Rewiring the A&D Supply Chain for Scale

Aerospace and defense has no shortage of demand. The five largest U.S. primes, Boeing, RTX, Lockheed Martin, General Dynamics, and Northrop Grumman, closed fiscal 2025 with $1,357.5 billion of combined backlog, up 23.7% in a single year, against $347.6 billion of revenue (Santiago & Company analysis of FY2018–FY2025 Form 10-K filings; PwC’s mid-2026 deal outlook cites the same $1.36 trillion figure). The commercial queue has passed 17,000 jets, about 60% of the active fleet against a historical norm of 30–40%, and IATA puts cumulative deliveries since 2019 at least 5,300 aircraft short of pre-pandemic trend, with normalization unlikely before 2031.

Output has not kept pace, and the constraint sits below the aggregate, at particular production nodes: a specific site, process, workforce, and part approved to deliver at the required rate and yield. These nodes can take years to create, are difficult to move, and often sit several tiers below the prime. That changes the management agenda: leaders need to map production nodes rather than supplier names, secure control without assuming ownership is always necessary, and raise yield inside qualified operations before adding nominal capacity.

Demand visibility has outgrown demonstrated throughput.

Eight years of SEC filings put a number on the wedge. The five primes’ revenue-to-opening-backlog ratio, a proxy for backlog burn, was 0.322 in 2021, slipped to 0.291 by 2024, and recovered to 0.317 in 2025. Measured the other way, revenue-equivalent backlog duration rose from 3.1 revenue years to 3.9 revenue years over the same period. Both labels matter: duration is an inventory measure rather than a delivery schedule, and the ratio is a proxy, since some revenue converts from orders booked within the year and backlog definitions differ across companies. The companies’ own disclosures point in the same direction: RTX expected about 30% of its backlog to convert to revenue within 12 months in 2021, about 25% by 2023, and again in 2025.

Order books are outrunning conversion: the five primes' combined backlog now equals 3.9 years of revenue, up from 3.1 in 2021

The aggregate conceals variation worth reading. A pure-play defense cut (Lockheed Martin, Northrop Grumman, General Dynamics) shows the same direction more mildly, from 0.53 in 2019 to roughly 0.47, and Lockheed Martin’s funded backlog has converted at a broadly stable rate. Aggregate duration has increased, but the pattern varies by company and contract mix; this makes the constraint most visible at the segment and production-node levels.

The commercial lane tells the same story in aircraft, most starkly in widebodies. Boeing and Airbus delivered 380 widebody aircraft in 2018 and 426 in 2019. Deliveries collapsed to 155 in 2021, were still just 172 in 2024, and reached 246 in 2025, a real recovery, up 43% year over year; that still leaves output 35% below the 2018 level. Held against the constant 2018 output, the cumulative gap since 2020 is about 1,100 aircraft; against the 2018–19 average, it is closer to 1,240. That is an output gap versus a pre-pandemic production rate, a counterfactual baseline rather than a count of unfulfilled orders, and it is striking on those terms.

Widebody deliveries have not recovered: 2025 output was still ~35% below 2018, a cumulative output gap of roughly 1,100 aircraft versus the 2018 rate

Slow conversion shows up in prime P&L and airline operating costs. Boeing recorded $5.0 billion of 2024 losses across five fixed-price defense development programs; its T-7A trainer loss was, in the company’s words, “primarily driven by projected increases in supplier cost estimates,” and the 777X program absorbed reach-forward losses of $3.5 billion in 2024 and $4.9 billion in 2025 as delays compounded. Citi’s November 2025 sector review notes backlogs stretching to seven years at some OEMs and cash conversion cycles longer than in almost any other industry. IATA, drawing on Oliver Wyman analysis, put the 2025 cost of supply chain dysfunction to airlines above $11 billion, with lease rates up 20–30% since 2019. This is not simply a story of weak prime-level execution; many of the documented delays originate in structural constraints several tiers below the prime.

The industry’s institutions each name part of the cause. IATA points to engines, certification timelines, skilled labor, and fragile supplier networks; the Pentagon’s industrial-base assessments to consolidation and single-source dependency; trade associations to attrition; banks to capital intensity and working capital. These constraints share a common denominator: output depends on approved capabilities embedded in particular sites, processes, and workforces. We call that shared scarcity qualified throughput, the ability of a specific site, process, workforce, and production line to deliver an approved part at the required rate and yield. That is the constraint this paper examines.

The constraint sits at the qualified production node.

Capital is necessary here, but it does not eliminate qualification and industrialization lead times. The unit of capacity in this industry is smaller than the factory and more specific than the company: the qualified production node, the approved combination of site, process, people, and part. Nominal capacity (buildings, machine tools, generalist labor) can be bought in quarters. A qualified node is built twice over, first through a stack of formal approvals and then through the industrialization that makes an approved process run at rate and yield. The moat is not paperwork; it is the combination of approval and industrialization, and much of both is lost in a move. This is why the industry’s real constraint is the node, not the factory.

Qualified capacity is four layers of approval plus rate readiness: all attached to the node, not the company

The approval stack has four formal layers. AS9100 certification indicates that a site operates an aerospace-grade quality management system. Nadcap accredits each special process (heat treating, chemical processing, welding, nondestructive testing, and composites sit among its 26 families) at that site through recurring industry-managed audits; industry consultants place a typical accreditation effort at 12 to 18 months including preparation. First article inspection under AS9102 ties a specific part to a specific line, and generally must be redone when process, tooling, materials, or location change. Customer and program approval sits above all three, and in U.S. government work it attaches to a specific legal entity: under FAR 42.1204, contracts do not transfer with an asset sale; they move only through a discretionary novation in which the buyer must demonstrate capability to perform. The regulatory layer above the supplier stack has moved in the same direction, with IATA reporting that new-aircraft type-certification timelines have stretched from 12–24 months to four or five years.

Beneath the approvals sits a layer of a different kind: rate readiness. The people who run approved processes at rate are a production-readiness asset rather than a formal approval, and not all of the critical ones hold certifications. This layer is the fastest to lose. Aerospace and defense turnover ran about 15% in 2024 per AIA and McKinsey’s workforce study, more than double the cross-industry average, so a node can lose its rate readiness faster than any approval can be rebuilt. Transfer is gated as well, which is why Boeing carved its reacquisition of Spirit AeroSystems along customer-qualification lines: Airbus work packages to Airbus, Boeing work to Boeing, defense programs held in a separate subsidiary.

An industry short of buildings can build them. One short of qualified nodes must grow them in place over the years and hold them together as they do so.

The documented losses trace to a handful of constrained nodes.

The node constraint binds hardest where few nodes remain, and the counts are stark. The Pentagon’s February 2022 State of Competition report documents the arithmetic of three decades of consolidation: since the 1990s, 51 aerospace and defense primes have become 5; tactical missile suppliers fell from 13 to 3; fixed-wing aircraft suppliers from 8 to 3; satellite suppliers from 8 to 4. Ninety percent of missiles now come from three sources, and domestic solid rocket motor production rests on two suppliers. Beneath the primes, small businesses in the defense industrial base shrank by more than 40% over the preceding decade, with Deputy Secretary Hicks warning that 15,000 more were at risk. The GAO added in 2025 that the department has limited visibility into a base of more than 200,000 suppliers.

Beneath the primes, the supplier base has thinned to single digits at critical nodes

Supplier counts understate the problem when read through the node lens. Three missile suppliers sounds like three interchangeable sources. In practice, it is a fixed set of qualified site-process-part combinations, each with its own approval stack, none of which can absorb a competitor’s work without rebuilding part of that stack.

The thinnest parts of the base are where order books are growing fastest. Lockheed Martin’s missiles and fire control segment backlog grew 45% in two years, from $32.2 billion to $46.7 billion. RTX’s defense backlog rose from $93 billion to $107 billion in 2025 alone. Boeing’s defense backlog, broadly flat for six years, jumped 32% to $84.8 billion in 2025. General Dynamics booked $20.1 billion of submarine awards in a single year. Even antitrust policy now treats remaining competition at such nodes as a strategic asset.

The documented failures each trace to a named node. Finished airframes sat parked awaiting engines through 2025, by IATA’s account. Boeing’s T-7A loss was driven primarily by supplier cost estimates, and the 777X’s combined $8.4 billion in 2024–25 reach-forward losses are attributable to production challenges, certification delays, and supplier costs. Missile demand is compounding fastest where solid rocket motor supply rests on two domestic producers.

The constraint is visible node by node: documented chains from bottleneck to cost

Three qualifications apply. The consolidation the Pentagon describes played out over three decades, much of it driven by the collapse of post-Cold War demand rather than by qualification friction, and the department also reports that the vast majority of its contract actions are competed. The workforce picture is about composition rather than headcount, since industry employment grew 2.9% in 2024. And the chains above are documented cases, not a decomposition of the whole gap; they show the mechanism at work, concentrated at the chokepoints where certifications renew during audit cycles, while experienced people can leave at 15% per year.

Capital can buy control of a node, but it cannot conjure new throughput.

Scarce, slow-to-replicate assets attract capital. Capital transfers ownership of an existing node; it rarely compresses the time required to build a new one. The current M&A wave is largely a redistribution of qualified throughput toward owners who value it most rather than a net addition to it.

The volumes are substantial. PwC counts about $27 billion of U.S. aerospace and defense deal value in 2024 and more than $30 billion in 2025, with the first half of 2026 tracking toward a similar pace, and describes the thesis in language close to our own: “qualified capacity remains scarce,” with vertical integration, private equity-backed roll-ups of Tier 2 and Tier 3 suppliers, and distressed acquisitions of qualified facilities among the most active areas. The Pentagon had flagged private equity roll-ups of subtier suppliers as a defining trend before this cycle began. The primes are moving in parallel. Boeing completed its reacquisition of Spirit AeroSystems in December 2025, bringing some 15,000 employees back in-house in a deal it framed as stabilizing commercial production and the supply chain. L3Harris paid $4.7 billion in 2023 for Aerojet Rocketdyne, which it described as a critical merchant supplier of propulsion systems.

What buyers are paying for is visible in the prices. TransDigm agreed to pay about $960 million for Stellant Systems, a business with about $300 million of revenue (roughly 3.2 times sales), citing its highly engineered, proprietary, aftermarket-heavy positions. One data point is descriptive rather than a market-wide premium, but it illustrates the right underwriting lens: qualification-adjusted replacement cost, meaning the replacement cost of the machines plus the time and risk of rebuilding the approval stack and rate readiness at a new node. Priced that way, a modest machine shop with sole-source qualified positions can rationally command a price that looks irrational on an equipment basis. Deal structure is part of the asset: an entity-preserving stock purchase generally carries government contracts and approvals intact, while an asset carve-out invites novation risk.

Ownership is capped as a strategy in three ways. Regulators police node consolidation: TransDigm abandoned the Stellant acquisition in July 2026 after the Justice Department said it would sue, on the grounds that the deal would leave the government a single source for critical radar components. Integration does not automatically transfer approvals, as the stack makes clear. And ownership demands an operator; a buyer that cannot run the node adds risk rather than removing it. Nor is ownership always the end state: L3Harris, three years after buying Aerojet, filed in April 2026 to take its missile solutions business public. PwC’s outlook observes that the market is penalizing acquirers who carry legacy complexity.

Four control models, two gates

Where ownership is unavailable or unwise, executives need a default control model for every critical node. Two questions position each one: how many qualified alternatives exist, and how long and difficult would it be to qualify another. Where alternatives are many, and qualification is fast, contract: long-term agreements, capacity reservations, take-or-pay where demand supports it. If alternatives are many but qualification is slow, develop and reserve capacity: supplier development engineers on the supplier’s floor, tooling investment, co-funded accreditation. Few alternatives with a qualifiable second source call for diversify: fund second-source qualification as an insurance premium priced off qualification-adjusted replacement cost. And when alternatives are few and qualification is slow, pursue deep partnership or ownership, structured to preserve the qualified entity.

The fewer the qualified alternatives and the harder the requalification, the heavier the control model must be.

Two gates come before any of these choices. The first is consequence of disruption: the revenue and program schedule at risk if the node fails, which determines how much control is worth paying for. The second is confidence in long-term demand, which determines whether take-or-pay commitments, second-source investments, or acquisitions can be underwritten at all. Capex intensity, regulatory constraints, and the buyer’s operating capability then modify the default within each quadrant.

Raise effective output before adding footprint.

Whichever model controls a node, the cheapest new capacity is usually inside it. Raising first-pass yield, inspection speed, schedule reliability, and machine uptime expands effective throughput without requiring qualification of an entirely new source, although material process changes can still require customer review or revalidation. The same logic applies to people: retention at bottleneck nodes is capacity protection, priced against the cost and time required to rebuild rate readiness.

Digital tools, including AI, belong in this part of the agenda rather than at the top. They can multiply output within an existing node, but the current evidence is strongest in inspection, maintenance, and decision support rather than autonomous production. The strongest public case is GE Aerospace’s AI-enabled blade inspection tooling, with inspection times cut in half and the GEnx version in production use for about three years, part of a five-year MRO investment program exceeding $1 billion; that is an inspection and MRO result, not factory yield. AIA and McKinsey report that companies are prioritizing AI and automation in response to talent scarcity, and IATA urges wider adoption of predictive maintenance. Scope the tools accordingly: decision support within the validated envelope first, and treat anything that changes process parameters as a qualification event, planned as such.

The Monday agenda

Five actions follow. Build the node-level map of your critical supply base, site, process, and part rather than supplier names, and flag every node with few qualified alternatives, slow requalification, and rising demand. Re-underwrite make-buy and M&A decisions on qualification-adjusted replacement cost, and structure any deal to preserve the qualified entity. Fund second sources at the flagged single-source nodes, priced as insurance rather than procurement. Run funded-backlog burn and first-pass yield as the operating scorecard for each critical node, because that is where the next three years of revenue live. And point retention money and digital tools at bottleneck stations inside qualified operations before approving new footprint.

The industry’s blueprints have rarely been better, and the order book proves customers believe them. The bottleneck sits between blueprint and delivery, built out of approvals and rate readiness that exist only in specific places. Companies that manage qualified throughput on a node-by-node basis will convert this backlog. The rest will keep winning orders and watch someone else deliver them.