China's own model says its drone output tops out near 1.9x, not 3x

Run the percentage back through the demand shock and the study reads less like a shortfall and more like a ceiling. Filling 63.4% of an order book three times its peacetime size means absolute output of roughly 1.9 baseline years' worth ^[17]. Hold the no-mobilisation case next to it: a median 34% of tripled demand works out to about 1.02 times peacetime output ^[18]. A supply chain that does nothing in particular produces about what it already produced.

So everything the model actually mobilises (released stored equipment, overtime, added lines, converted civilian plants ^[2]) buys something on the order of seven-tenths of one baseline year of extra drones, comparing the AI-led median with the do-nothing median ^[19]. The unfilled remainder in the headline scenario, 36.6% of a tripled book, is itself about 1.1 baseline years of demand that never gets met ^[21].

The money curve is blunter still. Raising available mobilisation funding from 60% to 140% of baseline is a 2.3-fold increase in cash, and it moves fulfilment 7.4 points, from 54.9% to 62.3% ^[8][20]. Against tripled demand, that is roughly 0.22 of a baseline year of additional output ^[22]. The researchers say why the curve flattens: factory construction times and conversion limits govern how fast new capacity becomes usable ^[9], and expansion and conversion arrive in phases rather than continuously ^[15]. Cash is not the scarce input in their model. Calendar time is.

Which is the same finding stated a second way when the team concludes that wartime output depends not only on assembling drones but on how quickly the supply of critical components can expand ^[16]. Their framing puts it directly: competition is "no longer merely about the performance of weapon systems, but more fundamentally about industrial capacity" ^[4]. That is a claim against the way most drone comparisons are written, and it comes from a group whose lead author directs a national defence mobilisation research centre and whose model explicitly links component and subsystem suppliers to assembly plants ^[3].

The dissent in the reporting is worth reading closely, because it does not contest the mechanism. A Chinese weapons specialist called the work an "argumentative study" rather than a forecast ^[10], and noted that "rapid production capability can fully meet the demand, but the supply of core components is not that simple" ^[12]. That is agreement on where the bottleneck sits, and disagreement about the roster. The simulation covered existing military suppliers plus civilian manufacturers capable of conversion ^[11], so 63.4% is a property of that roster's depth, not of Chinese manufacturing at large. Widen it to electronics, battery, vehicle, machinery and consumer technology firms and the number moves ^[11]. The Covid precedent cited is real, with non-medical companies turning quickly to masks and protective equipment ^[13], but drones need technical adaptation, suitable machinery and specialised parts before a converted line produces anything ^[14].

One small inconsistency in the reported figures is worth flagging: the AI strategy's advantage is given as 8.4% on averages ^[6], while the quoted medians for rule-based and AI mobilisation sit about seven points apart ^[23]. Either way, the allocation policy is the cheapest lever in the study, because it only re-sequences which suppliers expand and which civilian firms convert, using inventory, capacity, order pressure and funding as inputs ^[7]. It buys more than doubling the budget does ^[20][6].