Those capabilities require a distributed, software-defined factory network, built to a government-owned standard, that decouples design from the factory equipment that manufactures and assembles its physical structures. Five actions must be taken to build that network: finishing qualification reform so certification travels with the design file; consolidating the structures catalog; unifying component strategy; scoring industrial capabilities in solicitations; and bringing allies inside the network.
As the founder of a company that operates such factories today, I have a direct commercial interest in the architecture described here. That overlap is sharpest in the sections urging the government to own a standard, score industrial capacity and surge convertibility, and fund the network. My argument, however, is that the network needs to be far more expansive, and that the government should own the standard and ensure no company is the sole provider.
Sign Up for Our Newsletter
Factories of the Future
As we’ve seen in Ukraine and Iran, in a crisis against an adversary with deep-strike capabilities, massive munitions plants and single-source suppliers represent a focused target list. The future will demand networked factories that can immediately shift production priorities when assets are destroyed, and whose entire digital product catalogs are available on demand.
The weapons the United States needs on day 100 of a conflict may not be the weapons it needed on day one. Wartime adaptation adds momentum to the flywheel and scale benefits of a network rather than slowing them. Battlefield-generated data signals demand and iteration requirements to the network in real time, allowing for faster feedback loops, even if nodes (factory locations) are struck. Network scaling generates multiple benefits: Each new node raises the value of every design in the catalog because it can be built by any node, and each new catalog entry raises the value of every node because every node can build it, alone or in combination with other qualified nodes.
This can be done. Currently, for example, the same qualified part, additively-manufactured fuel nozzle tips, are already being printed across dozens of machines in a single GE plant, and the technology is being used for volume production. The challenge is scaling it into a network without replicating the analog factory architecture of the past.
For one, this will necessitate design-agnostic factories. Whether it’s automaking or munitions production, when capital equipment is hard-tooled to one design, the entire investment is a bet on demand for that one design. If that bet fails, invested capital is destroyed because it is non-fungible. The defense industry is even more complicated than other manufacturing sectors because it is limited by monopsony demand: Program-cancellation risk makes primes rationally under-invest in capacity, leaving the government to fund facilities. (So, design-dedicated missile lines are sized to peacetime minimum sustaining rates, surge requires new facilities, and design iteration stops the line.)
A large node of design-agnostic machines avoids this problem as it is not a bet on one design. The same capital equipment produces any product in the catalog. This network model is directly analogous to the network built around the semiconductor industry, albeit American-based and focused on mechanical structures.
Additionally, design-agnostic equipment does not require time-consuming changes of tooling for weapons iteration. While Russia, for example, has demonstrated weapons iteration cycles of days or weeks, driven by battlefield feedback, these are largely component swaps on an unchanged airframe, and the iteration cycle extends significantly when hard tooling changes are required (such as the swap of warheads or propulsion systems).
Finally, using digital manufacturing tools produces more optimized structures and supply chains. General Atomics Aeronautical Systems, using such a system, has achieved over 95 percent part consolidation on a small unmanned aircraft system aerostructure and tool-less robotic assembly in under 20 minutes, a capability it expects will enable near-theater ramp capacity. GE Aerospace consolidated 855 conventionally manufactured parts into 12 additive components in its Federal Aviation Administration-certified Catalyst turboprop. The supply chain collapses toward feedstock and a small subset of high-value components: energetics, seekers, and microelectronics.
Building the Network
Such networked factories will produce critical components for munitions and related systems. With the exception of energetics and electronics, all of these components are either in digital production or going into it: integrated airframe and related structures (fuselage, strongback, fuel tanks, avionics tray, empennage and control surfaces), turbojets (unitized combustor, shaft, turbine wheel and inlet housing), rocket motor (casings and nozzles), warhead casings (ex-energetics), electronic attack and electronic warfare cradles and related launch canisters.
The core of this network idea is that once a production process is certified, design files for mechanical structures can be sent to qualified factories across the United States and allied nations. These factories will operate functionally identical versioned machines and execute identical code to produce identical parts. That requires tight control of calibration, material lots, build parameters, post-processing, and in-process inspection, which has already been done for qualified jet engine and rocket engine parts. Design improvements are treated like software updates, as once changed, every factory can produce new versions immediately. The set of mechanical structures defines the boundary of the qualification standard. Under the specification stack, an organic-industrial-base depot, a qualified print shop, a machining shop, an assembly-only node, and an end-to-end platform are all the same thing to the network — conforming nodes of different scopes.
Under this model, companies offering a digital production system focus their efforts on the design, manufacturing, and assembly of missile system mechanical structures, while the prime is freed up to focus on its highest-value activities (analogous to Apple in the microelectronics sector). The prime also completes final assembly of the round, and the government inspects and accepts.
For a defense industrial base architecture to deliver the necessary regeneration rate and adaptive capability that survives attack, it must be founded on six capabilities: process-level certification; decoupling of factory equipment from design; functionally identical modular nodes; software-speed propagation and adaptability; fungibility of capacity; and dual-use by construction, so peacetime commercial volume funds wartime surge capacity. The government would have to compensate companies for displacing commercial production and pay for production of defense articles, but the government would not need to have invested in additional idle capacity, as traditional manufacturing would require, for surge. It also must be built to translatable and adaptable standards and processes.
Owning the Standard
To build these network capabilities into the defense industrial base, the government must own a standard, not pick a winner. By standard, I mean the formalization of the six capabilities above into a framework of process qualification (certifying once with certification traveling wherever nodes are deployed), interface conformance (node-agnostic catalog entries), and a requisite data and security baseline (the digital thread specified and defended like a weapons system).
The standard centers on the data package (the design file in its full, versioned form — i.e., the complete build recipe) and machine and process qualification. Standardization of the data package requires that it be versioned as one object, including model, material specification, build parameters, assembly instructions, and inspection and acceptance criteria for the executable article. Qualification certifies the machine-plus-process once, but the certification travels with the data package. Process qualification decouples chip design from physical fabrication. In the semiconductor industry, the process design kit acts as a highly formalized, digital contract that abstracts the physical complexities of a fabrication plant into a standardized data package.
In this network, qualification states what a node can produce. The government must accept this part, from this node, today without re-qualifying. Enforcing this standard separates design authority from production conformity, so contracting officers aren’t paralyzed by liability concerns. It defines the digital birth certificate: machine, process, and inspection provenance — replacing the resident inspector at a thousand distributed nodes. Nadcap already conducts one master technical audit per facility for aerospace suppliers, with all primes accepting the result.
Nodes also need conformity classes — definition of types of factories — that would include commercial-only, dual-use, and defense-cleared, and would be subject to different qualification requirements. Data-package classification levels would be mapped to node tiers, so controlled packages route only to cleared nodes while unclassified structural packages, which may still be controlled unclassified information subject to export controls, route with fewer restrictions depending on the tier. Structuring arsenals and depots as network nodes would provide the modernization pathway for government-owned bases, strengthening the organic industrial base rather than threatening it. There is also an operational layer of the dual-use network — tasking, priority, and surge activation — that largely exists. The Defense Priorities and Allocations System would determine who gets capacity first.
A fair objection is that by shifting to a standard approach, the digital thread and the data package every node runs can become vulnerable. But a data package can be escrowed, replicated, and patched network-wide in days, while it takes years to rebuild a destroyed plant. Each part travels with its birth certificate, so a corrupted build would surface at the part itself. This asymmetry exists only if the digital thread is defended like a weapons system — as the semiconductor design layer already is. A flawed or tampered design file is a different risk, because parts would match the bad file: Guarding against it requires cryptographically signed design files and independent verification. The qualification authority itself must be guarded against capture or paralysis.
Building the Factories
This is a specification plus qualification regime, and an integration effort. Every element has working precedent in the commercial world, and the Pentagon and Congress have already taken initial steps. The problem is no longer to persuade the government to begin, but to finish the job.
First, the government must finish qualification reform. The constraint is the acceptance regime, built for fixed plants, which qualifies a specific part from a specific machine in a specific building. The Fiscal Year 2025 National Defense Authorization Act commits the Department of Defense to expedited acceptance and qualification of advanced manufacturing processes. However, practice needs to catch up to mandate. In March 2026, Portsmouth Naval Shipyard installed one of the first welded, additively manufactured metal parts aboard a nuclear submarine. It is a single flange, qualified as one part, at one yard, for one boat, under a part-by-part model. Lack of speed in implementation of additive manufacturing beyond this single part has led to criticism of the part-by-part qualification process by Navy officials and issuance of a new process-oriented regulation in August 2026. However, no department-wide instruction yet certifies the production system once and enables that certification to travel with the design file.
Second, the Pentagon must consolidate the service branches’ structures catalogs. The same authorization act created a statutory category of critical readiness items and directed digital production paths for them. What does not exist is the comprehensive institutional authority to decide what enters the catalog as priorities — the munitions that empty first, unmanned systems needed at volume, and the few hundred out-of-production parts driving the readiness crisis.
Third, the Defense Department must unify the component strategy. With a digitized network, munitions bottlenecks are reduced to energetics, seekers, and microelectronics. Because structures are built to match the supply of energetics and electronics, no excess structures inventory builds up, and upstream suppliers get certainty about demand. Two programs are underway: a $435 million contract is restoring domestic TNT production in Kentucky, and around $120 million in wartime production authority investments support adding a third domestic supplier of solid rocket motors. NATO members also launched jointly funded studies to qualify European suppliers for priority missile components. But these efforts operate under separate authorities with no common deadline. There is not a dedicated national effort for seekers and microelectronics, components that most directly gate mass production. The lessons of Operation Warp Speed apply here: one concrete objective, one accountable manager, one deadline, allies inside the program from day one, and a willingness to buy outcomes rather than administer processes.
Fourth, the Pentagon needs to directly score the elements of deterrence. Procurement buys units, but deterrence encompasses magazine depth, regeneration rate, surge convertibility, catalog production fungibility, and asset dispersal and survival. The Pentagon’s November 2025 acquisition strategy prioritizes wartime surge capacity and provides that affordable-mass awards be deliberately split among multiple vendors. These actions are directionally correct but stop one step short. Surge convertibility and reconstitution rate are rarely, if ever, weighted criteria on which proposals are scored and contracts won. Until solicitations score and pay for them, no rational contractor will build a business on supplying them. For example, the ability to supply energetics loadable into any family of additively manufactured warhead casings would be weighted for its production capacity across that family and its surge capacity.
Fifth, while U.S. allies already participate in component supply and co-production, the next step is to make them full network partners. The NATO Engine is an alliance-wide initiative to expand defense production capacity by connecting defense and civilian manufacturing lines across the alliance. This is a necessary bridge, not tenth-salvo deterrence. The United States needs a mutual qualification-acceptance agreement among allies that mirrors its domestic standard and includes them in its digital network.
While America’s core technology edge is its hyperscaling bits network, it now has the mature technology to create an analogous network for advanced manufacturing. That standard is something no adversary can destroy, and no one company should control. It requires Congress to define a clear scope of responsibility and liability within the existing Pentagon organization and provide funding for staffing. Deterrence will rest on whether America’s magazines regenerate faster than they can be emptied.
Write for Cogs of War
Kevin Czinger is the founder and executive chairman of Divergent Technologies, an American company that provides advanced engineering and manufacturing capacity for defense primes and automakers through an integrated platform combining AI-generative design, metal additive manufacturing, and automated robotic assembly. He served earlier as an enlisted infantry rifleman in the Marine Corps Reserves. The views expressed here are his own and are not intended to represent the views of any companies with which he is affiliated.
***Please note, as a matter of house style, War on the Rocks will not use a different name for the U.S. Department of Defense until and unless the name is changed by statute by the U.S. Congress.
Image: Midjourney