Defence drone makers quietly become the largest customers for commercial AI

The aggregate compute bill for a serious defence drone programme is large. The combined compute bill for the half-dozen serious programmes underway is very large indeed.

What this changes for the industry

A few things, in subtle but real ways.

The first is the customer composition of advanced AI infrastructure. Defence customers have different procurement cycles, different security requirements, and different feedback loops than commercial AI labs. Vendors increasingly maintain dual product lines or accept different terms for defence buyers.

The second is the talent flow. AI engineers and roboticists working on defence drone programmes are increasingly well-paid, often at parity with commercial AI lab compensation. The pool of available engineering talent shifts accordingly.

The third is the dual-use research question. Models and methods developed for defence drones often have commercial applications, and vice versa. Where exactly the boundary should sit, both technically and ethically, is a conversation the field is starting to have more explicitly.

What is likely next

Expect more disclosure of defence procurement of commercial AI infrastructure. Expect at least one major commercial AI lab to formally adopt or reject defence-customer policies in the next year. Expect the export-control conversation around AI compute to focus increasingly on dual-use drone applications.

The compute is moving toward defence customers because the work being done there is large, well-funded, and operationally important. The implications are bigger than drones.

Rashid Amin · 3 May · 3 min