Two rocket companies walk into an antitrust review. They leave as a de facto monopoly. And somehow, the punchline may be that this was good for consumers, taxpayers, and maybe even competition.
A little context. In 2006, Boeing and Lockheed Martin combined their launch divisions into a joint venture called United Launch Alliance (ULA). That’s what I mean when I say they “combined.” It wasn’t technically a merger, and the distinction matters once you get deep into antitrust doctrine. For simplicity, though, think of it as a merger to monopoly. That sounds especially bad in an industry responsible for launching national-security payloads.
By every static measure, this is the textbook nightmare.
Readers know I think textbook metrics are often a bad starting point. So I was intrigued by a recent paper from Ruibing Su, Chenyu Yang, and Andrew Sweeting arguing that the deal was probably a good idea after all.
Their key insight—missing from the standard model—is that every launch teaches engineers something. The team flying its 50th mission knows things the team flying its first mission does not. Before the “merger,” that knowledge accumulation was duplicated across two separate programs, with separate engineering teams, supply chains, and production systems. That creates a clear possibility for efficiencies in a very specific sense—not just cost cutting, but faster learning.
The question is whether those efficiencies were large enough to outweigh the standard monopoly problem.
Su, Yang, and Sweeting argue yes. Using a structural model packed with all the usual modern industrial-organization bells and whistles, they analyze the space-launch industry from 1985 to 2024. Their conclusion: the learning synergies were real and large enough to offset the harms from increased market power.
They also find that when the government committed to multiyear block buys—instead of shopping for launches one mission at a time—costs fell dramatically. Forward-looking procurement gave the supplier stronger incentives to invest in improving future performance, even without a direct competitor applying pressure.
That’s a serious empirical result for one industry. The harder question is what regulators were supposed to do with that possibility at the time.
‘Too Speculative’ Usually Means ‘Too Hard’
Every merger in a learning-intensive industry raises the same questions. Will the merged firm keep innovating, or grow complacent? What happens to entry when the incumbent has a decade of accumulated know-how and the challenger starts from scratch? Will the dominant firm use its position to lock up critical inputs and foreclose rivals?
In practice, regulators mostly punt. Merger review tends to focus on static effects—prices, concentration levels, market shares—while treating “dynamic efficiencies” as something vague and speculative floating out in the ether. That bias consistently cuts in one direction: against mergers in the industries where dynamic effects may matter most.
To be fair, studying dynamic efficiency is hard.
Economists can build full dynamic models that try to capture learning, investment, and long-run competition. But even the people building those models openly acknowledge the tradeoffs. Ariel Pakes and Ulrich Doraszelski, in their handbook chapter on “Applied Dynamic Analysis in IO,” caution that the framework “delivers very little in the way of analytic results of applied interest.” Steven Berry and Giovanni Compiani, surveying the same literature, are not much more optimistic: “[T]he attempt to add dynamics may create enough compromises that the result is not better than the static model.”
So, yes, if the only way to incorporate dynamics into merger review is through a five-year structural-modeling project, regulators will mostly keep ignoring them.
You Can’t Buy Experience
Let’s work through some alternatives. Instead of going full structural industrial organization, let’s think in simple price-theory terms. Strip away the details for a moment. There’s a relatively straightforward way to capture many of the important dynamics here.
The key idea is that output today affects costs tomorrow. A firm that produces more accumulates something valuable over time.
That’s a reduced-form way to capture several different phenomena. The rocket example relies on learning by doing, but it’s hardly unique. Retailers benefit from denser distribution networks. Airlines build advantages through route density. Manufacturers refine tooling and supplier relationships through sustained production volumes in ways smaller rivals struggle to match. In platform markets, the equivalent is an installed base.
The engineering details differ. Don’t let that distract from the economics. In each case, output today builds a productive stock that lowers future costs.
Notice that this differs from standard capital accumulation. There, investment and production are separate decisions that compete for scarce resources. Cash spent on a new machine is cash not spent making products. Time spent building the next semiconductor fabrication plant is time not spent operating the current one.
The productive stock here works differently. You can’t simply write a check for a year of launch experience or a denser airline route map. The only way to build those assets is by producing. Output and investment therefore do not compete with each other. They are the same decision. Producing more today is investing more today.
To circle back to mergers specifically, this framework does not explain every dynamic-efficiency claim. Some dynamics involve patent races, product repositioning, entry timing, demand-side network effects, or strategic investments chosen separately from output. Those may require different tools.
Most merger-efficiency claims, though, are more mundane. They involve scale, experience, density, know-how, or installed bases. Take T-Mobile and Sprint in 2019. The companies argued that combining their spectrum holdings and cell sites would allow them to build a higher-quality 5G network than either could alone. That’s fundamentally a network-density claim. The “stock” is network capacity, and producing more output helps build it.
Competition Falls. Production Might Rise.
A merger involving this kind of productive stock pulls in two directions at once.
Start with the familiar concern: less competition. After a merger, the combined firm no longer worries as much about losing customers to its closest rival because, in a sense, it owns the rival too. Before common control, if Boeing’s launch division raised prices, some customers would switch to Lockheed Martin. After common control, many of those “lost” sales stay inside the same organization. The merged firm recaptures business it used to lose.
That weakens the incentive to fight for the marginal customer. The predictable result is higher markups and less output.
You can think of this graphically. Common control rotates the marginal-benefit curve inward. Before the merger, winning a launch contract from your rival is a real gain. Afterward, winning business from another division of your own company is partly just stealing from yourself. The merged firm internalizes cross-product diversion, so the marginal benefit of expanding output falls. Quantity falls with it.
That’s the standard merger story.
But this framework introduces a competing force. If producing today builds productive capability for tomorrow, then every additional launch also generates experience, operational knowledge, and learning by doing. Those gains lower the effective marginal cost of future production.
Push that effect far enough, and the supply curve can actually slope downward. The firm still pays the immediate cost of the launch, but it is also effectively purchasing future cost reductions.
Under that logic, consolidation can increase the incentive to produce. A combined firm captures more of the returns from building productive capability, so producing today becomes more valuable. Larger production volumes accelerate learning. Greater internal coordination makes it more likely the resulting efficiencies will actually materialize, instead of being duplicated across separate organizations and supply chains.
Where the Textbook Graph Starts Misbehaving
To see both forces in a single picture, we need to think about the marginal cost of building this productive stock. How does learning change the firm’s marginal-cost curve? What’s the “price” of experience?
Physical capital has a rental rate—the amount you would pay each period to use it. Dale Jorgenson called this the “user cost of capital”: the implicit rental value of an asset the firm owns rather than leases. Experience and operating capability have a similar structure, even if nobody literally sends the firm a bill.
What’s the value of owning that stock? Each unit of output today adds to it and lowers future costs. The present value of those future savings is effectively the rental value of the stock. Producing one additional launch is therefore cheaper than the accounting cost suggests, because part of the expenditure is really purchasing future productivity improvements.
Call the normal accounting cost per launch the static marginal cost. Call the static cost minus this implicit “rental rebate” the dynamic marginal cost, net of rental. Static marginal cost is what shows up on the invoice. Dynamic marginal cost is what actually drives the firm’s production decision once it accounts for learning effects.
Pre-merger, in the top panel, the firm faces residual demand for its own product and a standard single-product marginal-revenue curve. Static marginal cost sits at 3. Dynamic marginal cost, net of the rental rebate, sits closer to 2. The firm produces where marginal revenue intersects dynamic marginal cost, and we can read off price and quantity from there.
Post-merger, two things change simultaneously.
First, marginal revenue rotates inward to “owned-product marginal revenue.” This is the standard diversion effect in picture form. If that were the only change, the merger would mechanically reduce output and raise prices.
Second, dynamic marginal cost rotates further downward to “post-merger dynamic marginal cost.” The combined firm captures more of the value from producing today. There is less duplication across engineering organizations. Higher combined output pushes the firm farther up the learning curve—or, equivalently, farther down the cost curve. More of the future cost reduction stays inside the firm instead of leaking to a competitor. The rental rebate grows larger, so the firm’s perceived marginal cost falls.
In the figure, the dynamic-marginal-cost shift dominates.
The picture also clarifies when the result flips. If the learning curve is relatively flat, dynamic marginal cost barely moves, and the marginal-revenue rotation wins. Quantity falls, prices rise…cats and dogs living together, mass hysteria.
In one sense, this is all almost tautological. If the force pushing quantity upward exceeds the force pushing it downward, output rises. Not exactly a profound insight. The value of the framework is that it helps us think systematically about when each force dominates.
That means thinking more carefully about how learning actually works. If experience spills over heavily to competitors, a merger does little to change how much of the rental rebate the firm captures. Dynamic marginal cost barely moves. The industry may already exhibit a downward-sloping cost structure even if no individual firm fully internalizes it. The outcome then depends on how steep the learning curve is, how private the learning remains, and how durable the resulting advantage proves to be. The theory helps organize the investigation.
It also pushes us to think about alternatives. A merger is one way to increase the returns from producing today, but it is hardly the only one.
Su, Yang, and Sweeting find that costs fell sharply when the government shifted from buying launches individually to committing to multiyear block purchases. That policy effectively guaranteed suppliers a stream of future orders. The mechanism is the same one illustrated in the figure. A larger committed order book increases the value of investing in learning today because the firm expects to produce the future launches that benefit from those improvements. The rental rebate gets larger. Dynamic marginal cost rotates downward.
Not Every Market Gets a SpaceX
The figure shows when a merger is most likely to generate genuine cost savings: when learning curves are steep, experience remains private, and the productive stock is durable.
We should be careful, though. Those same conditions also make entry harder. A new entrant starts with zero accumulated experience, while the merged firm sits on years of operational knowledge. If learning is steep, the cost gap will be large. If learning is private, entrants cannot easily catch up by poaching engineers. If the stock is durable, the advantage persists.
It is tempting to treat those entry barriers as the offsetting “cost” of the efficiency and simply net the two effects against each other. I’ve argued before that this gets the analysis backwards. Achieving scale is not an antitrust harm. Preventing rivals from achieving scale through better products or lower prices is not an antitrust harm either. That is what competition on the merits looks like.
The same logic applies to accumulated learning. If ULA’s experience made it harder for entrants to win contracts, that productive stock was doing exactly what productive stocks are supposed to do. The merger accelerated the accumulation of that stock by combining output. It did not do so by sabotaging competitors.
The actual antitrust harm must come from some separate exclusionary mechanism. The restraint is the problem, not the stock itself. Foreclosing key inputs. Locking up distribution. Raising rivals’ costs through means unrelated to the merged firm’s own productivity.
As it happens, ex post, entry turned out to be possible. SpaceX arrived and detonated the market structure so completely that much of this now feels almost quaint. Not every market gets a SpaceX, though.
Still, we have tools for thinking carefully about these problems. This kind of simplified framework does not replace a full structural model, but it gives you a way to reason through the question before building one—and a way to understand what the model is actually doing once you have.
Most importantly, it forces specificity. Which curves are shifting? What mechanism is moving them? Which effects are observable ex ante, and which only become visible after the fact?
If we can answer those questions clearly, we are already a long way toward understanding the market.
The post In Space, No One Can See Your HHI appeared first on Truth on the Market.
Nicole Byers is an entertainment enthusiast! Nicole is an entertainment journalist for the Maple Grove Report.
