Starship, Starlink, and some SpaceX-conomics

The economics of Starship are subtler than you might think

Starship/Super Heavy is an awesome rocket in the original sense of the word. The Boca Chica test got a lot of coverage given the explosion and the lack of environmental precautions, but it would have gotten a lot of coverage regardless.

Starship/Super Heavy taking off from Boca Chica in April 2023. Source.

I won’t rehash the engineering details here. You can find plenty of writeups online and I have nothing to add on that. No, I want to focus on the economics of Starship. Specifically, how it relates to Starlink and SpaceX’s increasingly prominent position in the launch market.

There are many misconceptions about the economics of Starship and Starlink. Even as people recognize that SpaceX is “vertically integrated”—they own rockets and satellites—and that SpaceX is facing less and less competition in launch markets, it seems almost no one has really thought through what these facts mean for Starship’s economics. There are two major errors I see people make when thinking about this:

1. Assuming that Starship slots will be priced at their average total cost (ATC, which is often conflated with its “marginal cost”, MC); and

2. Neglecting the effect of opportunity cost (i.e. what SpaceX gives up) on Starship slot pricing.

Casey Handmer’s old post about why Starship will be a big deal is a great case in point. I don’t mean to pick on Handmer but the post commits both of these errors in a big way. I’ll go through these errors separately, then together at the end with some rough order-of-magnitude numbers. There’s some calculus underlying my statements; I’ve kicked that to the appendix in case you’re interested.

At the outset I also want to list the three key modeling assumptions that will drive just about everything I say here¹:

1. SpaceX is a vertically-integrated firm, owning both rockets and satellites. The satellites are part of a (presumably) lucrative constellation, Starlink.

2. SpaceX will effectively be a monopolist in the launch sector and they won’t be constrained by their own launch capacity.

3. SpaceX will try to maximize profits.

1 is just an observable fact plus a common belief about Starlink profitability. 2 seems to be a widely-held belief (see Handmer’s piece). 3 is often not stated, but is consistent with the belief (again, widely held and reinforced by Musk) that SpaceX is trying to accumulate resources and capabilities to go to Mars. 1 and 2 are more or less baked into Handmer’s post; 3 is one interpretation of the Mars plan, which Handmer explicitly mentions.

Price is not cost; not all markets are perfectly competitive

The first thing to get out of the way is simple: price is not cost.² Introductory economics textbooks will often present a nice simple supply-and-demand diagram and say something like “in a perfectly competitive market, the supply curve reflects the marginal cost of producing a good, and the price of the good has to be equal to its marginal cost”.³ If you attend enough of the class, somewhere along the way you’ll see this generalized slightly: in the long run, the marginal cost of a good is also its average total cost, and competitive forces equate the price with the average total cost.⁴

Cost curves for a perfectly competitive firm from an intro microeconomics class. MC is marginal cost, ATC is average total cost, MR is marginal revenue. Source: Wikipedia.

This is a very useful result because actual marginal costs are notoriously hard to measure. What we more often see from accounting statements are total costs, sometimes with a fixed/variable breakdown, but it’s rare to see the costs attributed to incremental units of production in a way that lets one estimate the true marginal cost of the good. So if the market is competitive enough, you can use the “ATC = MC” result to estimate marginal costs by just taking the total cost and dividing by the quantity produced (ATC = TC/Q). Boom, simple, done.

But this assumes the market is (close to) perfectly competitive, meaning the firm is earning (close to) zero profits. That may be a good description of the market for, I dunno, some agricultural commodities. Launch markets—with high fixed costs of rocket development (billions this year alone for Starship) and few competitors (two companies, SpaceX and Rocket Lab, account for most US launches)—are not perfectly competitive.

The dude. You know. Abiding.

This isn’t just my opinion (1, 2). Again, Handmer makes a nod to this in his post but doesn’t really internalize what this means for Starship pricing.⁵ In imperfectly competitive markets, the “ATC=MC=p” rule breaks down. ATC isn’t the same as MC, and firms charge markups (p > MC). They earn positive profits. Handmer makes a nod to this in his post (“If flights were sold at the going rate, Starship could launch any payload currently in the market, while making money.”), but doesn’t seem to take this to its logical conclusion. We’ll get there. For now, let’s just acknowledge that the ATC of a Starship slot isn’t going to be the price; the price will be higher.

Opportunity costs are real

The next piece to acknowledge is that there are opportunity costs from launching a Starlink satellite rather than an external customer’s payloads, and vice versa. Since slots are rival (i.e. a slot can only hold one thing), each slot allocated toward Starlink satellites means giving up the revenues from one commercial payload, and each commercial payload means delaying or giving up the revenues from one Starlink satellite.

Opportunity cost: when you choose to party instead of study for your econ exam, you’re about to learn what this means.

Opportunity costs are tricky to reason about, and trickier to measure. We don’t get to see the folks who would have bought a marginal Starship slot but didn’t because it wasn’t available, and we don’t get to see SpaceX’s internal calculations on how much revenue one more Starlink satellite would bring if launched now rather than later (or not at all). Making matters even more complicated, these opportunity costs should be discounted by the time value of money facing SpaceX—and outside observers like us certainly don’t observe the internal rate of return that SpaceX faces.⁶ But with some math and guesstimation, we can figure out how opportunity costs connect to the marginal cost of Starship slots and ballpark the numbers.

Putting it all together

Ok, so price isn’t cost and opportunity costs are real. Given our three assumptions, what does this mean for the price and marginal cost of Starship slots?

The fundamental result here can be boiled down to two statements:

1. The price of a Starship slot is going to be the marginal cost plus a markup. If SpaceX is a monopolist, the markup will be (proportional to) the inverse of the price elasticity of slot demand.

2. The marginal cost of a Starship slot is going to be exactly the net present value (NPV) of profits from a marginal Starlink satellite. This is the opportunity cost of allocating the slot to a non-Starlink payload.

Result 1 is pretty standard, nothing new here. Handmer might even say “duh, I said their costs would go super low, of course they could charge more!” But the specific form of the pricing rule is informative. The less elastic the demand for launch capacity is, the higher the price will be.⁷ So if you think that cheap launch will unlock hitherto-unseen space development, that Starship will “[obliterate] the mass constraint and every last vestige of cultural baggage that constraint has gouged into the minds of spacecraft designers”, then sure, you should think SpaceX-as-monopolist will price Starship close to marginal cost. But it doesn’t work the other way—SpaceX can’t guarantee high launch demand just by dropping the price. The (often-unstated) demand elasticity assumption (and it is, for now, an assumption) comes first.

Next, marginal costs. The equality between the NPV of marginal Starlink satellite profits and the marginal cost of Starship slots means that the more profitable Starlink satellites get, the more expensive Starship slots will get. Conversely, if you think that Starship slots are going to be cheap because marginal costs will be cheap, you must also believe that Starlink satellites will not be that lucrative.

Why? Opportunity costs. Every single Starlink satellite not launched on a Starship is forgone profits; they may only be pushed out into the future, but that’s still a loss. Think of it like this. If the marginal costs were lower than the Starlink profits, SpaceX could make money by spinning up more Starships (even as scarcity drives MC higher) and make money. If marginal costs exceeded Starlink profits, SpaceX could make money by retiring Starships (relaxing scarcity to lower the MC). So not equating the NPV of marginal Starlink satellite profits with the marginal cost of Starship slots means SpaceX is, one way or another, leaving money on the table.

Numbers numbers numbers

Ok, let’s put some numbers to this mess. We can’t know the elasticity of demand yet—it’s one of the big unknowns here—so we can’t say much about pricing directly. But we can use the second condition, linking Starship marginal costs to Starlink marginal revenues, to get a sense of how cheap Starship slots might get. In his 2019 blog post, Handmer gives us a specific prediction to work with: “the marginal cost per [Starship] flight could fall to $5M or below”. Does this make sense?

Payload Space estimates Starlink’s 2023 revenues at around $3.7B. Let’s call that $4B for 2023 to err on the side of “Starlink is fantastically lucrative”. Starlink currently has about 3,500 satellites in orbit. Let’s suppose by the end of 2023 Starlink has 4,500 satellites in orbit, again consistent with Payload’s assumptions. That gives a 2023 revenue per satellite per year of around $888,889. If you take Musk’s projections at face value (ok baka) and assume $30B in total revenues with 12,000 satellites in orbit we get about revenues of $2.5M per satellite per year. Let’s say $2.5M at the high end and and $750K at the low end. Starlink satellites have ~5 year lifespans, and let’s assume a 5% discount rate.⁸

Starlink satellite revenue estimates with numbers above. Spreadsheet here—play with your own copy, let me know if you see a mistake.

Osoro and Oughton (2021) estimate the marginal cost of one Starlink satellite—including manufacture, spectrum, and other capex/opex, but excluding launch—on the order of $0.5M/satellite. (We exclude launch here because, as shown in the appendix, those costs are accounted for separately in the optimization problem.) So we’re looking at something like $3-11M in operating profits (“cashflow” may be a better term) per Starlink satellite.

Table 2 from Osoro and Oughton (2021). The capex costs and staff/R&D/marketing costs are fixed costs with respect to satellites, so become negligible as the constellation grows.

Each Falcon 9 can carry 60 Starlink satellites, and Starship can surely carry more—TechCrunch is estimating around 400 Starlinks/flight here. But look, even at 60 Starlinks/flight and low revenue estimates, the marginal cost of a Starship launch is going to be driven up well past $5M/flight.

“Oh but SpaceX faces a much higher cost of capital because they’re venture funded, so they must use a higher discount rate, you’re being unreasonable!” Ok, sure. Crank the discount rate up to 50%. Hell, crank it higher. Even at a 100% discount rate the low-end revenues/satellite/year are on the order of $1.4M. $0.9M per slot is pretty cheap by today’s standards—around $54M/flight.

Again, $5M/flight doesn’t make any sense. Could the average total cost fall that low? Sure. But as we’ve discussed, absent competitive forces that make it so, ATC is not MC. Handmer’s assumptions/beliefs are internally inconsistent. I’m not trying to pick on Handmer specifically here—if anything, I’m grateful that he spelled out his assumptions and logic so clearly. It’s more than most folks do. But that thinking seems to be shared by many in the space world—here’s a random screenshot of someone expressing similar thoughts on Twitter:

A random Twitter user confidently expressing beliefs that Starship+Starlink will make SpaceX an unbeatable space company.

Anyway, look, you can’t believe all three:

1. Musk’s projections about Starlink and the general idea that Starlink is/will be crazy lucrative;

2. SpaceX is a profit maximizer that will soon be effectively a monopolist in the launch sector (or will generally have significant market power)⁹;

3. The MC of a Starship flight (not slot) will be ~$5M.

At least one of these has gotta go.

---

---

Appendix: the integrated monopolist’s problem

Consider a monopoly space launch provider who also operates their own satellites. The firm is not constrained in the amount of rocket slots they can provide. Each rocket slot sells for P, with market demand for launch slots given by Q(P) where Q' <= 0. The firms own satellites are denoted Q_o and each provides operating profit (i.e. cashflow net of operating costs and capex other than launch) of \pi. The total cost of a rocket slot is C(Q_o + Q(P)) with C' >0, C'' > 0.

\(\max_{P, Q_o} ~~ PQ(P) - C(Q_o + Q(P)) + \pi Q_o\)

Taking first-order conditions, we get

\(P^* : P^* = C'(Q_o^* + Q(P^*)) - \frac{Q(P^*)}{Q'(P^*)}\)

\(Q_o^* : \pi = C'(Q_o^* + Q(P^*))\)

which are exactly the math behind the two statements of the fundamental result.

1. You can disagree with these assumptions, and I’ll try to sketch out how violations of those assumptions might affect things, but at least I’ll have been explicit about my economic modeling setup.

2. This is one of my pet peeves with economics commentary on the space sector. Too many folks conflate these concepts. They are not the same!

3. Marginal cost” just means “the cost of producing one more unit”. “Incremental cost” is a synonym. It’s surprisingly a bit subtle to measure. “Unit cost” is not always the same, since it may not refer to the cost of an incremental unit of production. “Variable cost” describes the things that go into the cost, like labor or lighting, and doesn’t attribute those costs to incremental units of output production.

4. Marginal costs are eventually increasing in the number of units produced as various forces of scarcity come into play: the plants sustain wear and tear, there are only so many hours in the day and overtime rates kick in, materials get exhausted or more expensive, etc. When there are fixed costs, average total costs first decline; when marginal costs are increasing, the ATC is also eventually increasing. Most posts about Starship (like Handmer’s) focus on the declining portion of ATC while ignoring the increasing portion.

5. “SpaceX believes (correctly) that there is almost infinite upside to flying satellites to meet global telecommunications demand, provided you have a monopoly on launches that are orders of magnitude cheaper than the competition.”

6. “Discounting” and “time value of money” just mean that a dollar today is worth more than a dollar tomorrow. It’s a pretty standard finance/economics concept—wiki for the general concept, RFF for applications to climate change.

7. For a good time, ask a procurement officer how much they’re actually paying for SpaceX launch slots. Yes yes, it’s cheaper than ULA. But is it anything close to marginal cost? Watch their reaction closely.

8. 5% is a very low internal rate of return for venture capital projects in the US, but it’s in the range of weighted-average cost of capital estimates for aerospace/defense and wireless telecom services. See cost of capital estimates here for a range of WACCs estimated for different regions and sectors. Starlink probably has a much higher IRR given projections made to investors—I wouldn’t be surprised if it’s on the order of 50%.

9. If SpaceX is only a launch duopolist or one of 3-4, the pricing rule will be closer to marginal cost (the markup will be lower than inverse elasticity), but the marginal cost will still have to be equated to the forgone revenues from a Starlink satellite. The key issue here is vertical integration, not monopoly.