Simulations run by western think tanks and military offices increasingly place unmanned systems at the center of any credible defence of Taiwan. The image is familiar now: thousands of cheap, networked air, surface and subsurface vehicles forming a distributed sensing and effects layer ahead of manned assets, complicating an adversary’s targeting problem and buying time for reinforcement. This is not science fiction. Senior U.S. commanders have publicly described this contingency in stark terms and planners have begun funded attempts to make that contingency tangible.
What the wargames actually show is less theatrical than the phrase “hellscape” implies and more revealing for doctrine. RAND and allied Air Force simulations that have been discussed in public fora simulate the swarm as a layered, cooperative system: sensor nodes, jammers, loitering munitions and strike nodes that share local processing and nominate targets for kinetic effectors. In those runs, a distributed mesh of unmanned aerial systems often functions as the initial detection grid. That grid enables surface and subsurface unmanned units to concentrate fires against high value maritime targets or to tie up an invasion fleet’s escorts and logistics at favorable cost exchange ratios. These results explain why unmanned maritime platforms are becoming operational priorities in conceptual modeling.
The policy driver is clear. The Pentagon’s Replicator initiative intends to field attritable autonomous systems at scale to blunt a numerically superior adversary. That program has explicit maritime strands and its first publicly announced buys and budget signals show the department is serious about operationalising massed autonomy rather than leaving it an intellectual exercise. The commercial and industrial actions connected to Replicator also matter because having a plan is not the same thing as having the surge capacity to execute it.
When we move from air to sea the operational calculus shifts in three important ways. First, the ocean is a slower, denser battlespace. Surface drones can loiter longer and present persistent threats to shipping and amphibious formations in ways that short endurance aerial platforms cannot. Second, the signature environment is different. Surface vessels are harder to disguise than subsonic aircraft but are also susceptible to low cost kinetic or explosive effects that have been demonstrated in other theatres. Third, command relations and launch vectors change. USVs and UUVs can be launched from shore, over the horizon from host vessels, or from mother platforms, which creates a richer menu for distributed employment but also a wider attack surface for countermeasures. These maritime characteristics are why defence commentators have singled out USVs and related maritime nodes as crucial components of the broader swarm concept.
Simulations nevertheless expose frictions that are often elided in advocacy pieces. The mesh networks that make distributed autonomy resilient are themselves fragile in contested electromagnetic and cyber environments. RAND modelers and simulation participants explicitly note electronic warfare, jamming and deception as key threats that reduce the effectiveness of an autonomous grid unless resilience is engineered from the start. Likewise, attritability is a double edged sword. Cheap expendable vehicles are politically and operationally attractive, but mass attrition still creates logistical and manufacture demands that must be met in wartime. The policy consequence is obvious: simulation success depends as much on industrial capacity and spectrum robustness as it does on AI algorithms.
China and other actors are not passive bystanders in these models. Public reporting and commentary from Chinese outlets and analysts underscore that Beijing is investing in counter-swarm tools, including electronic warfare, directed energy prototypes and massed unmanned formations of its own. Simulations therefore become a contest of not only numbers but of resilience, deception, and how quickly each side can reconstitute lost edge nodes in a degraded environment. The strategic implication is that maritime drone deployments will not simply invert the cross-Strait balance overnight. They will change the geometry of escalation and attrition.
There is also an ethical and legal dimension that simulations cannot resolve for us but that they do force onto the table. Maritime drones employed against surface vessels cross clear thresholds in distinction, proportionality and attribution. Autonomous target nomination, at scale, complicates accountability. If a distributed mesh nominates a target and a loitering munition executes the strike with only local autonomy intervening, who bears legal responsibility for a mistaken hit on a neutral or commercial vessel? Simulations can highlight probability distributions and expected outcomes but they cannot substitute for doctrine, rules of engagement and legal frameworks that assign decision authority. This is a governance gap that strategy must confront before operational deployment becomes routine.
What should policymakers and technologists take from the simulations, in pragmatic terms? First, invest in layered resiliency. Mesh networking, edge processing and frequency diversity are not optional. Second, exercise the industrial base as aggressively as the tactical concepts. Replicator and similar efforts are steps in the right direction because they couple procurement with experimentation, but simulations show that procurement velocity must match assumed attrition rates. Third, prosecute doctrine and legal policy in parallel with tech. The law of armed conflict and alliance burden sharing will both be tested the day any massed autonomous maritime campaign is launched. Finally, do not fetishise unmanned systems as magic. Simulations make their value plain. They also make plain the costs, the vulnerabilities and the political choices.
In the end the simulations are a useful corrective to both optimism and fatalism. They show that autonomous maritime systems can be a decisive factor in the defence of Taiwan only if those systems are embedded inside resilient networks, supported by industrial surge capacity, and governed by clear human-in-the-loop doctrines. The moral responsibility for those choices rests with strategists and statesmen, not with the algorithms that will one day be their instruments.