Defense: the vertical where the rules invert

01Why defense breaks the rest of the thesis, on purpose

Everywhere else in this sector, Orien underwrites unit economics: payback periods, net revenue retention, the cost curve. Defense is the deliberate exception, and holding it up against the master framework is the fastest way to find out whether that framework is real or just a story we like. The buyer here does not optimise cost per task. It optimises capability, mass, and the certainty of a domestic, trusted supply. That single change in the buyer's utility function rewrites almost every verdict, and, usefully, it rewrites them in ways the framework predicts.

Three forces turned defense into a venture category

Ukraine proved the doctrine. A war fought with $300–500 first-person-view drones produced by the million has demonstrated that autonomous mass beats exquisite scarcity, that the side which can field, coordinate and attrit cheap smart systems at scale wins engagements against far costlier hardware. Ukraine went from roughly ten drone makers in 2022 to over five hundred, and now builds ten million drones a year. The lesson has been absorbed in every Western capital.

The budget turned. For the first time, the US FY2026 request broke out autonomy as a dedicated ~$13.4B line, and the FY2027 request escalates it dramatically, roughly $54B for autonomous and remotely-operated systems, a ~$39B "Drone Dominance" tranche, $3.1B for counter-UAS. (We flag that the FY2027 topline mixes discretionary with mandatory and reconciliation money, so the growth rates quoted in the press are not clean; the direction, however, is unambiguous.)

Procurement began to reform. The historic killer of defense startups, the "valley of death" between a prototype and a program of record, is being addressed by the SPEED and FoRGED Acts and a large expansion of Commercial Solutions Openings, including a "commercial-first" mandate requiring the department to prove commercial options cannot meet a need before building bespoke. This is the plumbing that lets venture-scale companies reach government-scale revenue.

02The framework, stress-tested

Here is the interesting result. Defense does not break the master framework, it runs it forward at double speed, and exposes a subtlety the commercial verticals hide.

Recombination is already here, the body is designed to be disposable

In the master thesis, the open question was when the intelligence–hardware interface standardises and the stack unbundles. In defense, doctrine has answered it. Attritable mass means the airframe is deliberately cheap, commoditised and expendable, while the durable value sits in the autonomy that flies it and the command-and-control software that coordinates the swarm. The body is not a moat; it is ammunition. Shield AI's Hivemind runs across 26+ vehicle classes; Auterion positions AuterionOS as "the Android for drones" and is shipping tens of thousands of software strike-kits; Anduril's Lattice is sold as the operating system, not the missile. Cross-platform transfer, the master's swing variable, is not a future hope here; it is the business model. Defense is the vertical where World B has already won at the layer where value accrues.

Yet the biggest outcomes are still full-stack, for a different reason than commercial

And here is the twist that makes defense worth studying. If the software layer captures the value, why is the $61B company (Anduril) a full-stack prime that builds both the drones and Lattice? In the commercial world, integration wins because the interface is immature. In defense, integration wins for a different reason: the buyer rewards it. The Department of War would rather hand a trusted prime a ten-year, $20B-ceiling enterprise agreement than stitch together a best-of-breed stack itself, because the prime absorbs the integration risk, carries the security and ITAR burden, and can be held accountable for the whole system. Integration here is not a technical necessity; it is a procurement preference. That distinction matters, because procurement preferences can change faster than physics, which is exactly why the pure software layer (Palantir, AuterionOS, NODA) is the other place value pools, and why the two coexist rather than one killing the other.

The security overlay, at its absolute maximum

If the master thesis argued that provenance gets priced in the West, defense is where provenance is mandated. The American Security Drone Act already bars Chinese drones from federal use; the American Security Robotics Act (March 2026) extends the same logic to ground robots; Blue UAS certification and ITAR draw a hard border around trusted suppliers. This walls off DJI, some 70%+ of the commercial drone market, and manufactures captive demand for domestic autonomy. The overlay that is a friction elsewhere is, here, the entire market structure: a Chinese bill of materials is not a cost disadvantage, it is a disqualification. Defense is the proof case for the master's claim that the West runs a separate, trust-gated supply chain.

03The cost-asymmetry engine

The whole vertical runs on one number: the ratio between the cost of an effect and the cost of the thing that used to deliver or defend against it. When a $500 drone can do what a $170,000 loitering munition did, and a $35,000 drone can force a $3,000,000 interceptor to be spent, the economics of warfare re-price, and so does the question of where a fund should own the value.

For an investor the implication is precise. If the airframe cost is racing toward zero, then owning airframe manufacturing is owning a commodity, margins compress toward contract-manufacturing economics. The value migrates to the two things that do not commoditise: the autonomy and coordination software that turns a cheap body into a lethal, swarming, jam-resistant system, and the scarce hard enablers that mass production still bottlenecks on, solid rocket motors, seekers, effectors, trusted compute. This is the master's stack logic, sharpened by a buyer who actively wants the body to be cheap.

04Where the value is accruing

The market is pricing this in real time, and the pattern confirms the framework: capital concentrates in the software-and-systems layer and in the new full-stack primes, at multiples that look like enterprise software rather than defense hardware, Palantir trades around 65× forward revenue, roughly ten times the software-cohort median.

Underneath the platform names sits the layer we find most interesting for Orien's edge: the autonomy-software and orchestration companies selling the coordination brain rather than the body. Auterion (AuterionOS, $130M Series B led by Bessemer, 33,000 strike-kits to Ukraine), NODA AI (cross-vendor swarm orchestration, $25M Series A), and the C2 frontier generally are where the attritable-mass doctrine creates a durable, recurring, software-margin business, precisely because the platforms beneath them are becoming interchangeable.

05Orien's verdict

Defense signposts we track

• Does the software layer hold its premium? If Lattice / AuterionOS / Palantir keep winning cross-platform mandates at software multiples, the "value is in the OS" call holds. If primes re-bundle and squeeze the independent software layer, revisit.
• Do the new primes reach durable government-scale revenue, or does the valley of death reassert once reform enthusiasm fades? Watch program-of-record conversions, not prototype awards.
• The exit window. First defense-tech IPOs are printing (Karman, Voyager in 2025; Ukraine's Swarmer up 452% on debut). If Anduril or Shield AI file, the "generational new primes" thesis is validated in public markets, the clearest read on whether the private marks are real.
• Budget durability. The FY2027 autonomy surge leans on mandatory and reconciliation money. Watch whether it survives into enacted appropriations, or proves a one-cycle spike.

06What breaks this call

The "Iron Bubble." PitchBook's own framing is the honest bear case: median defense-tech valuations up ~240% year-on-year, repeated 2–2.4× step-ups on undisclosed revenue, capital arriving faster than programs convert. Much of the cohort is priced for a procurement transformation that has been legislated but not yet delivered. Entry discipline is the whole game here.

Procurement and political risk. Reform can stall; a budget cycle can turn; a program can be cancelled between prototype and production. Defense revenue is lumpy, contract-timed, and politically exposed in a way SaaS is not, the master thesis's warning about hardware capital intensity applies with the added variable of an electorate.

ITAR cuts both ways. Export controls are a moat against foreign competition and a tax on domestic startups, and increasingly a strategic liability: European buyers now market "ITAR-free" as a feature, and the largest European rounds (Helsing at $18B) are partly a bet on a sovereign, US-independent defense-tech stack. A US-centric portfolio inherits that boundary.

Concentration. A handful of names, Anduril, Palantir, are capturing a disproportionate share of both value and program access, and late entry into the obvious winners is expensive. The edge is in the software and enabler layers one level down, before the market has fully priced them.

07Sources