The Orbital Assembly Line: How Software-Defined Payloads, Sovereign Constellations, and the Serial Factory Broke Space's Five-Decade Guild
1. The Cleanroom at Montreal: The Death of the Artisanal Satellite
On the western tip of the island of Montreal, in Sainte-Anne-de-Bellevue, the satellite has stopped being a monument and started being a product.
MDA Space ($MDA) built its high-rate satellite manufacturing centre there for a single purpose: to produce spacecraft on a moving line, at a cadence the company describes in units per day rather than months per unit.2 The physical layout is the argument. In the old model, one spacecraft sat in one bay for two or three years while technicians in white smocks hand-built a wiring harness around it. In the new one, the work moves β payload panels populated, tested, and handed downstream, then integrated onto a bus that arrives configured rather than invented.
What comes off that line is the 225-satellite Lightspeed broadband constellation for Telesat and the replenishment spacecraft for Globalstar ($GSAT), the network that carries Apple's ($AAPL) satellite messaging feature and whose renewal Apple helped fund.210 In August 2026 Telesat expanded its order by C$474 million for 27 additional satellites, on top of an initial award above C$2.1 billion.2
Treat the cadence claim as a claim. What is audited is narrower and more useful: MDA reported second-quarter 2026 revenue of C$498.6 million, up 33.6% year over year, against a pro forma backlog of C$4.4 billion and full-year guidance of C$1.8β1.9 billion.2 A factory that converts a multi-year backlog into revenue at that rate is behaving like a manufacturer. A factory that does not is behaving like a research programme with a marketing department. The distinction between those two things is, in the end, the entire investment case for this industry.
To see why the Montreal line represents a break rather than an upgrade, look at what a satellite used to be.
A commercial geostationary communications satellite of the 2000s cost $200β400 million, weighed four to six tonnes, took 36 to 48 months to design, build, and test, and was expected to work for fifteen years without a service call 36,000 kilometres above the equator.9 Launch cost $15,000 to $25,000 per kilogram, so every gram bought with structure was a gram not sold as revenue-generating payload.9 Engineers responded rationally: exotic lightweight alloys, machined titanium, beryllium where the budget allowed, radiation-hardened components qualified over years to military specifications, and redundancy layered on redundancy because a single latch-up in orbit destroyed a nine-figure asset.
That engineering logic produced an industrial logic. Because each satellite was effectively a prototype, the work was organised as a guild: a handful of primes, cost-plus or heavily negotiated contracts, percentage-of-completion accounting, milestone billing, and program managers whose job was to absorb schedule risk. Learning curves β the thing that makes manufacturing profitable β barely formed, because you cannot descend a learning curve you climb only once every few years.
Then the price of the ride collapsed. Reusable boosters pushed launch cost per kilogram down by roughly an order of magnitude, and the scarce resource in the space economy moved. When getting to orbit is cheap and routine, the binding constraint becomes how fast you can build the thing that goes up, and how much useful capability you can pack into each watt of power it radiates.
The numbers describe a genuine industrial regime change. The Satellite Industry Association counted 4,434 satellites launched in 2025, and independent orbital tracking put 2026 on a run-rate above 4,800.14 Low Earth orbit now accounts for roughly 88% of units manufactured but only about 58% of manufacturing value, while geostationary orbit is down to some 2% of units and still 28% of value.9 That gap between where the units are and where the dollars are is the single most important fact in this industry, and most of this article is an explanation of it.
The economics of the two worlds are not comparable in the way league tables pretend. A modern proliferated-LEO satellite costs roughly $1.5β5 million, weighs 300 to 1,500 kilograms, and is expected to be replaced after three to five years.9 The satellite is no longer a capital monument to be depreciated over fifteen years; it is a consumable, and its replacement is a recurring order. Merchant builders β firms that sell hardware to third parties rather than to themselves β run gross margins in the 22β32% range, while diversified defence primes report space segment operating margins closer to 10β14%.10 Those are different accounting lines and should not be stacked, but the direction is real: a company that owns the design and sells it repeatedly earns differently from a company that assembles someone else's design once.
The deeper change is what the money buys. A satellite has become an industrialised consumer of advanced silicon and dense radio-frequency electronics wrapped in a structural chassis. Competitive advantage has migrated from the ability to survive launch risk to the ability to run a cleanroom at throughput without wrecking yield.
Which raises the obvious question, and the one a professional investor should ask before anything else: why should anyone believe the world needs several thousand new satellites every year, indefinitely?
Subscribe on LinkedIn2. The Upstream Belief: Why the Ground Must Move to Orbit
Every thematic investment rests on a belief about the physical world that is held one layer upstream of the companies involved. Stated plainly, and in a form that can be proved wrong, this one reads:
Terrestrial computing, defence sensing, and mobile connectivity have hit topological limits on the ground, and the classes of service that matter next β real-time tactical sensor-to-shooter loops, global edge data transport, and ubiquitous direct-to-device coverage β require physical infrastructure in low orbit operating under roughly 30 milliseconds of latency. If that is true, orbital hardware becomes a recurring industrial requirement rather than a discretionary capital project, and the constraint on the whole system moves from launch mass to payload manufacturing cadence and power density.
The belief is testable because each of its legs leaves evidence in different places, produced by parties with no incentive to corroborate one another.
The first leg is sovereign. The United States Department of Defense abandoned its own century-defining procurement habit. Instead of buying a handful of exquisite billion-dollar satellites β the SBIRS missile-warning and AEHF communications lineages β the Space Development Agency now buys proliferated constellations in two-year spiral tranches under firm-fixed-price contracts, with mandatory compliance to a published optical inter-satellite link standard so that hardware from competing vendors can talk to each other in orbit.5 The FY2026 budget request for the agency exceeded $4.2 billion, and Tranches 1 and 2 are fully obligated.5 A defence bureaucracy does not restructure its contracting model, surrender vendor lock-in, and accept fixed-price risk unless it has concluded that the capability is both necessary and buildable at industrial rates.
The second leg is commercial, and it arrived through a standards body rather than a boardroom. The 3GPP releases that define mobile networks β Release 17 and Release 18 β incorporated non-terrestrial networks into the standard, which means a satellite link is now a feature specification for ordinary smartphone basebands rather than a separate product category served by a $3,000 handset.10 Once the modem in a mass-market phone can, in principle, close a link with something in orbit, mobile operators start treating satellite capacity as a coverage line item β a substitute for building towers in places where towers never pay back β and that turns satellite payload capacity into an operating decision made by telecom procurement departments.
The third leg is a component story borrowed from the car industry. Because a LEO satellite is now designed to live three to five years rather than fifteen, builders can use automotive- and industrial-grade parts screened for radiation tolerance and protected by software-level error correction, instead of the slow, scarce, military-specification radiation-hardened silicon that space demanded for decades. Industry estimates put the resulting collapse in payload electronic bill-of-materials cost at 70% to 90%.9 That is the mechanism by which a satellite's brain got cheap without getting stupid.
Three independent constituencies β a defence procurement agency, a global telecoms standards process, and a semiconductor supply chain built for cars β arrived at the same conclusion within roughly five years of each other. That is what triangulation looks like, and it is why this theme graduated from the speculative listings boom of 2020β2021 into something an institution can underwrite: the demand is now anchored in funded programmes of record and signed commercial contracts rather than in slide decks.
The transmission into this industry's economics is direct. Sovereign programmes convert political decisions into multi-year funded backlogs with delivery milestones. Commercial constellations convert spectrum licences and financing into fixed-price orders. Both convert into factory throughput requirements, and throughput requirements convert into demand for the scarce physical inputs β space-qualified solar cells, radiation-tolerant processors, laser terminals, thermal vacuum test chambers β that determine who in the chain has pricing power. The same belief, incidentally, implicates several sibling industries an investor should expect to move in sympathy: terrestrial tower and rural-broadband operators facing orbital bypass, defence command-and-control systems being rearchitected around continuous low-latency downlinks, and compound semiconductor and advanced packaging suppliers selling gallium arsenide and gallium nitride into space-qualified power amplifiers.
A word on the decision context, because it governs everything that follows. This analysis is written for a global, multi-year thematic public-equity mandate, and it deliberately stops short of prescribing security weights. That matters because a long-only investor measured against a benchmark and a long/short investor measured on absolute return do not face the same problem here: the first is asking which listed vehicles carry genuine, disclosed exposure to a structural change; the second is asking where expectations have overshot the operating evidence. Both questions are answerable from the same facts. Neither is answered by the size of the end market.
The structural force is the migration of digital infrastructure into orbit. The investable theme is the manufacture of the payloads and buses that constitute it. The subthemes are software-defined communications payloads, defence sensing payloads, and merchant subsystems. The industry boundary excludes launch, ground networks, user terminals, and service operators β distinctions that will do real analytical work later, because several of the most expensive mistakes available in this sector come from crossing them carelessly.
And the clock is explicit. The base case requires the world to build and launch something like 4,800 to 6,200 satellites a year by 2028, with merchant manufacturers demonstrating durable double-digit operating margins on fixed-price work.9 If annual net additions to the operational fleet fall below roughly 1,500, or if fixed-price defence programmes generate the kind of late-stage rework charges that have historically savaged aerospace developmental contracts, the belief is fraying regardless of how many satellites are already up there.
Before the money can be traced, though, the object itself has to be understood β because the reason profit is moving is buried in the hardware.
3. Physics, Silicon, and SWaP-C: The Anatomy of a Modern Payload
Put a 500-kilogram proliferated-LEO satellite on a workbench and take it apart, and it separates into two economically distinct machines.
The first is the bus, or platform: the chassis. It holds solar arrays and batteries, thrusters and propellant, reaction wheels and star trackers for pointing, radiators and heat pipes for thermal control, and the flight computer that keeps the whole thing alive. The bus is infrastructure. It is what makes the satellite a satellite rather than a box of electronics tumbling through vacuum.
The second is the payload: the reason the mission exists. On a communications satellite that means antennas, radios, digital processors, and laser terminals. On an observation satellite it means telescopes, radar arrays, or spectrometers. The payload is what the customer is actually buying, and it typically accounts for 50% to 70% of the bill-of-materials value of a modern spacecraft.9
Engineers reconcile the two through an equation known as SWaP-C β size, weight, power, and cost. Every subsystem competes for the same fixed budget of mass, volume, and watts, and every design decision is a trade against the other three. The environment is unforgiving in ways that are easy to underestimate. In vacuum there is no air to carry heat away, so a satellite can only cool itself by radiating infrared into space; a dense digital payload is therefore constrained less by how much processing it can afford than by how much waste heat it can shed. Radiation degrades semiconductors continuously and occasionally flips a bit at random. Launch shakes the structure like an earthquake for eight minutes. Nothing can be repaired.
Four technologies inside the payload explain where the money now goes.
Active electronically scanned arrays are antennas made of hundreds or thousands of small transmit-receive elements whose signals are combined with controlled timing so the beam can be aimed electronically, in microseconds, without anything moving. The useful analogy is a stadium crowd doing a wave: no one runs anywhere, but the pattern travels. The analogy's limit is that the crowd makes one wave, while a phased array can synthesise many independent beams at once β which is precisely why it can serve thousands of separate users on the ground from a satellite crossing overhead at 27,000 kilometres per hour.
Digital channelizers are the payload's brain. A legacy satellite was, functionally, a mirror: it received analogue radio signals and reflected them back down on a different frequency, with the beam pattern and capacity allocation fixed by hardware at the moment of manufacture. A channelizer digitises incoming signals, routes and filters them in software, and reconstructs the downlink beams. That converts a satellite from a fixed asset into a reconfigurable one, capable of moving capacity to wherever demand appears after it is already in orbit β a hurricane, a shipping lane, a battlefield. It is closer to a router than a mirror, though the comparison understates the constraint: unlike a terrestrial router, it must do this within a power budget of a few kilowatts and never be rebooted by hand.
Optical inter-satellite links are laser transceivers that connect spacecraft to each other in orbit at gigabit rates. Their significance is architectural. Without them, every satellite must be in view of a ground station to be useful, which means coverage over oceans and denied territory requires an impossible ground footprint. With them, a constellation becomes a mesh β data entering over the Pacific can traverse orbit and exit over Virginia. They are also the industry's most acute manufacturing bottleneck, because pointing a laser at a moving object hundreds of kilometres away, from another moving object, tolerates almost no error, and because the Space Development Agency's interoperability standard limits how many vendors can qualify at all.5
The fourth is the quiet one: screened commercial silicon. Traditional space parts were built to military specifications, qualified over years, and priced accordingly. Modern LEO payloads increasingly use automotive- and industrial-grade components screened for radiation tolerance, with reliability recovered in software through techniques such as triple redundancy, where three copies of a computation vote on the answer. On a fifteen-year geostationary asset that trade is reckless. On a satellite designed to be deorbited in five years and replaced by a better one, it is the only sane choice β and it is the single largest reason payload costs fell.9
Now follow the physics into the pricing power, because this is where the industry's profit map is drawn.
The structural bus is being standardised into a configurable product. Rocket Lab's Photon, Lockheed Martin's LM 400, MDA's platform line, and the off-the-shelf buses from venture-backed entrants all converge on the same idea: a catalogue chassis with defined power, pointing, and propulsion options.10 Standardisation is good for customers and corrosive for margins, and bus platform work generally clears at roughly 18β28% gross margin.9
The payload does not standardise the same way. Software-defined channelizers and phased arrays run 30β45% gross margin, and the specialised subsystems one tier upstream β radiation-tolerant processors, triple-junction gallium arsenide solar cells, laser terminals, precision attitude-control hardware β run 35β55%.9 The barrier is not cleverness; it is qualification. Once a component is qualified onto a prime's bus and has flight heritage, replacing it means requalifying the interface, the thermal model, the software, and the test campaign. Switching costs of that kind are why component suppliers, not integrators, dictate schedules.
The supply chain evidence is unambiguous about who those suppliers are. Radiation-tolerant field-programmable gate arrays β the reconfigurable chips at the heart of digital payloads β come overwhelmingly from AMD ($AMD) through its Xilinx space-grade lines, with lead times of 36 to 48 weeks.10 Space-qualified triple-junction solar cells come from a duopoly of Rocket Lab ($RKLB), through its SolAero business, and Boeing ($BA), through Spectrolab, with lead times of 40 to 52 weeks.10 Roll-out solar arrays and deployable booms come from Redwire ($RDW).10 Laser terminals come from a short list including Mynaric, now inside Rocket Lab, and CACI.10 High-precision reaction wheels, control moment gyroscopes, and inertial measurement units for pointing-critical defence and observation payloads come substantially from Honeywell ($HON).10 Electric propulsion comes from Safran's space propulsion arm in Europe and from Busek and Enpulsion for smaller platforms.10
Those lead times are the industry's real clock. A prime can win a contract in a quarter and staff a factory in two; it cannot compress a 52-week solar cell queue, and a satellite missing its power system is not 95% complete, it is inventory.
The value of a modern satellite is measured in processing and optical throughput per watt radiated, not in tonnes of aluminium. Which is exactly what killed the companies that thought they were in the aluminium business.
4. The Value Chain and the Squeeze on Metal-Benders
In October 2024, Lockheed Martin ($LMT) completed the purchase of Terran Orbital for roughly $450 million.7 Terran was not a failed idea. It had a large factory in Irvine, California, a headline backlog in the hundreds of millions, and real customers, including Lockheed itself and defence constellation programmes. What it did not have was cash, and what it could not do was generate enough gross profit per satellite to fund the working capital that serial manufacturing consumes.
That is the industry's most instructive corpse, and it is worth being precise about the cause of death. Building bus structures without owning either the payload electronics or a sole-source component is a business that buys expensive inputs on supplier terms, adds labour and overhead, and sells to a small number of powerful customers on fixed-price terms with milestone billing. Cash goes out at the start of a build and comes back at delivery. Grow that business quickly and it consumes cash faster than it earns it. Grow it slowly and it never covers its factory overhead. Lockheed bought the capacity, folded it into a defence prime that can finance inventory out of a balance sheet generating billions in free cash flow, and the standalone small-satellite assembler as a public business model quietly ceased to exist.
Walk the chain from the bottom and the trap becomes structural rather than unlucky.
At the base sit raw materials and space-grade substrates: aluminium-lithium alloys, carbon composites, beryllium where mass matters most, and the high-purity germanium and gallium arsenide wafers on which space solar cells are grown. These are commodity businesses with specialised certification layered on top, earning perhaps 15β25% gross margin, and they have limited leverage over the tier above them.9
One layer up sit the specialised subsystems, and this is where the industry's pricing power actually lives. Radiation-tolerant processors, triple-junction solar cells, laser terminals, reaction wheels and star trackers, Hall-effect thrusters, deployable booms and roll-out arrays. Gross margins of 35β55%, extreme qualification barriers, frequently one or two qualified vendors per part, and lead times measured in seasons rather than weeks.910 A prime that needs these parts has almost no short-run alternative, and the parts are a small enough share of programme cost that fighting over price is rarely worth the schedule risk.
Above that sit payloads and buses. The payload β channelizers, phased arrays, optical and radar sensors β carries most of the satellite's value and 30β45% gross margins, because the capability is defined by proprietary radio-frequency engineering and software.9 The bus carries less and earns less, and is standardising toward catalogue economics.
At the top sits prime integration: assembly, integration, and testing. Enormous cleanrooms, vibration tables, acoustic chambers, thermal vacuum facilities, and the contractual obligation to deliver a working spacecraft on a date. Gross margins of 10β18%, and the position in the chain is structurally squeezed β powerful suppliers below, concentrated and price-setting customers above, and all of the schedule risk in the middle.9
The following exhibit shows what that squeeze has done to the industry's revenue mix.
Exhibit 1 β Global satellite manufacturing revenue pool, by segment (US$ billions, calendar years, worldwide, manufacturer-recognised hardware revenue)
| Year | Total | Proliferated LEO bus & payloads | Defence / sovereign bespoke | Commercial GEO platforms | Merchant subsystem sales |
|---|---|---|---|---|---|
| 2020 | $12.2B | $2.8B | $5.4B | $2.6B | $1.4B |
| 2022 | $15.8B | $5.2B | $6.8B | $2.0B | $1.8B |
| 2024 | $20.0B | $8.9B | $7.5B | $1.6B | $2.0B |
| 2025 | $22.8B | $11.2B | $8.4B | $1.1B | $2.1B |
| 2026E | $25.4B | $13.4B | $9.2B | $0.9B | $1.9B |
Definition: revenue recognised by manufacturers for satellite platforms, payloads, and subsystems; excludes launch, ground systems, terminals, and services. 2020β2025 are observed estimates from specialist industry census work; 2026E is a forecast, not observed data. Source: Satellite Industry Association and Novaspace/Euroconsult manufacturing market data.19
Read that aloud and the story is simple. The industry's revenue roughly doubled in six years, but almost none of the growth came from the segment that used to define it. Commercial geostationary platform revenue fell by roughly two-thirds, from $2.6 billion to under $1 billion, while proliferated-LEO hardware went from $2.8 billion to a forecast $13.4 billion. The industry did not grow; it was replaced.
The second exhibit shows where the profit inside that revenue now sits.
Exhibit 2 β Estimated share of industry gross profit pool by value-chain tier (percent of total, 2018 versus 2026E, global)
| Value-chain tier | Typical gross margin | Share of profit pool, 2018 | Share of profit pool, 2026E |
|---|---|---|---|
| Specialised subsystems (Tier 2) | 35β55% | ~22% | ~38% |
| Mission payloads (software-defined RF, sensors) | 30β45% | ~20% | ~29% |
| Bus platforms | 18β28% | ~23% | ~16% |
| Prime assembly, integration & testing | 10β18% | ~25% | ~11% |
| Raw materials & substrates | 15β25% | ~10% | ~6% |
Definition: estimated distribution of industry-wide gross profit across value-chain tiers. Margin ranges are drawn from specialist manufacturing market data and company segment disclosure; tier shares are Empor estimates derived from those inputs and are directional rather than measured. Evidence status: analytical estimate, not audited data.910
The reading is the thesis in one line: the tier that physically builds the satellite has gone from earning a quarter of the industry's gross profit to roughly a tenth of it, while the components and the payload electronics between them now take about two-thirds. Terran Orbital was positioned exactly where the profit was draining out.
Porter's framework is usually a way of avoiding specifics, so let it earn its keep here with actual mechanisms. Supplier power is high and verifiable: a 52-week solar cell queue and a handful of qualified laser terminal vendors are not abstractions.10 Buyer power is high but asymmetric β the Space Development Agency sets fixed prices and delivery dates, yet deliberately dual-sources across vendors, which prevents any single prime from being squeezed into monopsony ruin while also preventing any prime from extracting rents.5 In commercial constellations the buyer list is short enough to name β Amazon, Telesat, Globalstar β which gives each of them real leverage at the moment of contract award and none at all afterwards, once the design is locked.
Substitution barely exists. Nothing on Earth delivers wide-area sensing over denied territory or sub-30-millisecond global links from a tower. New entry is harder than the software-industry analogy suggests: designing a bus is a year of engineering, but security clearances, classified facilities, thermal vacuum chambers, and flight heritage are hundreds of millions of dollars and several years, and a customer who has never flown your hardware has an entirely rational reason to prefer someone who has.
Rivalry, finally, is bifurcated. In commodity bus assembly it is intense enough to produce bids with almost no margin. In classified radio-frequency and optical payloads, two or three credible vendors exist globally, and pricing behaves accordingly.
That bifurcation explains why the same headline β "satellite demand is booming" β has produced record backlogs at some companies and insolvency at others. It is also why the industry's history matters, because the last time capital rushed at orbital connectivity, almost everyone lost.
5. History's Ghosts: From the Iridium Crash to the Starlink Shock
In August 1999, Iridium filed for Chapter 11 bankruptcy protection. Sixty-six satellites were working in orbit, the network functioned as designed, and roughly $5 billion of capital was destroyed anyway.
The engineering had succeeded and the market had not appeared. Handsets were bricks costing thousands of dollars, calls cost several dollars a minute, and the phone did not work indoors β a defect that, for a device sold to travelling executives, was close to disqualifying. Globalstar followed into bankruptcy. Teledesic, the most ambitious of the lot, was cancelled before it flew. The lesson the industry drew was blunt: orbital capacity built ahead of terrestrial demand is a way of setting fire to money.
The consequences shaped satellite manufacturing for the next two decades. Capital fled commercial LEO. Manufacturers retreated to what was left β military programmes and direct-to-home broadcast satellites in geostationary orbit β where contracts were cost-plus or heavily protected, volumes were tiny, and the guild model was rewarded for exactly that. An entire generation of aerospace engineers learned that the way to survive was to build one flawless satellite very slowly.
Three developments broke that equilibrium, in sequence.
The first was small and unglamorous. Between roughly 2015 and 2018, imaging startups proved that commercial electronics could survive short missions in low orbit. Planet Labs ($PL) launched flocks of shoebox-sized Doves built with parts closer to a smartphone's than a satellite's, accepted that individual units would fail, and replaced them. Spire did the same for weather and ship-tracking. The insight was statistical rather than technical: with enough cheap units, reliability becomes a fleet property instead of a unit property. Planet remains an instructive hybrid today β it designs and builds its own hardware, including the higher-resolution Pelican and the hyperspectral Tanager spacecraft, on 2026 estimated revenue of roughly $260β290 million β but it sells imagery subscriptions, not satellites, so its manufacturing skill shows up as a cost advantage rather than as hardware revenue.10
The second was industrial, and it changed the price of everything. SpaceX built a satellite factory in Redmond, Washington, and ran it like a consumer electronics plant, reaching production above 2,000 satellites a year and, at peak, more than 40 units a week.10 Industry estimates put its unit cost below $800,000 for the Gen2 mini class.10 Those estimates deserve a caveat that matters: SpaceX is private, publishes no audited bill of materials, and does not disclose transfer pricing between its launch and manufacturing arms, so the figure is an inference from mass, manifest, and supplier evidence rather than a comparable accounting fact.10 It should be read as a directional cost floor, and it functions as one. Every merchant builder now competes against a number its customers believe to be true.
The third was geopolitical. Between 2022 and 2024, the war in Ukraine demonstrated in public what defence planners had argued in private: distributed commercial constellations kept working when centralised infrastructure was destroyed or jammed, and commercial radar imagery altered tactical decisions within hours. The Pentagon's response was institutional β the Space Development Agency and its proliferated architecture β and it converted a doctrinal argument into a funded, recurring procurement line.5
Alongside all this ran the capital cycle's contribution: the 2020β2021 listings boom, which financed a dozen sub-scale builders on the theory that a cleanroom plus a backlog equalled a business. When rates rose and the equity window shut, the ones without differentiated payload technology or defence programmes ran out of working capital. Terran Orbital's sale was the emblematic outcome, not an isolated misfortune.7
The volume record puts the eras side by side.
Exhibit 3 β Satellites launched worldwide, by mission category (units, calendar years)
| Year | Total launched | Commercial LEO constellation units | Civil / sovereign / defence units | Commercial GEO units |
|---|---|---|---|---|
| 2018 | 461 | 124 | 315 | 22 |
| 2020 | 1,283 | 982 | 286 | 15 |
| 2022 | 2,325 | 1,935 | 378 | 12 |
| 2024 | 3,145 | 2,680 | 455 | 10 |
| 2025 | 4,434 | 3,870 | 556 | 8 |
| 2026E | 4,950 | 4,280 | 660 | 10 |
Definition: spacecraft successfully placed in orbit, counted at launch, worldwide. 2018β2025 observed from orbital launch logs and industry census; 2026E is a run-rate projection, not observed data. Source: BryceTech orbital launch and payload log; Satellite Industry Association.14
Said aloud: in seven years the world went from launching under 500 satellites a year to more than 4,400, and essentially all of the growth came from commercial constellations. Meanwhile the commercial geostationary business β the segment that defined satellite manufacturing for forty years and still generates roughly a quarter of its value β shrank from 22 units a year to eight. Sovereign and defence launches nearly doubled, which is a smaller number telling a more important story, because those units carry far more revenue each.
So is this 1999 again? The comparison is the most common bear argument, and it deserves a fair hearing rather than a straw man. The structural similarities are real: capital-hungry operators, unproven consumer demand at scale, and constellations that must be replaced every few years whether or not subscribers arrive. Several of today's operators do not yet cover their own replacement capital expenditure from operations, and manufacturers booking their orders are therefore exposed to those operators' financing conditions as much as to their stated plans.
But three things differ materially. Demand no longer depends on persuading consumers to buy a special device, because the standards bodies put the capability inside phones that people already own. The largest single buyer is now a defence establishment that has rewritten its procurement doctrine around the architecture, which gives the industry a demand floor Iridium never had. And the cost base has fallen by more than an order of magnitude on both launch and payload electronics, which changes the revenue required to justify a constellation. Iridium's error was building supply thirty years before the silicon and the standards were ready. The risk today is different in kind: not that the market fails to appear, but that it appears and the economics of serving it are captured by two or three vertically integrated giants and a handful of component owners, leaving the assemblers in between with volume and no profit.
That risk is not distributed evenly around the world, because the industry has reorganised itself into three separate industrial systems that barely trade with one another.
6. The Tri-Polar Industrial Order: Primes, Consolidators, and the State
By the summer of 2026 the satellite manufacturing industry has resolved into three blocs, each with a different owner of capital, a different definition of success, and almost no ability to sell into the others.
The American system: fixed price, dual sourced, spiral.
The Space Development Agency is the most consequential institutional actor in this industry, and its power is procedural rather than technological. Under director Derek Tournear, the agency structured its Proliferated Warfighter Space Architecture around three rules that changed vendor economics.5 Contracts are firm fixed price, which transfers cost overruns from the taxpayer to the contractor. Capability is bought in two-year spiral tranches β Tranche 0, 1, 2, and a planned Tranche 3 β so that production lines stay warm and each generation can be improved rather than perfected in advance. And every vendor must comply with a published optical inter-satellite link standard, so satellites from different manufacturers interoperate in orbit.5
Each rule redistributed power. Fixed pricing rewards manufacturers with low costs and punishes those with cost-plus habits, which is precisely why merchant builders can compete with primes. Spiral tranches convert defence procurement into recurring industrial demand β the thing a factory needs and a job shop cannot use. The interoperability standard strips away the vendor lock-in that primes historically used to defend positions, and moves the moat one tier down, to whoever can actually build a compliant laser terminal.
The result is a genuinely contested market. Lockheed Martin has taken the largest cumulative dollar share of the architecture's awards across Tranches 0 through 2 and now owns internal small-satellite capacity through Terran Orbital, on a space segment generating roughly $12.5 billion in annual revenue.710 Northrop Grumman ($NOC) is a prime on Tranche 1 and 2 with a space business of roughly $11.2 billion, built on deep pedigree in national security payloads, missile-warning infrared sensors, and geostationary defence platforms β a portfolio that also carries the sector's characteristic hazard, since fixed-price development work on novel payloads has historically been where aerospace primes take charges.10 L3Harris ($LHX) leads the tracking side, where its electro-optical and infrared sensors detect and follow missiles against a cluttered thermal background. Boeing occupies an odd dual position: its legacy 702 geostationary platform sits in the shrinking high-value segment and has produced painful rework on the O3b mPOWER payloads, while its Millennium Space subsidiary builds small defence satellites quickly for classified customers.10
Against these sit merchant challengers with different cost structures. Rocket Lab won a Space Development Agency Tranche 2 transport layer award worth more than $515 million for 18 satellites and has accumulated over $1.3 billion in cumulative agency contracts.3 York Space Systems, majority-owned by BlackRock's private equity arm and therefore invisible to public market investors, was an early winner on Tranche 1 and 2 with more than $800 million in total contract value, using a deliberately commercial, low-cost bus architecture.10 Apex, a venture-backed Los Angeles builder with more than $100 million raised, sells standardised catalogue buses β Aries, Nova, Comet β on lead times under nine months, an approach that tests whether the bus can be a product rather than a project.10
And behind all of them are the two captive giants. SpaceX builds for Starlink and sells externally only through its Starshield defence line, which means its manufacturing scale sets the industry's cost expectations without ever competing for a merchant contract. Amazon ($AMZN) built its Kuiper satellite plant in Kirkland, Washington, targeting production rates of up to five satellites a day, financed from a parent balance sheet rather than from project economics.10 Kuiper's deployment obligations under its US spectrum licence included a milestone requiring roughly half the constellation β 1,618 satellites β by July 2026, a date that has now passed; the regulatory record on extensions and compliance is not established in the evidence available here, and investors should treat the outcome as open rather than assume either enforcement or forgiveness.10
The European system: consolidation under sovereign protection.
Europe arrived at the same industrial problem from the opposite direction. Its primes were losing money in commercial geostationary telecommunications, the segment collapsing fastest, while competing against a vertically integrated American rival with its own rockets. In October 2025, Airbus ($AIR), Thales ($HO), and Leonardo ($LDO) signed a memorandum of understanding to combine their space activities into a single venture β informally, Project Buomo β with Airbus holding 35% and Thales and Leonardo 32.5% each, targeted to become operational in 2027 and combining more than β¬5.5 billion of space systems revenue.6
The political economy behind it is explicit. The European Commission and the European Space Agency structured the IRISΒ² sovereign constellation to guarantee industrial workload for European manufacturers, which supplies the merged entity with an anchor programme that no non-European competitor may bid for. What the merger buys is the elimination of duplicated research and development across three national champions; what it must survive is multi-country governance, labour agreements in three jurisdictions, antitrust remedies, and the restructuring charges attached to legacy geostationary capacity. Final approvals were still pending as of publication, with completion expected around mid-2027.6
An investor should hold both facts at once: this is a protected business with contracted sovereign demand, and it is a restructuring with execution risk, assembled from divisions that were losing money on the products being consolidated.
The Chinese system: the state as customer, financier, and spectrum strategist.
China's approach is the most vertically integrated of the three. In 2020 the State Council and the state assets regulator classified satellite internet as part of national "new infrastructure" (ζ°εΊε»Ί), a designation that in the Chinese system directs financing, land, and provincial support toward a sector.10 Two megaconstellations followed. China Satellite Network Group β Guowang β is a centrally owned enterprise managing a national constellation planned at around 13,000 satellites, supplied largely by the state aerospace prime CASC and its satellite arm CAST, which also builds China's geostationary platforms and remote sensing fleets.10 The second, Qianfan, also called the G60 constellation, is run by Shanghai Spacecom Satellite Technology with backing from the Shanghai municipal government and manufacturing through the automated Gesi Aerospace line, reported at roughly 300 satellites a year of capacity.10
Here the source hierarchy matters. That satellite internet is a state priority, and that Shanghai has organised municipal industrial policy around Qianfan, is established by official policy channels and is authoritative evidence of intent. Performance is a separate question requiring independent evidence. Independent orbital tracking shows 238 or more Qianfan satellites deployed as of August 2026 against a stated 2030 target of 15,000 β real progress, and far behind the pace the target implies.48 In August 2026 the company closed a Series B of 7 billion yuan, roughly $1.94 billion, at a reported valuation near $7 billion.8 Claims of 300β500 satellites per year of automated capacity come from company and municipal statements; on-orbit payload reliability, achieved throughput, and yield cannot be independently verified to Western audit standards, and should be treated as unconfirmed.
The strategic logic is spectrum. International Telecommunication Union rules require constellation operators to deploy 10% of a licensed system within two years of filing, 50% within five, and 100% within seven, or forfeit the allocation.10 Orbital shells and frequencies are assigned in the order they are brought into use, which turns manufacturing cadence into a claim on a permanent global resource. That is why states are financing factories rather than waiting for markets.
Three systems, three sets of rules, and export controls between them. American and European regulations wall Western payloads off from Chinese vehicles and non-allied customers; Chinese manufacturers are excluded from Western commercial markets by the same regime. Prices will not converge globally, and scale achieved in one bloc does not translate into share in another. Leadership in this industry is therefore always a claim about a defined parameter inside a defined bloc β and the merchant leaders in the Western bloc got there by two very different routes.
7. How the Merchant Leaders Built Their Leads
Two companies have solved the merchant manufacturer's problem β how to sell hardware to other people and still earn a return β and they solved it in opposite directions. One went deep into the payload. The other went wide across the components.
MDA Space: own the brain, rent out the body.
MDA's inheritance was robotics and radar. It built the Canadarm manipulators for the Space Shuttle and the International Space Station and the RADARSAT synthetic aperture radar satellites for the Canadian government β high-prestige, low-volume programmes that produced world-class radio-frequency and control engineering and almost no manufacturing scale.10
The strategic decision that produced today's position was a judgement about where commoditisation would land. Management concluded that satellite structures would become catalogue products and that the durable differentiation would sit in the payload: digitally reconfigurable channelizers and active phased arrays that let an operator move capacity around the planet after launch. So MDA spent heavily on that architecture and on an automated production facility capable of building it at rate, rather than on becoming the cheapest bender of aluminium.2
The payoff is legible in the order book. MDA is prime contractor for Telesat's 225-satellite Lightspeed constellation under an award above C$2.1 billion, expanded in August 2026 by C$474 million for 27 more satellites, and it is building the Globalstar replenishment fleet whose service Apple underwrites.2 Backlog stood at C$4.4 billion on a pro forma basis in the second quarter of 2026 against full-year revenue guidance of C$1.8β1.9 billion β roughly two and a half years of production already contracted.2
State the leadership claim precisely, because vague crowns are worthless. On the parameter of merchant serial manufacturing scale for commercial communications satellites with proprietary digital payloads, as of August 2026, MDA leads, and its closest challenger is Rocket Lab's space systems division. Customers care because a software-defined payload lets an operator redeploy capacity toward demand that did not exist when the satellite was ordered β a materially different asset from a satellite whose beams were fixed in a factory years earlier. MDA leads rather than its rivals because digital payload engineering requires decades of accumulated radio-frequency and signal-processing know-how combined with the capital to build automated test infrastructure, and because building at rate requires the two capabilities in the same building. Competitors who buy payload electronics from third parties remain, structurally, assemblers of other people's value.
What could erase it? Concentration, mainly. A large share of MDA's medium-term backlog rests on Telesat, and Telesat is a capital-hungry operator financing a constellation in a market with a shorter track record than its business plan requires. That is not a criticism of MDA's execution; it is a statement about where the risk actually sits, and it is the first thing a sceptical investor should test.
Rocket Lab: own the parts everyone else needs.
Peter Beck's company began as a launch business and drew an uncomfortable conclusion from its own income statement: launch revenue is lumpy, capital-intensive, and structurally deflationary. So Rocket Lab went shopping in the tier of the value chain where margins are highest and buyers have no alternatives.
The acquisitions read like a component catalogue: Sinclair Interplanetary for reaction wheels and star trackers, SolAero for space-grade triple-junction solar cells, Advanced Solutions for flight software, Planetary Systems for separation systems, and the assets of the laser terminal specialist Mynaric.310 Each purchase bought a qualified, flight-proven part with an installed base at other primes.
The result is an unusual dual monetisation. Rocket Lab now sources more than 60% of a satellite bus internally by value, which lets it bid aggressively for prime contracts such as the $515 million Tranche 2 transport layer award.310 Simultaneously, it sells those same components to the companies bidding against it β SolAero cells fly on spacecraft built by Lockheed Martin, Northrop Grumman, and MDA, among others.10 When a rival wins, Rocket Lab still ships product.
The leadership claim here is narrower and better evidenced than "Rocket Lab is winning space." On the parameter of component vertical integration among Western merchant satellite builders, measured as internally sourced share of bus value, Rocket Lab leads with more than 60%, and its closest merchant comparison is Redwire, which owns critical parts but does not build complete spacecraft at scale.10 Customers care because internal sourcing shortens the lead times that otherwise govern delivery, and because it converts supplier margin into own margin. Rocket Lab accumulated the position by buying distressed or sub-scale specialists during the post-2021 capital drought at prices no strategic buyer would have accepted in 2021 β timing, capital access, and a willingness to integrate unglamorous businesses.
That lead is real, and it has a mirror-image vulnerability: the company is simultaneously executing large fixed-price defence builds and developing a new launch vehicle, Neutron, and those two programmes compete for engineering attention and cash. A merchant supplier that misses component deliveries because its own prime programme took priority would damage the very relationships that make the roll-up valuable.
L3Harris: the cornered resource that cannot be bought.
The third lead is the least visible and possibly the most durable. On the parameter of electro-optical and infrared missile-tracking payloads for proliferated low-orbit defence constellations, L3Harris leads, with Northrop Grumman the closest challenger.10 The capability is the ability to detect a hypersonic vehicle or ballistic missile against the thermal clutter of the Earth's surface from low orbit, which depends on sensor design, calibration technique, and processing algorithms refined against real on-orbit data over many years. New entrants cannot simulate their way to that; they need flight heritage they do not have, and they need the classified facilities and cleared personnel to do the work at all.
Run these three through Hamilton Helmer's categories and the powers are concrete rather than decorative. L3Harris holds a cornered resource β classified calibration data and cleared infrastructure β that cannot be purchased at any price by an uncleared competitor. MDA holds switching costs: once a constellation's ground network, flight software, and spectrum plan are designed around a specific digital payload architecture, changing prime mid-constellation costs years and hundreds of millions. Rocket Lab holds scale economies in components, spreading qualification and tooling costs across units sold to itself and to its competitors. All three hold process power in the narrow sense that yield in space hardware manufacturing is accumulated slowly and cannot be hired.
Contrast this with what the primes hold. Lockheed Martin's advantage is capital and relationships: an AAA-tier balance sheet generating more than $6 billion of consolidated free cash flow a year can finance inventory through a launch delay that would bankrupt a small manufacturer, and decades of classified programme execution buys access no newcomer can bid for.10 That is a real and durable advantage. It is also, in this particular industry, an advantage of a lower order than owning the payload, because it does not expand margins β it merely permits survival at 10β11% operating margin while others earn more per dollar of revenue.
Which brings the story to the place where these differences are supposed to be adjudicated: the price the public market puts on them.
8. Financial Read-Through, Valuation Dispersion, and the Expectations Gap
Put the listed universe side by side and the dispersion is striking enough to demand an explanation.
Exhibit 4 β Financial and valuation read-through, listed satellite manufacturing exposure (as at August 2026)
| Company | Revenue base and growth | Margin (basis stated) | Backlog | Valuation |
|---|---|---|---|---|
| MDA Space (TSX) | C$1.8β1.9B FY2026 guidance; Q2 2026 revenue C$498.6M, +33.6% y/y | ~28% gross; ~20% adjusted EBITDA (consolidated) | C$4.4B pro forma | ~12.5x 2026E EV/EBITDA |
| Rocket Lab (Nasdaq) | Q2 2026 revenue $234.1M, +62% y/y; Space Systems ~$185M, roughly doubled y/y | ~32% gross, Space Systems segment; consolidated adjusted EBITDA near breakeven | $2.36B total, incl. >$1.3B cumulative SDA | ~5.8x 2026E EV/sales |
| Redwire (NYSE) | $300β320M 2026E revenue, +18.5% y/y | ~24% gross; ~8.5% adjusted EBITDA (consolidated) | $350M+ | ~1.2x EV/sales; ~9.5x EV/EBITDA |
| L3Harris (NYSE) | ~$4.5B Space & Airborne Systems segment, +6.8% y/y (of ~$21B group) | ~15.5% segment operating margin | >$32B company-wide | ~14.8x 2026E EV/EBITDA |
| Lockheed Martin (NYSE) | ~$12.5B space segment, +4.5% y/y | 10.2β11.0% segment operating margin | Multi-billion, largely classified | ~13.2x 2026E EV/EBITDA |
Definitions and comparability: MDA figures are in Canadian dollars and consolidated; Rocket Lab's margin is a segment figure and its EBITDA is consolidated including launch development; L3Harris and Lockheed Martin figures are segment operating margins, which sit below gross margin and are not comparable to the gross margins shown for MDA, Rocket Lab, and Redwire. Backlog definitions differ by issuer and include funded and unfunded elements. Multiples are market-observed as at August 2026. Sources: company quarterly disclosures and investor materials.2310
Read aloud, that table says something specific. A company growing revenue above 30% with roughly 20% EBITDA margins and two and a half years of contracted work trades at about 12.5 times EBITDA β slightly below a defence prime growing at 4.5% with an 11% segment operating margin. The market is applying, roughly, a defence contractor multiple to a business whose operating characteristics increasingly resemble an industrial equipment manufacturer.
There are respectable reasons for that, and an investor should walk through them before concluding anything is mispriced.
The first is revenue recognition. Satellite manufacturers book much of their revenue on percentage of completion, which means reported revenue reflects estimated progress and estimated total cost. When those estimates prove wrong, the correction lands in a single quarter as a charge. Aerospace history is littered with them, Boeing's mPOWER payload rework being a recent example inside this very industry.10 A market that discounts percentage-of-completion revenue is not being stupid; it is pricing a known accounting hazard.
The second is customer concentration. MDA's backlog leans heavily on Telesat's Lightspeed programme, and Telesat is an operator that must finance a constellation before it earns from it.2 Backlog from a customer whose funding depends on capital markets is worth less than backlog from a defence appropriation, and the discount is rational.
The third is fixed-price risk. The Space Development Agency's contracting model transfers overruns to contractors.5 Primes assert that efficient builders can earn 15β20% operating margins under it. That assertion is a claim, not a result: the deliveries that would prove it were still moving through environmental acceptance testing in late 2026, and no full tranche has yet been delivered, flown, and margin-reconciled across a complete programme.10 Treating management's target margin as an established outcome is exactly the error that turns a correct industry forecast into a poor security outcome.
Now the other side, stated with equal discipline. If serial production holds β and the observable evidence for it is backlog converting to revenue at 38.5% to 44% a year across MDA and Rocket Lab's space systems business, with gross margins at 28% and 32% respectively β then these are not project businesses at all.2310 They are manufacturers with a replacement cycle: satellites die in three to five years and must be rebuilt, which produces recurring rather than one-off demand. Businesses with that profile and those margins typically clear at premium industrial multiples. The variant view here is about business model classification, and it is falsifiable: if merchant gross margins fail to expand as unit volumes rise across hundreds of deliveries, the classification is wrong and the current multiple is correct.
Each expression needs its own rejection test, because a theme-level conclusion is not a security-level one.
For MDA, the first thing a senior investor would challenge is whether C$4.4 billion of backlog is a manufacturing pipeline or a financing bet on one customer. What would make it deeper work: evidence of order diversification beyond Telesat and Globalstar at comparable margins. What would kill it: a Lightspeed funding interruption, or gross margin compressing below 20% on volume deliveries.
For Rocket Lab, the challenge is that consolidated free cash flow is negative and the reported growth blends acquired components, a launch business, and prime contracts with different economics. The company does disclose Space Systems separately, which is the right granularity and more than most peers offer.3 What would make it deeper work: Space Systems sustaining low-30s gross margins while prime programme deliveries are at peak. What would kill it: component delivery slippage to third-party primes, or a Tranche 2 schedule breach that triggers liquidated damages.
For Redwire, the exposure is the purest in the listed universe β roughly 90% of revenue from satellite hardware, solar arrays, and space mechanisms β and the constraint is equally pure: a cash buffer in the $35β50 million range against working capital swings on large deliveries.10 A company that must fund inventory before it bills can be right about the industry and still be forced to raise equity at a bad price. That is a capital structure risk, not a thesis risk, and it is the reason its exposure purity and its valuation of roughly 1.2 times sales coexist.
For L3Harris and Lockheed Martin, the honest description is different. Space represents roughly 20% and 18% of revenue respectively, so a correct thematic call produces a diluted result; these are defensive holdings whose cash flows are insulated from commercial constellation financing cycles, and whose upside from the theme is real but modest relative to group size.10
Three false positives deserve naming, because they are how money gets lost in themes like this.
The first is the ecosystem trap: companies whose association with space is marketing rather than revenue β terrestrial fabrication shops, additive manufacturing bureaus, software vendors β with no qualified flight heritage and no disclosed prime contracts. Compatibility with an industry is not exposure to it, and available disclosure for such firms typically does not establish measurable revenue from this theme at all.
The second is the unfunded constellation trap. Announced procurement plans from operators without committed financing or secured spectrum should not be credited to a manufacturer's backlog, however impressive the headline number.
The third is treating launch as a proxy for satellites. Launch pricing is deflationary by design and its capacity is being expanded aggressively; payload electronics carry qualification-based moats that launch does not. A portfolio built on the assumption that these two move together owns one bet twice, badly.
One more evidentiary distinction matters. Private valuations are not comparable to public ones: Shanghai Spacecom's roughly $7 billion valuation in an August 2026 primary round, or Apex's venture rounds, reflect negotiated preference structures, strategic and state objectives, and illiquidity, and cannot be read across to a listed multiple.810 They are evidence about capital availability, not about value.
All of which leaves the question of how much hardware the world will actually order.
9. Adoption Scenarios, S-Curves, and Capital Cycles (2026β2030)
The denominator for adoption in this industry is unusual: it is not households or subscribers but operational satellites in orbit, and the replacement rate of that fleet.
Roughly 8,500 commercial and defence satellites were operating in low orbit as of August 2026, against annual production tracking near 4,950 units for the year.410 Because a proliferated-LEO satellite is designed for three to five years of life, a stable fleet of that size generates replacement demand of perhaps 1,700 to 2,800 units a year on its own, before any new constellation is deployed. That is the structural floor beneath the industry, and it is the single most important adoption fact: this business has crossed from installation demand into recurring replacement demand for the first time in its history.
The prerequisites governing further growth are physical and regulatory rather than commercial. Spectrum must be secured under the International Telecommunication Union's deployment milestones, which forfeit allocations for missed schedules. Orbital slots must be licensed, and in the United States satellites must carry the propulsion and passivation needed to deorbit within five years of mission end.10 Launch capacity must be available on the date the satellite is finished, because a completed satellite waiting in storage is uninvoiced inventory. And the physical supply of solar cells, radiation-tolerant processors, and laser terminals must scale, because none of them can be conjured in under a year.
Three scenarios follow, and they differ by causal path rather than by percentage.
The bear world is a financing failure, not a technology failure. Commercial constellation operators fail to generate operating cash flow exceeding their four-year replacement capital expenditure. Credit spreads widen, the equity window closes, and the operators that were going to place the next round of orders instead defer them. Defence budgets flatten. Annual production settles at 2,500 to 3,200 units, the hardware market contracts toward $18.5 billion, and the industry's operating margin pool falls to 6β8.5%.9 In that world the survivors are the diversified defence primes whose balance sheets and appropriated backlogs are indifferent to commercial credit conditions, and the casualties are venture-backed integrators without sovereign programmes. Note the asymmetry: adoption of satellite services could still rise in this world while manufacturers destroy capital, because the operators would be sweating existing assets rather than ordering new ones.
The base world is sovereign demand carrying commercial demand. Tranches 2 and 3 are funded and awarded, Lightspeed and Kuiper deploy roughly on plan, China continues expanding Qianfan, and the three-to-five-year replacement cycle establishes itself as a routine order flow. Production runs 4,800 to 6,200 units a year, the market reaches about $31 billion by 2028, and the industry operating margin pool sits at 13.5β17%.9 Scaled merchant builders with payload technology and vertically integrated component suppliers take a disproportionate share of that pool; un-automated machine shops and manual geostationary integration lines are stranded.
The bull world requires the phone in your pocket. Direct-to-device connectivity becomes a standard feature across 3GPP releases, mobile operators contract for orbital capacity as routine coverage infrastructure, and defence spending on space domain awareness and electronic warfare compounds. Production reaches 8,500 to 12,000 units annually, the market approaches $48.5 billion, and margins reach 20β25.5%.9 The beneficiaries shift upward: at that volume, the constraint becomes laser terminals and high-density digital processing, so the component owners capture more than the integrators.
Now place the capital cycle beside that adoption curve, because they are not the same curve and confusing them is the classic error in infrastructure investing.
The evidence says this industry is in early deployment following a shakeout. The 2020β2022 boom funded far more bus assembly capacity than the profit pool could support; rates rose; the sub-scale builders were absorbed or liquidated, with Terran Orbital's distressed sale the clearest marker.7 Capital is now concentrating in fewer hands and being spent on automation β MDA's Montreal expansion, Rocket Lab's Long Beach and Albuquerque facilities β rather than on new entrants' first cleanrooms.23
That is the healthy phase of a capital cycle, and it contains its own seed of decay. Automated capacity is being installed against forecast volumes. If the base case arrives, utilisation is high and the operating leverage is spectacular. If the bear case arrives, the same automation becomes fixed cost against declining volume, and a business that looked like a manufacturer reverts to looking like a job shop with an expensive building.
The specific mechanism to watch is the coupling between factory output and launch manifest. Manufacturing is now capable of running faster than the launch fleet in some quarters. When that happens, finished satellites accumulate on manufacturers' balance sheets awaiting delivery milestones, working capital balloons, and the companies least able to fund it are precisely the small pure-plays with the purest thematic exposure. Falling launch prices are wonderful for constellation operators and neutral-to-negative for a manufacturer whose cash is tied up in a spacecraft that cannot ship.
Who bears which risk is worth stating plainly, because the answers differ. Operators bear demand and financing risk. Manufacturers bear schedule, working capital, and fixed-price execution risk. Component suppliers bear capacity risk and enjoy the pricing power. Governments bear the residual β and, through appropriations, provide the floor that makes the whole structure investable.
Which means the useful monitoring set is small, physical, and mostly not found in earnings headlines.
10. The Crux KPIs and the Falsification Dashboard
Most of what this industry publishes about itself is noise: memoranda of understanding, constellation announcements, artist's renderings, and target production rates. Four observables actually sit on the binding constraints, and each of them moves before revenue, margin, or share does.
One: delivery cadence on fixed-price defence programmes, measured in months from contract award to shipment to the launch site. The latest readings run 23.5 to 26 months against a 24-month design target for Space Development Agency tranches.510 This is the leading indicator for whether serial manufacturing discipline actually survives contact with government payload requirements, because under fixed-price terms every month of slip is contractor cost. It discriminates the central bull-bear disagreement of the whole theme: bulls believe these are factories, bears believe they are development programmes wearing a factory's clothes. Source: agency programme milestones, disclosed quarterly. Confirmation is delivery at or under 24 months; the thesis breaks if slippage exceeds 32 months across more than half of primes, because at that point liquidated damages and rework charges convert the industry's flagship demand source into a margin sink.
Two: payload value density, the dollar value of payload electronics per kilogram of payload mass, currently $8,500 to $14,000 per kilogram.9 This measures directly whether value is still migrating from structure into signal processing and optics, which is the mechanical heart of the profit-pool argument. It leads margins because pricing per kilogram changes at contract award, quarters before it shows up in a gross margin line. It sits at the industry level of the hierarchy rather than the company level. The measurement caveat is real and should not be glossed: this is a specialist estimate derived from manufacturing price surveys, not an audited series, and it is best used as a directional proxy. Source: Novaspace/Euroconsult manufacturing price data, semi-annual. Expansion above roughly 8% a year confirms; two consecutive years of contraction would indicate payloads are commoditising, which would move the whole thesis one tier upstream to components.
Three: backlog conversion velocity β the share of opening twelve-month funded backlog recognised as revenue over the following four quarters β currently 38.5% to 44% at MDA and Rocket Lab's space systems business.2310 This is the cleanest available test of whether a backlog is manufacturing activity or paper. It leads cash flow directly, and it is company-level, disclosed quarterly in audited filings under US and Canadian rules, which makes it the most reliable of the four. Sustained conversion at or above 40% with stable gross margins confirms industrialisation; conversion falling below 28% would say the factories are bottlenecked, almost certainly by components, regardless of how large the order book looks.
Four: space-qualified triple-junction gallium arsenide solar cell output, in megawatts delivered per quarter, currently around 1.85 megawatts globally in merchant capacity.10 Every satellite needs power, the qualified supply is a duopoly of Rocket Lab's SolAero and Boeing's Spectrolab, and lead times already run 40 to 52 weeks. This is the hardest physical ceiling on total Western satellite production, and it is upstream of everything else in this article. Source: supplier production disclosure, quarterly, with the honest caveat that neither supplier publishes a clean megawatt series and the figure is assembled from partial disclosure. Expansion above 2.5 megawatts a quarter by mid-2027 would confirm that the constellation ramp is physically feasible; stagnation below 1.5 megawatts with rising spot prices would cap global output no matter how many rockets are available or how many contracts are signed.
Arrange these into a hierarchy and the monitoring discipline becomes clear. At the structural level sit defence appropriations β sustained funding above roughly $3.5 billion a year for the proliferated architecture confirms; a cut exceeding 25%, or abandonment of the spiral tranche model, would kill the theme's demand floor.5 At the adoption level sit net additions to the operational fleet, tracked monthly by orbital catalogues: above 3,500 a year confirms, below 1,500 breaks.4 At the industry level sit merchant book-to-bill, currently running 1.3 to 1.6 times, and component lead times.10 At the company level sit segment gross margins and backlog conversion. At the market level sits credit availability for constellation operators, which is the transmission channel through which interest rates reach this industry's order intake.
Separate theme kill criteria from security kill criteria, because they are different events. The theme dies if defence procurement reverses or if annual deployment collapses. An individual security can die while the theme thrives β through a customer's financing failure, a fixed-price charge, an equity raise forced by working capital, or simply by having been bought at a price that already assumed flawless execution.
The portfolio risks worth naming are mostly correlations hiding as diversification. Owning a merchant manufacturer, a component supplier, and a defence prime looks like three positions and is substantially one bet on the appropriations cycle for a single architecture. Owning a manufacturer and its customer doubles exposure to the same constellation's financing. Owning a satellite builder and a launch company doubles exposure to the same manifest while adding a business with structurally worse economics. And the entire complex carries embedded sensitivity to real interest rates through operator financing, to export control policy through addressable market, and to a small number of flagship programmes whose schedules move several income statements at once.
The rotation logic follows from the KPIs rather than from opinion. If payload architectures standardise into open, interoperable modules β which the interoperability mandates make more likely, not less β then the payload's margin premium compresses and the defensible economics move upstream to laser terminals, radiation-tolerant processing, and compound semiconductor packaging. The payload value density series is the signal that would say so, and it would say so before any income statement did.
11. Game Changers, Future Risks, and the Next Frontier
Three developments could invalidate parts of this analysis, and each works through a different mechanism.
Ultra-heavy reusable launch removes the constraint the entire discipline was built around. If fully reusable heavy vehicles reach routine cadence and launch cost per kilogram falls toward $200, mass stops being scarce.9 The engineering consequences are counterintuitive: satellite builders would stop spending millions on exotic lightweight structures and start building spacecraft out of cheap heavy materials with generous margins, the way terrestrial industries do. The commercial consequence is that bus cost falls further and faster than payload cost, accelerating the migration of value into apertures and electronics. Larger fairings also permit physically larger antennas and optics, which improves link budgets and sensor resolution in ways that no amount of signal processing can substitute for.
The beneficiaries would be payload specialists and aperture builders; the losers would be precision machining suppliers and any manufacturer whose competitive advantage is mass efficiency. The observable milestone is not a test flight but a published, repeatedly achieved commercial price per kilogram at a sustained cadence, which does not yet exist. Separate the demonstrated from the scaled: a vehicle that flies is not a price list.
Direct-to-device standardisation could turn payloads into cellular infrastructure. Successive 3GPP releases have been pulling non-terrestrial networks deeper into mainstream mobile standards, and the endpoint is a satellite payload that behaves like a standardised base station in orbit rather than a proprietary communications system.10 For manufacturers with software-defined channelizer technology, that would be a demand surge, because mobile operators buy standardised infrastructure in volume and on schedule.
It cuts both ways. Standardisation is what destroyed proprietary margins in terrestrial telecoms equipment, and there is no obvious reason space would be exempt. The plausible resolution is that the payload becomes standardised and the money moves to whoever supplies the highest-performance components within the standard β the same pattern that made a handful of chip and optics suppliers wealthy while base station assembly commoditised. The milestone to watch is commercial service revenue from standards-based satellite connectivity carried on operators' own income statements, not handset feature announcements.
On-orbit servicing could break the replacement cycle that makes this industry investable. Autonomous rendezvous and docking β demonstrated by Northrop Grumman's mission extension vehicles and pursued commercially by Astroscale and Starfish Space β would decouple a satellite's life from its propellant.10 Design lives could lengthen, and payloads could in principle be swapped rather than replaced.
Here the investor should notice a tension the enthusiasts skip. The recurring three-to-five-year replacement order is precisely the characteristic that converts satellite manufacturing from a project business into an industrial one. Extend asset life materially and the replacement flow thins. The offsetting possibility is that servicing changes what gets built β spacecraft designed with standard docking interfaces and modular payload bays β which would shift value further toward payload modules and away from complete vehicles. Either way, the profit pool moves; the question is which direction, and the honest answer today is that commercial servicing revenue remains small and the demonstrations do not yet constitute a business.
Three risks sit alongside these opportunities, and they are not symmetrical.
Orbital debris is the tail risk with no hedge. The nightmare case, in which collision fragments cascade through a populated shell, does not need to occur to hurt the industry; a serious incident in the 500β600 kilometre band or in sun-synchronous orbit would be enough to trigger deployment moratoria, mandatory insurance requirements, or licensing delays. That would hit manufacturers through deferred orders while their fixed costs continued.
Spectrum politics is the slower risk. Orbital frequencies are allocated internationally and claimed by deployment. As Chinese and Western constellations both scale, coordination disputes at the International Telecommunication Union become a channel through which geopolitics reaches manufacturing schedules directly, since a constellation that cannot coordinate spectrum cannot justify its build.
Export control is the risk most likely to be underestimated by investors modelling global market size. American and European regulations define which customers exist for Western payloads, and the same regime excludes Chinese manufacturers from Western markets entirely.10 Any global total addressable market calculation that ignores this is describing a market no company can actually sell into. A tightening of controls would shrink the addressable market for merchant builders; a loosening β improbable in the current environment β would introduce a low-cost competitor into markets that currently exclude it.
Underneath all three sits the destination the technology keeps pointing toward: satellites as standardised compute and communications nodes, with the boundary between a terrestrial data centre and an orbital platform becoming an engineering detail rather than a category. That is a long way from being demonstrated commercially, and it should be labelled as direction rather than forecast. But it is the direction in which every one of the constraints described in this article is being relaxed.
12. Conclusion: Testing the Upstream Belief
Return to the cleanroom outside Montreal, where a completed digital payload is sealed into a shipping container bound for a launch site in Florida. Nothing about the moment is heroic. A serial number is recorded, a container is loaded, and the line behind it keeps moving. That is what industrialisation looks like when it finally arrives in an industry that spent fifty years insisting it was impossible.
The belief this article set out to test held that terrestrial computing, defence sensing, and mobile connectivity had reached limits that could only be relieved by physical infrastructure in low orbit, and that this would move the space economy's binding constraint from launch mass to payload manufacturing cadence and power density. Three parts of that claim can now be assessed against evidence rather than assertion.
The physical claim is holding. Nothing on the ground substitutes for wide-area sensing over territory you do not control or for global links under 30 milliseconds, and the buyers who most need those capabilities have restructured their procurement around them rather than merely funding studies. Annual launches rose from 461 in 2018 to 4,434 in 2025, and the fleet has passed the point where replacement alone sustains a substantial industry.14
The transmission claim is holding, and it is visible in the revenue mix rather than in rhetoric. Commercial geostationary platform revenue fell from $2.6 billion in 2020 to a forecast $0.9 billion in 2026, while proliferated-LEO hardware went from $2.8 billion to $13.4 billion over the same period.19 The industry's centre of gravity relocated, and the companies that could not move their factories with it were absorbed or wound down. Terran Orbital's sale to Lockheed Martin remains the clearest single piece of evidence, because it showed that a full order book cannot rescue a position in the wrong tier of the value chain.7
The profit-pool claim is holding, with an important qualification. Value has plainly drained out of structural assembly and pooled in payload technology and sole-source components β the margin structure, the lead times, and the bargaining behaviour all confirm it. What has not yet been demonstrated is the leg that matters most to a shareholder: that merchant manufacturers can convert this position into durable, audited operating margins on fixed-price work at volume. The evidence available in August 2026 is encouraging and incomplete. Backlog is converting at 38.5β44% a year, gross margins stand at 28% and 32% at the two leading merchant builders, and defence deliveries are tracking near their 24-month targets.235 But no full defence tranche has been delivered, flown, and reconciled, and aerospace history offers ample precedent for late-stage charges on exactly this kind of programme. On that specific leg, the belief is unproven rather than confirmed or fraying.
What the whole exercise illustrates is the distinction that governs thematic investing generally. A correct view about society or technology produces a poor outcome if it is expressed at the wrong layer of the value chain, in the wrong company, at the wrong capital structure, or at the wrong price. The forecast that satellites would proliferate was widely held by 2021 and was correct. Acting on it by owning bus assemblers still destroyed capital, because the tier being described was the tier losing profit share. The layer, not the trend, determined the outcome.
For a global thematic mandate, the analytical structure that follows is a barbell rather than a basket. At one end sit scaled merchant builders whose payload technology is proprietary and whose backlog is converting β with concentration and fixed-price execution as the specific things to monitor rather than the specific things to assume away. At the other end sit component owners whose qualification barriers give them pricing power regardless of which prime wins, with balance sheet strength as the limiting variable. Between them, diversified defence primes offer cash-flow resilience and diluted thematic exposure, which is a trade-off to be made consciously rather than discovered later. And the whole structure carries one correlated bet on a single appropriations cycle, which is worth counting honestly as one bet.
The industry's own history supplies the discipline. Iridium was engineered beautifully and financed into bankruptcy because the world was not ready. Terran Orbital had customers, a factory, and a backlog, and ran out of cash in the wrong tier. Both failed while being directionally right about where space was going.
The era of the artisanal satellite has ended, and the orbital assembly line has replaced it. Whether that assembly line earns its cost of capital is a separate question, and it will be answered in the next eight quarters by delivery dates, component lead times, and the margin on the first fully reconciled tranche β not by anything anyone says about the future of space.
Glossary
Bus (satellite platform). The structural and infrastructural body of a satellite: power, propulsion, thermal control, pointing, and flight computing. It is standardising into a catalogue product, which is why its margins are compressing toward 18β28%.
Payload. The mission equipment a customer actually buys β antennas, digital processors, sensors, laser terminals. Typically 50β70% of a satellite's bill-of-materials value and the location of most of its profit.
AESA (active electronically scanned array). An antenna of many small elements whose beams are steered electronically in microseconds. It allows one satellite to serve thousands of separate users on the ground while moving at orbital velocity.
Digital channelizer. The software-defined processor that digitises, routes, and reconstructs radio signals on board. It turns a satellite's capacity into something reallocatable after launch, and it is the single most defensible piece of communications payload technology.
OISL (optical inter-satellite link). A laser terminal connecting satellites to one another in orbit. It converts a set of independent satellites into a routing mesh and removes the need for continuous ground station visibility. Also the industry's sharpest supply bottleneck.
pLEO (proliferated low Earth orbit). An architecture using hundreds or thousands of satellites at 300β1,200 kilometres to achieve low latency, resilience through numbers, and continuous coverage. Its three-to-five-year replacement cycle is what makes satellite manufacturing a recurring business.
SWaP-C (size, weight, power, cost). The four-way engineering trade every payload design resolves. Power, and specifically the ability to radiate waste heat in vacuum, is usually the binding term for dense digital payloads.
COTS+ (screened commercial off-the-shelf). Automotive- or industrial-grade electronics screened for radiation tolerance and protected by software error correction, replacing military-specification radiation-hardened parts on short-life missions. The main reason payload electronics costs fell 70β90%.
AI&T (assembly, integration and testing). The cleanroom phase where a spacecraft is built up and subjected to vibration, acoustic, and thermal vacuum testing. Capital-intensive, schedule-risky, and the lowest-margin tier of the chain.
Percentage of completion. The accounting method by which manufacturers recognise revenue as work progresses against estimated total cost. It makes reported revenue sensitive to estimates, and it is why cost reassessments arrive as sudden charges.
Backlog conversion velocity. The share of opening twelve-month funded backlog turned into revenue over the next four quarters. The most reliable public test of whether an order book represents factory activity or paper.
Bringing into use (BIU). The International Telecommunication Union requirement to deploy a defined share of a licensed constellation β 10% in two years, 50% in five, 100% in seven β or forfeit the spectrum. It is why manufacturing cadence is a claim on a permanent global resource.
References
-
State of the Satellite Industry Report, 29th edition β Satellite Industry Association, May 13, 2026 ↩↩↩↩↩
-
MDA Space investor relations: Q2 2026 results, FY2026 guidance, backlog and Telesat Lightspeed contract disclosures β MDA Space Ltd., August 2026 ↩↩↩↩↩↩↩↩↩↩↩↩↩
-
Rocket Lab USA investor relations: Q2 2026 financial results, Space Systems segment and backlog disclosures β Rocket Lab USA, Inc., August 10, 2026 ↩↩↩↩↩↩↩↩↩
-
Global orbital launch and payload log, Q2 2026 β BryceTech ↩↩↩↩↩↩
-
Proliferated Warfighter Space Architecture programme materials, tranche solicitations and optical interoperability standards β U.S. Space Development Agency, 2025β2026 ↩↩↩↩↩↩↩↩↩↩↩
-
Airbus, Thales and Leonardo memorandum of understanding to combine space activities β Airbus newsroom, October 2025 ↩↩
-
Lockheed Martin completes acquisition of Terran Orbital β Lockheed Martin news, October 30, 2024 ↩↩↩↩↩
-
Shanghai Spacecom Satellite Technology closes 7 billion yuan Series B for the Qianfan/G60 constellation β SatNews, August 2026 ↩↩↩
-
Novaspace (Euroconsult), Satellite Manufacturing and Launch Markets 2026 β specialist subscription dataset covering manufacturing revenue pools, orbit mix, unit costs and payload price indices; no public canonical URL. ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩
-
Empor industry dossier, Space β Satellite Companies, August 24, 2026 β synthesis of company segment disclosures, defence procurement records, supplier lead-time surveys and specialist trade reporting. ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩
LinkedIn
WhatsApp
Reddit
Hacker News
Email
Kindle
RSS Feed
Spotify
Apple Podcasts
Amazon Music
Audible
YouTube