The 1-to-100 Trap
Why defense can't scale innovation.
Stuart Lodge is the CEO of Lodge Systems, a Canadian company building a domestic, software-defined manufacturing stack for flight-ready composite airframes and UAV fuselages.
The Pentagon is built around the Program of Record. Identify one problem, select one solution, lock the design, then scale it for years. That approach exists for good reasons. It creates accountability, supports training and sustainment, and keeps complex programs safe enough to operate.
Autonomy behaves less like a platform and more like a living ecosystem. The advantage comes from speed of adaptation, not from perfecting one design in a conference room and hoping it survives the next countermeasure. Likewise, in software we do not appoint a single point of truth and ban everyone else. We let thousands of teams build, ship, fail, learn, and compete. The ecosystem becomes a meritocracy because reality selects what works.
Defense hardware often does the opposite. We select early, freeze early, and optimize against a stable set of metrics that can look convincing while drifting away from the messy reality of mud, cold, jamming, maintenance, and exhausted operators. The result is a monoculture: one architecture, one supply chain, one brittle bet.
That may be changing as the United States and its allies race to build arsenals of small drones and throw out their old acquisition playbook.
In the United States, the Drone Dominance Program is targeting more than 300,000 small drones over the next several years, with military planners estimating each Corps will need 1,500 to 2,000 per day during active operations. In Canada, the Army’s MINERVA Initiative is building a framework to integrate small uncrewed systems across land, air, and maritime roles, with an explicit focus on iterative development alongside domestic industry.
The demand signal is real on both sides of the border. What neither country has is a manufacturing base that can move hardware from first prototype to fieldable production without forcing teams to start over with new materials, new tooling, and new suppliers. The gap exists because the West’s two default approaches, additive manufacturing and traditional aerospace composites, each solve half the problem. One iterates but cannot scale. The other scales but cannot iterate.
Technologists often speak about going from zero to one. The urgent challenge today is figuring out how to go from prototype to production. Solving this problem means building a manufacturing stack where the same materials and processes carry a design from unit one to unit ten thousand, so that every flight hour at prototype scale still counts at production scale.
The 1-to-100 Trap
The West does not need a million of one drone. We need the capacity to let a hundred teams build a hundred variants at the same time, then prove which ones survive real use. But we are failing at the most important phase of hardware learning: the jump from Unit 1 to Unit 100.
This is the hardware valley of death. If you cannot build production-representative batches, you compensate with models, lab demos, and a handful of prototypes. Then the first honest stress test arrives in operational use and the learning shows up late, when it is most expensive to change.
U.S. operational test guidance is blunt: decisions tied to full-rate production are supposed to be informed by testing with production systems or production-representative test articles. If you cannot afford to build enough representative units to test, you are not doing the kind of validation the system expects. You are guessing, and hoping the guess holds up under fire.
Why the Obvious Manufacturing Answers Miss the Mark
Drone Dominance and MINERVA set the demand signal. Translating it into hardware means hitting hard manufacturing constraints. Per-unit costs for attritable systems need to land around $5,000. The Army’s SkyFoundry initiative is planning facilities that can produce 10,000 systems per month. At those price points and production rates, the airframe cannot depend on materials or processes that were designed for platforms built to last thirty years. It needs to be cheap, fast, structurally sound, and producible on a line that can change geometry between production runs. Neither of the two manufacturing approaches the defense world defaults to can deliver.
Additive manufacturing is a false summit. It works well for getting to the first prototype, and carbon-fiber-reinforced filaments have pushed 3D-printed parts closer to structural relevance in recent years. But the gap between a printed demo and a fieldable airframe is wide, and it gets wider under exactly the conditions that matter: vibration, moisture, temperature cycling, sustained operational stress. A printed part is built layer by layer, and each layer boundary is a potential failure plane. Moisture absorption in printed composites runs significantly higher than in compression-molded equivalents because the microstructure is more porous. Dimensional consistency drifts from part to part, with shrinkage rates exceeding two percent on larger components. The physics of the process makes this unavoidable.
The rate problem is just as severe. Even the most ambitious deployable additive systems—containerized print factories designed to operate near the point of need—top out at roughly fifty Group 2 airframes per month. That is a meaningful prototyping capability. Against wartime consumption rates measured in thousands of drones per day, it is a rounding error. You cannot 3D-print your way to a wartime production base.
The traditional aerospace approach fails in the opposite direction. The composites infrastructure the West has spent decades building was designed for platforms that fly for thirty years: F-35s, 787s, Global Hawks. That entire stack is optimized for durability and low volume. Invar steel molds can take twelve months to develop. Autoclave cure cycles run hours per part. The supply chains depend on a handful of specialized firms in Japan, the United States, and increasingly China. A CSIS analysis published in late 2025 found that carbon fiber production cannot be surged, and that any disruption would ripple across every composite-dependent program in the defense industrial base. Aerospace-grade carbon fiber is the right answer for a fighter jet that needs to last decades. For an attritable drone that costs five thousand dollars and may not come back from its first flight, it is the wrong material, at the wrong price, from the wrong supply chain.
So the defense innovation ecosystem is stuck between two bad options: a prototyping method that cannot scale, and a production method that cannot iterate.
One company we work with is caught in exactly this gap. They are 3D printing medium-sized quadcopter airframes. It works fine for development, but they hit a wall after about five units. Each part needs extensive support structures and post-processing. The per-unit cost stays stubbornly high. The parts are not production-representative. Their only conventional path forward is to re-engineer the entire system for injection molding, spend fifty thousand dollars or more on permanent steel tooling, and lock a geometry before they have had enough flight hours to know if it is right.
Every team trying to enter the defence UAS space hits the same wall. It is happening right now, across dozens of startups, in the path of every program from Minerva to Drone Dominance to Canada’s IDEaS challenges. The West keeps producing impressive demos that stall before they reach operational relevance. The designs are often good. The manufacturing options force a choice between learning and scaling, and the current industrial base does not let you do both at the same time.
Mass Iteration
Real hardening requires feedback at a volume where failure shows up with enough frequency to act upon. One prototype tells you whether the concept works. Five prototypes tell you whether the build is repeatable. Fifty units, in the hands of real operators, in bad weather, under jamming, is where you actually start learning how a system fails. Below that threshold, you are still guessing.
This is a different manufacturing requirement than rapid prototyping, and it is a different requirement than mass production. Rapid prototyping optimizes for speed to first article. Mass production optimizes for cost at volume. Neither one is designed for the thing that actually matters in a fast-moving conflict: the ability to build, test, break, learn, and rebuild in operationally relevant batches, on a cycle measured in weeks rather than years.
Call it mass iteration. The side that can run this loop fastest will have a compounding advantage. Every batch of fifty reveals something the previous batch could not. Multiply that across dozens of teams running parallel experiments and you get an innovation rate that no single program of record can match, no matter how well funded.
We already know what happens when one side has this capacity and the other does not. Ukraine’s Ministry of Defence reported delivering over 1.2 million drones of various types in the first eleven months of 2024. The designs are evolving month to month. The teams that cannot iterate fast enough get replaced by teams that can. When consumption is measured in mass, the advantage belongs to whoever replenishes and improves fastest.
The same logic applies outside of conflict. Wildfire response, infrastructure inspection, offshore energy, logistics. Any domain where you need rugged, mission-specific airframes at moderate volumes benefits from a manufacturing approach built around iteration speed.
The Real Requirement: No Second Switch
Most hardware teams experience not one valley of death, but two. They prototype in whatever process is fast and available, usually 3D printing or hand layup. Then, if they get traction, they are forced into a completely different manufacturing process to reach production scale. New materials, new tooling, new suppliers, new qualification, new failure modes. That second switch is where startups go to die.
The problem goes deeper than cost or schedule. Every material system behaves differently under load, under heat, under moisture. A part validated through hundreds of flight hours in printed nylon tells you almost nothing about how the same geometry will perform when injection-molded in glass-filled polypropylene. Beyond delaying production, the second switch resets your engineering knowledge to near zero. All the test data, all the failure analysis, all the hard-won confidence in your design goes out the window the moment you change the material.
We see this pattern playing out across the industry right now. Teams printing airframes that work well enough to demonstrate capability, then stalling because their only path to volume requires them to start over with a different process. Some spend months re-qualifying. Others burn through their runway trying to bridge the gap. The ones that make it to production typically do so by raising enough capital to brute-force the transition. The ones that do not have the capital simply stay small, no matter how good their design is.
The goal should be a single manufacturing stack that carries you from Unit 1 to Unit 100 to Unit 10,000 without changing the material system or the process logic. Same material. Same forming method. Same failure modes. Every hour of flight testing at prototype scale becomes data that is still valid at production scale. No second switch. No reset.
I started my company to build this stack for composite UAV airframes. The core idea is to treat tooling the way software treats code: versioned, disposable, and fast to produce.
Our software models the structural loads an airframe will encounter to define the most efficient reinforcement patterns, producing composite parts with the strength of metals at the weight of plastics. From there, we produce tooling in hours rather than the weeks or months that traditional steel dies require. Our material properties are compatible with additively manufactured tooling modelled on high-throughput automotive stamping, letting us calibrate tooling fidelity to the run size rather than committing to permanent steel dies upfront. If the geometry needs to change, we produce new tooling. There is no six-figure commitment locking us to a shape we are not yet confident in.
The structural material is a thermoplastic biofibre composite sourced from Canadian forestry by-products. The entire supply chain is domestic and sovereign. The result is a drop-in airframe: flight-ready, structurally validated, and made from the same material system at unit five as at unit five thousand.
A team using this process does not re-engineer when they move from prototype to production. The material stays the same. The process stays the same. The failure modes stay the same. The only things that change are tooling fidelity and production rate. And because the tooling is transient, the cost of changing your mind is low enough that you can actually afford to learn from the field before you commit.
What the Ecosystem Needs
The defense industrial base lacks the capacity to build fieldable, production-representative hardware in batches of 50 to 500, learn from real operational use, and then scale the survivors without switching materials, processes, or suppliers.
This is the structural gap that programs like Minerva, Drone Dominance, and IDEaS are running into right now. They can fund the design work. They can fund the procurement. What they cannot do is force a hundred drone startups through a manufacturing transition that the industrial base does not yet support.
Building that missing middle is what I’m focused on. But the need is bigger than one company or one material system. The West needs manufacturing infrastructure that treats iteration as a first-class requirement, treated with the same seriousness as cost and schedule. The teams that can iterate in hardware the way software teams iterate in code will be the ones that produce systems capable of surviving real conflict. The ones that cannot will keep producing demos.
We should stop trying to pick a single winner in advance. We should instead build the manufacturing infrastructure that lets many teams compete, fail fast, and scale the survivors. That requires domestic materials, transient tooling, and production systems designed for learning.
The 1-to-100 gap is a mass iteration problem. And it is solvable.
 | A guest post by
|