# Drone Swarms
> Coordinated masses of autonomous unmanned systems that the US and China are racing to deploy at scale, reshaping military doctrine and outpacing governance frameworks worldwide.

**Meta:** type: reference · date: 2026-07-03 · heads:  · 4 takes · 2 lenses · 2 regions

## What it is

A drone swarm is a coordinated group of unmanned systems, typically dozens to hundreds of units, that operate using distributed algorithms to share sensor data, divide tasks, and adapt to losses without a human directing each unit individually. The military proposition is volume over exquisite platforms: a US$1 million interceptor missile cannot economically defeat a swarm of US$500 drones. Three domains are active as of mid-2026: aerial (the most mature), unmanned surface vessels, and ground robots. Key actors are the US Department of Defense, China's People's Liberation Army, and a cluster of US defence-technology firms including Anduril, Auterion, and Shield AI.

## History

DARPA's OFFensive Swarm-Enabled Tactics (OFFSET) program, launched in 2017, tested swarms of up to 250 small unmanned aircraft and ground robots with US Army infantry units across six field experiments of increasing complexity, completing in 2021. The program's core insight was that swarm commanders need a "swarm grammar" to issue intent, not individual maneuver orders. Ukraine's mass FPV drone use from February 2022 proved the attritable-at-scale logic on a live battlefield, pushing the concept from research into active doctrine. The US DoD formalised the industrial bet with the Replicator initiative, announced August 28, 2023, targeting thousands of all-domain attritable autonomous systems by August 2025 to offset China's numerical advantages in mass. A second Replicator line, focused on counter-UAS, was added by US Defense Secretary Lloyd Austin in September 2024. The US Congressional Research Service subsequently found that only "hundreds" of systems were delivered by the August 2025 target date.

## Current state

As of July 2026, the competition is a live US-China race. The US DoD Chief Digital and AI Office is running Swarm Forge, a solicitation seeking AI agents that autonomously coordinate the Find-Fix-Finish targeting cycle, named a pace-setter in US Defense Secretary Hegseth's January 2026 AI memo. In June 2026, the DoD held "Crucible," an industry swarm stress-test backed by a US$100 million Pentagon challenge. At the AWE26 allied exercise, US, UK and Australian swarms shared sensor data live for the first time, a step toward interoperable allied autonomy. China's People's Liberation Army ran its first autonomous maritime swarm drill on March 25, 2026 off Zhuhai, using L30 unmanned surface vessels that detected and contained a simulated intruder with limited operator input after launch. The PLA has also demonstrated a 200-drone aerial swarm commanded by one soldier, with units cooperating autonomously after losing operator contact.

## Relationships

[Swarm Forge and Crucible](/zh/n/pentagon-swarm-forge-crucible) are the US DoD's most aggressive push toward full autonomous targeting to date; the explicit goal is "end-to-end autonomous" completion of Find-Fix-Finish, pushing against human-control norms before US policy has settled what that means in practice. The [US Army's US$20B Anduril enterprise vehicle](/zh/n/anduril-army-20bn-counter-drone) is the counter-swarm mirror: Anduril's Lattice AI suite is now the US Army's primary counter-UAS command-and-control platform, built on the assumption that adversary swarms must be defeated at machine speed. The [China L30 maritime drill](/zh/n/china-l30-maritime-swarm) is Beijing's clearest public signal that autonomous naval swarms are a Taiwan-contingency asset. SIPRI has tracked autonomous-weapons governance under the UN Convention on Certain Conventional Weapons since 2014; as of July 2026, no binding multilateral treaty governs lethal autonomous systems.

## What to watch

Three questions will define the swarm trajectory into 2027. First, Crucible results: which vendors advance toward Swarm Forge awards, and whether "end-to-end autonomous targeting" clears US human-control policy without triggering a formal override. Second, whether Replicator's deployment gap, hundreds versus thousands, narrows before China's maritime and aerial swarm programs scale to contested-sea use in the Western Pacific. Third, governance: the UN CCW Group of Governmental Experts' 2026 session is the next formal venue where states must decide whether to open binding treaty negotiations on autonomous weapons, a threshold no major military power has yet accepted.

## Regional takes (batched by bias / lens)

### official record
- **DARPA: OFFensive Swarm-Enabled Tactics (OFFSET)** (United States, en) — DARPA's official OFFSET program page documenting the 2017-2021 foundational US swarm program: six field experiments testing swarms of up to 250 UAS and ground robots with small-unit infantry in urban terrain, structured around a swarm-grammar human-machine interface.
  Source: https://www.darpa.mil/research/programs/offensive-swarm-enabled-tactics
- **SIPRI: Autonomy in Weapon Systems** (Global, en) — SIPRI's institutional research hub on autonomous weapons governance, tracking technological trajectories and legal challenges under the UN CCW framework since 2014, funded by Canada, Germany, Netherlands, Sweden and Switzerland.
  Source: https://www.sipri.org/research/armament-and-disarmament/emerging-military-and-security-technologies/autonomy-weapon-systems
- **Congressional Research Service: DOD Replicator Initiative (IF12611)** (United States, en) — CRS analysis of the US DoD Replicator initiative, announced August 2023, targeting thousands of attritable autonomous systems by August 2025 to offset China's numerical advantages; notes only hundreds of systems were delivered by the deadline.
  Source: https://www.congress.gov/crs-product/IF12611

### US defence-policy analysis
- **Defense One: Pentagon leans on drone swarms with $100M challenge** (United States, en) — Details the Pentagon's US$100 million swarm challenge and the doctrinal bet that AI-coordinated swarms can locate and destroy targets faster than human-paced systems, while noting Replicator's earlier swarm promises remain partly unmet.
  Source: https://www.defenseone.com/technology/2026/01/pentagon-leans-drone-swarms-100m-challenge/410742/

## Across the graph
- Related: [[pentagon-swarm-forge-crucible]], [[china-l30-maritime-swarm]], [[anduril-army-20bn-counter-drone]]
- Entities: Drone Swarms, US Department of Defense, China, Anduril, Counter Uas

---
Canonical: https://rbtfl.xyz/zh/n/drone-swarms-dossier