as a cohesive, adaptive whole. This principle is the key to understanding the specific engineering choices that follow.
So let’s look at these systems from the ground up, examining the core components that make them work: datalinks, onboard processors, algorithms, and the C4ISR architecture that enables human-machine partnerships. By comparing the People's Liberation Army (PLA) approach with parallel U.S. programs, we can see how two different philosophies are shaping the future of warfare.
Datalinks and Networking
An autonomous swarm is a network first and a collection of airframes second. Its ability to collaborate is entirely dependent on a resilient, secure “nervous system” capable of handling immense data flows in a hostile electromagnetic environment.
The PLA’s Approach
The PLA’s networking philosophy is built for a fight in contested airspace. The core technology is the Mobile Ad-Hoc Network (MANET/自组网), a “self-organizing network” that allows individual drones (无人机) to connect directly with each other without relying on fixed ground stations. Research from PLA-affiliated institutions focuses heavily on developing bespoke military-grade protocols to manage the high mobility and frequent link breakages inherent in a dynamic swarm.
Resilience against jamming is a primary design driver. The state-owned enterprise China Electronics Technology Group Corporation (CETC) employs foundational anti-jamming (抗干扰) techniques like frequency-hopping spread spectrum (FHSS). For operations over the horizon, the PLA relies on its sovereign Tiantong-1 (天通一号) satellite communication system. While this provides independent command and control, its data rates peak at ~384 kbit/s. This represents a critical constraint, insufficient for streaming high-fidelity sensor data from multiple assets and a key factor driving China’s emphasis on local, onboard processing.
U.S. Architecture
The U.S. approach builds on its legacy of tactical datalinks but is rapidly evolving. While the ubiquitous Link 16 provides a baseline for interoperability, the key enabler for modern CCA operations is the Tactical Targeting Network Technology (TTNT), a high-speed, low-latency, IP-based mesh network. However, the most significant differentiator is the top-down mandate for a Modular Open Systems Approach (MOSA). For airborne platforms, the crucial standard is Open Mission Systems (OMS), which functions like a universal software adapter. This allows DoD to use a Lego-like approach, integrating the best sensor from any vendor onto any compliant platform, preventing vendor lock and accelerating upgrades. To handle BLOS connectivity, the U.S. is deploying the Proliferated Warfighter Space Architecture (PWSA), a large LEO satellite constellation that acts as a powerful space-based relay for datalinks like TTNT.
Onboard Processing and AI Hardware
If datalinks are the nervous system, onboard processors are the distributed brains. The ability to perceive, decide, and act at the tactical edge depends entirely on the size, weight, power, and computational performance (SWaP-C) of this hardware.
The PLA’s Approach
Driven by the need to perform complex AI tasks locally, China is pursuing a hardware-first strategy focused on highly specialized, power-efficient Application-Specific Integrated Circuits (ASICs). This is a direct response to the physical constraints of their intended platforms. A small, attritable kamikaze drone has a tiny power budget and physical footprint, making it impossible to use power-hungry, general-purpose GPUs. Its edge processor must be hyper-efficient to run computer vision algorithms on its simple electro-optical/infrared (EO/IR) camera feed. This has spurred a military-civil fusion effort to create novel, hyper-efficient chips.
Companies like Houmo Intelligence (后摩智能) are pioneering compute-in-memory (CIM/存算一体), an architecture that performs AI calculations directly within the memory to drastically cut power consumption. Its H30 chip claims an exceptional 7.3 TOPS/W (Tera Operations Per Second per Watt), an efficiency ideal for small, battery-powered drones carrying explosive payloads. This focused investment in custom hardware is a strategic bet that architectural innovation can leapfrog established Western chip designs.
U.S. Architecture
The U.S. approach is less about designing new chips and more about leveraging its world-leading commercial semiconductor industry and fostering a new layer of “autonomy” companies. This is possible because U.S. CCAs, like the YFQ-42A and YFQ-44A, are often quite large (comparable in size to a small fighter jet). This larger airframe allows for more weight, power, and cooling, enabling the use of high-performance Commercial-Off-The-Shelf (COTS) processors from giants like NVIDIA and Intel. These powerful modules can process massive amounts of data from advanced payloads, such as high-resolution Synthetic Aperture Radar (SAR) or sophisticated signals intelligence (SIGINT) packages.
The defining feature of the U.S. ecosystem is the rise of companies like Shield AI and Anduril, which provide complete, vertically integrated autonomy solutions. Shield AI’s Hivemind, for example, is a platform-agnostic “AI pilot” delivered as a tightly integrated hardware-software system. Their value is not just a chip, but the entire software development kit, simulation environment, and validated autonomy stack that runs on powerful COTS hardware.
Logic and Algorithms
Moving from silicon to software, the effectiveness of any autonomous system is ultimately determined by the sophistication of its algorithms. Here, the two countries’ research priorities reveal potentially divergent operational concepts.
The PLA’s Approach
A survey of Chinese academic literature reveals a strong preference for decentralized, bio-inspired algorithms suitable for managing large swarms of relatively simple agents. Ant Colony Optimization (ACO) is frequently cited for decentralized path planning, simulating the way ants use pheromone trails to collectively find optimal routes. These choices suggest a focus on emergent coordination. The goal is to solve the key challenges of deploying a large, attritable swarm—optimizing routes, avoiding collisions, and achieving collective behavior from simple, local rules. This logic is well-suited for platforms with limited individual sensing and processing power.
U.S. Architecture
The U.S. has pursued a more program-driven approach, with DARPA and the Air Force Research Laboratory (AFRL) running foundational initiatives to create expert AI agents. DARPA’s Air Combat Evolution (ACE) program famously trained a deep reinforcement learning agent to defeat an experienced human F-16 pilot in a simulated dogfight. This demonstrates a mastery of creating a highly capable, individual AI agent that can achieve superhuman tactical performance. This suggests a “quality over quantity” dynamic at the algorithmic level. While China appears to be optimizing for the logic of the swarm, the U.S. is optimizing for the logic of the individual expert wingman.
Manned-Unmanned Teaming
The ultimate expression of this technology is Manned-Unmanned Teaming (MUM-T), or what the PLA refers to as manned-unmanned collaboration (有人/无人协同), where a human pilot orchestrates the actions of autonomous wingmen.
The PLA’s Platform-Centric Vision: The PLA’s MUM-T concept is centered on the Chengdu J-20S, the world’s first announced twin-seat stealth fighter. The second seat is explicitly for a Weapon Systems Officer (WSO) to manage the unmanned formation. This approach allows a wingman like the GJ-11 or FH-97A to act as a forward sensor node, using its own sensors to detect a target and pass “fire-control grade” data back to the J-20S, which can launch a weapon from a safer distance. This system is deeply integrated but risks being rigid and closed.
The U.S. Architecture-Centric Vision: The U.S. approach, mandated by OMS, is fundamentally different. The goal is to create a universal standard where any compliant CCA from any vendor can team with any compliant manned aircraft. This prioritizes flexibility and competition. The U.S. is betting that its more mature AI will enable a single pilot to act as a “human-on-the-loop” commander, trusting the AI to handle the complex tactical execution and thus offloading the cognitive burden. The “second seat” in the U.S. model is the AI itself.
Technological Hurdles
A review of Chinese technical journals reveals a candid internal discussion about the significant hurdles the PLA faces. Chinese sources admit that maintaining stable network links for large, dynamic swarms is a “great challenge” that can lead to mission-compromising collisions. The limited endurance of battery-powered UAVs is consistently cited as a primary bottleneck. Perhaps most significant is the acknowledged gap between lab simulations and real-world performance. As a China Aerospace Studies Institute report notes, multiple sources point to a lack of authoritative doctrine and training, with one researcher bluntly stating that many soldiers “do not dare to use and do not know how to use” (不敢用、不会用) UAVs. This reveals that the primary hurdle may not be inventing the technology, but building the human and organizational systems to wield it effectively.
Conclusion
This assessment reveals a fundamental divergence in development philosophies. The United States’ advantage is rooted in its mature, software-defined ecosystem, leveraging open standards and world-class commercial hardware to build flexible, high-performance systems. This approach excels in creating sophisticated, high-end capabilities like the individual AI-piloted Collaborative Combat Aircraft.
In contrast, China is pursuing a hardware-first strategy, making focused bets on novel, power-efficient edge ASICs. This approach, born of necessity and strategic ambition, is optimized to solve the critical SWaP constraints of massed, attritable drones. It could enable the PLA to field intelligent swarms at a scale the U.S. might struggle to match.
But technology alone does not confer victory. The most advanced hardware and software are only as effective as the doctrine that guides their use. The engineering choices detailed here directly shape and are shaped by operational concepts. This raises the critical next question: how will these competing technical architectures be employed on the battlefield?
In Part II of this series, we will answer that question. We will shift our focus from engineering to execution, analyzing the PLA’s operational concepts for employing these autonomous systems across multiple domains.
Thanks for reading. If you found this analysis valuable, the best way to support my work is to like and share it. For more analysis on C4ISR and military technology, please subscribe to my Substack, Orders and Observations, and follow me on X/Twitter @mcasey_asia for more frequent commentary.
Thanks. Were I a tad younger I can imagine this article would change my career plans. I wonder about the application of the ideas here, including the tension between two approaches, to non-military applications. Am I mistaken to think, for example, that dynamic supply challenges have something in common with dog fights?