Chapter 38: SpaceX and the Reusability Revolution

"If things are not failing, you are not innovating enough." — widely attributed to Elon Musk

Overview

For sixty years, every orbital rocket was a match: you struck it once and threw it away. The most sophisticated machines humanity had ever built — engines that out-power a battleship, tanks welded to the thickness of a coin, guidance that hits a moving target thousands of kilometers away — were used a single time and dropped into the ocean. We noted this quiet scandal back in Chapter 22: the propellant in a Falcon 9 costs a fraction of a percent of the launch, yet the launch cost a fortune, because almost all of it was hardware built once and discarded. The Space Shuttle tried to change this and, as we saw in Chapter 37, half-succeeded and half-failed — reusable in principle, but so laborious to refurbish that it never delivered the cheap access it promised. This chapter is the story of the company that finished the job, and of the two vehicles — Falcon 9 and Starship — that are the climax of every anchor thread this book has been pulling since Chapter 3.

The story is not really about one company. It is about an idea, and the idea is the fifth recurring theme of this book: reusability is changing everything. Reaching orbit has always been governed by the tyranny of the rocket equation — the exponential from Chapter 3 that makes rockets ninety percent propellant and leaves almost no room for payload. Reusability does not repeal that tyranny. A recovered booster still obeys $\Delta v = v_e \ln(m_0/m_f)$; it simply pays a slice of its precious mass ratio to buy back the whole vehicle. What changes is not the physics but the economics, and when the economics of launch fall by a factor of ten or twenty, the map of what is affordable in space redraws itself. That is the revolution, and this chapter is where we finally do its arithmetic.

We will move the way SpaceX did: start small and iterate. First the development philosophy — why "build, fly, fail, fix, repeat" beat "design for a decade, fly once perfectly." Then the evolution of Falcon 9 across its block upgrades, the engineered choreography of a booster flying itself home, and the cost model that decides whether any of it was worth doing. Finally we turn to the vehicle that takes the idea to its conclusion — Starship, fully reusable, stainless-steel, methane-fueled, caught out of the air by its own launch tower — and read every one of those strange choices as a consequence of physics and economics you already understand. We close inside the Raptor engine, the first full-flow staged-combustion engine ever to fly, where the propulsion of Part III and the reuse of this chapter meet.

In this chapter, you will learn to:

• Explain why iterating fast on cheap, expendable prototypes beats perfecting a vehicle you cannot afford to lose — and how that method connects to the reliability engineering of Chapter 32.
• Follow Falcon 9 from v1.0 to Block 5 and say what each block bought.
• Compute the payload a first stage gives up to come home, and why the trade still pays.
• Put real numbers on the cost of launch, and see where reuse stops helping and full reuse must begin.
• Trace Starship's steel, methane, catch, and refueling back to the rocket equation, thermodynamics, and the balance sheet — the systems-engineering case study the whole book has pointed toward.

Learning Paths

🚀 Space Enthusiast: Read 38.1 for the philosophy that changed the industry, then spend your time on 38.3 (how a booster lands) and 38.5 (Starship). The payload-cost worked example in 38.3 and the "why steel, why methane, why catch" walk-through in 38.5 are the two things that will change how you watch a launch. You can skim the cost algebra in 38.4 and keep the punchline.

📐 Engineering Student: This is a synthesis chapter — it uses tools from Chapters 3, 17, 22, and 23 all at once. Work the reuse-payload penalty in 38.3 and the cost model in 38.4 yourself, and be able to justify each Starship choice in 38.5 from first principles. The ⭐⭐/⭐⭐⭐ exercises tie straight back to your staging and propulsion work.

🎮 KSP Player: You have flown every idea here. Recovering a booster is the reserve-fuel-versus-payload trade you feel every time you try to land a first stage and still make orbit (38.3); Starship's belly-flop is a real re-entry profile you can attempt. Focus on 38.3, 38.5, and the Raptor cluster in 38.6 (engine-out, just like your many-engine designs).

🛰️ Industry Prep: 38.4 (cost per kilogram, amortization, the price-versus-cost distinction) is the commercial heart of modern launch, and the Mission Design Checkpoint asks you to redo your own mission's economics under reusable launch. Section 38.5's systems-engineering reasoning is exactly the kind of trade study a program defends at review (Chapter 29).

38.1 The approach: start simple, iterate fast

Before any hardware, understand the method, because the method is the innovation. Traditional launch- vehicle development — the way NASA built the Shuttle and the way most of the aerospace industry still works — treats a rocket like a cathedral. You design it exhaustively on paper, analyze every load case, review every drawing, and build the flight vehicle essentially once, because each one costs hundreds of millions of dollars and years of schedule. You cannot afford to lose one, so you test sparingly and fly conservatively. The result is superbly analyzed hardware that is also slow, expensive, and — crucially — almost impossible to improve, because every change ripples back through the whole paper edifice.

SpaceX inverted this. Its method — sometimes called iterative or hardware-rich development — treats a rocket like software: build a version, fly it, watch it break, learn exactly why, fix that, and fly the next version soon after. Failures are not disasters to be avoided at all costs; they are the fastest way to buy information, provided each failure is cheap and you actually learn from it. This is the same philosophy from a different angle than Chapter 32's "test like you fly" — not a contradiction of rigorous testing but a bet about where to spend the testing: on real flights of real hardware, early and often, rather than on ever-deeper analysis of a vehicle that flies late and rarely.

📜 From History: learning on Falcon 1. SpaceX did not begin with a giant rocket. It began with Falcon 1, a small two-stage kerolox vehicle with a single Merlin engine — the cheapest orbital rocket the company could build, precisely so that losing one would not end the company. It needed to lose several. Falcon 1's first flight (2006) failed seconds after liftoff (a corroded fuel-line nut). The second (2007) reached space but lost control late in the second-stage burn when unchecked propellant slosh coupled into the guidance — the very demon we named in Chapter 22. The third (2008) failed at stage separation. Only the fourth flight, in September 2008, reached orbit — becoming the first privately developed liquid-fueled rocket to do so, with the company reportedly weeks from bankruptcy. Three failures were the tuition. The lesson SpaceX drew was not "test more before flying" but "fly sooner, on hardware cheap enough to lose, and instrument it so every failure teaches."

The contrast with the Shuttle is exact and instructive. The Shuttle carried astronauts on its very first orbital flight, in 1981, with no uncrewed test flights at all — it could not iterate, because it was too expensive to fly without a crew and too complex to fly without a purpose. Every Shuttle was essentially hand-built and irreplaceable; the program could not casually lose one to learn from it, and twice it lost one anyway, at the cost of fourteen lives. The iterative method is, among other things, an argument that it is safer in the long run to break cheap, uncrewed vehicles early than to stake everything on a vehicle too precious to test to destruction.

💡 Intuition: the learning rate is what compounds. Imagine two teams building a hard machine. Team A ships a new version every two months and learns something real from each; Team B ships once every five years. After a decade, Team A has iterated sixty times and Team B twice. Even if each of Team A's steps is smaller, sixty compounding improvements crush two. Reusability turbo-charges this: a recovered vehicle is not just cheaper to fly again, it is a returned test article you can inspect for wear and feed straight into the next design. Reuse and iteration reinforce each other — every landed booster is data.

There is a subtlety worth stating so you do not misread the philosophy as recklessness. Iterating fast works only when three conditions hold: each failure is inexpensive (small or mass-produced hardware), each failure is survivable (uncrewed, and safely contained), and each failure is diagnosable (heavily instrumented, so you learn the true cause rather than guessing). Remove any one and "fly and fail" becomes just "fail." SpaceX flew crew only after Falcon 9 had flown dozens of times and its failure modes were understood — the iteration happened on cargo and satellites, and the reliability it bought was then spent on people. Fast iteration and hard-won reliability are not opposites; the first is how you earn the second.

🔄 Check Your Understanding 1. Why did SpaceX begin with the small Falcon 1 rather than going straight to an orbital-class booster? 2. Give the three conditions under which "fly, fail, fix, repeat" is a virtue rather than recklessness.

Answers

1. A small, cheap vehicle can be lost without ending the company, so the team can afford to fly early, fail, and iterate — buying information at low cost. A large vehicle is too expensive to lose, which forces the slow, analysis-heavy, fly-rarely method the iterative approach is trying to escape. 2. Each failure must be inexpensive (cheap or mass-produced hardware), survivable (uncrewed and safely contained, so no one is hurt), and diagnosable (well instrumented, so you learn the real root cause). Remove any one and iteration degenerates into mere failure.

38.2 Falcon 9 evolution: v1.0 to Block 5

The rocket that carried the whole revolution is the one we have been computing since Chapter 3. It is time to define it properly and watch it grow.

Definition (Falcon 9). Falcon 9 is SpaceX's partially reusable, two-stage, kerolox launch vehicle: a first stage powered by nine Merlin engines (the "9" in the name) whose booster is recovered by propulsive landing, and an expendable second stage powered by a single vacuum-optimized Merlin. It is the first orbital-class rocket to fly its first stage repeatedly, and by the mid-2020s it had become the most-flown and most reliable launch vehicle in operation. Its payload is roughly 17 t to low Earth orbit recovering the booster, up to about 22.8 t expended (Tier 2, version- and mission-dependent).

Falcon 9 is not one rocket but a family that iterated in flight, and the vehicle that lands boosters today is substantially different from the one that first flew in 2010 — same silhouette, deeply changed machine. The instrument of that change is the block upgrade.

Definition (block upgrade). A block upgrade (or block) is a discrete, numbered revision of a vehicle that bundles a set of design improvements — more thrust, denser propellant, tougher structure, better reuse hardware — into a new production standard while keeping the vehicle externally and operationally similar. Blocks are how an iteratively developed vehicle captures its accumulated lessons without a clean-sheet redesign: each block is the current "version number" of the rocket.

Trace the evolution, and you are watching the iterative method of 38.1 applied to a single vehicle across a decade (dates and figures Tier 2, widely reported):

Two of these steps deserve a closer look, because each solved a problem the rocket equation created.

Densified propellant (Full Thrust). Reuse costs mass ratio — we will make this precise in 38.3 — so where do you find the mass ratio to pay for it? One answer is chemistry, not structure. SpaceX chills its liquid oxygen well below its boiling point (and its kerosene too), making both denser. Denser propellant means more mass of it fits in the same tank volume, which raises the stage's mass ratio for free — no heavier tanks, just colder fluid.

Worked Example: buying back delta-v by chilling the propellant. Take the Chapter 3 Falcon 9 first stage: $395{,}700\ \text{kg}$ of propellant, dry mass $25{,}600\ \text{kg}$, exhaust velocity $v_e \approx 2{,}840\ \text{m/s}$, carrying an upper stack (fueled second stage + a $15\ \text{t}$ payload) of $121{,}000\ \text{kg}$. Now subcool the propellant so about $8\%$ more mass fits the same tanks — roughly $427{,}000\ \text{kg}$ (Tier 3, illustrative). The wet mass rises to $$m_{0,1} = 15{,}000 + 96{,}000 + 25{,}600 + 427{,}000 = 563{,}600\ \text{kg},$$ while the burnout mass $m_{f,1} = 136{,}600\ \text{kg}$ is unchanged. The first-stage delta-v becomes $$\Delta v_1 = 2{,}840 \times \ln\!\left(\frac{563{,}600}{136{,}600}\right) = 2{,}840 \times \ln(4.13) > = 2{,}840 \times 1.418 = 4{,}027\ \text{m/s},$$ up from the $3{,}862\ \text{m/s}$ of Chapter 3's nominal stage — a gain of about $165\ \text{m/s}$, won purely by making the oxygen colder. As we will see in 38.3, recovering the booster costs several hundred m/s; densification quietly hands a chunk of it back. (Real Full Thrust also uprated the engines and stretched the stage; densification was one lever of several.)

Block 5 (reuse as a design goal, not a stunt). The earlier blocks could land a booster; Block 5 was built so landing was routine and cheap to turn around. Its improvements read like a checklist against the Shuttle's failures (Chapter 37): a heat-shielded base and titanium grid fins that survive re-entry without replacement; retractable (not jettisoned) legs; engines qualified to fly repeatedly without teardown; and a general campaign to eliminate the hands-on refurbishment that made Shuttle reuse so ruinously expensive. The design target — a booster flying ten times with minimal work between flights, and many more with periodic overhaul — is the difference between "technically reusable" and "actually cheaper," and it is the hinge on which 38.4's economics turn.

🔗 Connection: the octaweb and engine-out. The v1.1 "octaweb" — eight Merlins in a ring around a ninth — was not just a packaging choice. Nine engines give Falcon 9 the engine-out capability we analyzed in Chapter 22: it can lose an engine and still reach orbit, as it did on the CRS-1 mission in 2012. Many small, mass-produced engines simultaneously buy engine-out redundancy, a single high-volume production line (cheap engines), and the deep-throttle-on-a-few-engines needed to land — one design decision serving reliability, cost, and reuse at once. Hold that idea; Starship's thirty-three engines take it to the limit.

🔄 Check Your Understanding 1. In what sense is Falcon 9 "one rocket," and in what sense is it many? 2. Densifying propellant raises a stage's mass ratio. Why does that specifically help a reusable rocket more than an expendable one?

Answers

1. It keeps one silhouette, name, and operational concept, but across block upgrades (v1.0 → v1.1 → Full Thrust → Block 5) its engines, propellant density, structure, and reuse hardware changed substantially — it is a family that iterated in flight, not a frozen design. 2. A reusable rocket must spend mass ratio on recovery (reserved landing propellant, legs, grid fins), which eats into the payload it can deliver; densification adds mass ratio back without adding tank structure, helping offset the reuse penalty. An expendable rocket has no recovery penalty to offset, so the same gain simply adds a little payload rather than paying a debt.

38.3 Propulsive landing engineered

We defined propulsive landing in Chapter 22 — decelerating and landing a stage vertically on its own engines — and walked the Falcon 9 booster's four-act return: separation, boostback burn, entry burn, and the final grid-fin descent and landing burn. We even proved the booster cannot hover (its lightest possible thrust exceeds its empty weight, forcing the "hoverslam"). This section does not redefine any of that; it engineers it, by answering the question Chapter 22 set up and left for the climax: exactly what does coming home cost, and why is the trade worth it?

The cost is paid in the only currency the rocket equation recognizes — delta-v. Every one of those return burns consumes propellant, and propellant reserved for landing is propellant not spent accelerating the stack toward orbit. A recovered booster therefore burns out slower than an expended one, stages lower and earlier, and hands its upper stage a harder job. Since the upper stage is fixed, the only variable left to give is payload. Let us put a number on it, using the very Falcon 9 we have carried since Chapter 3.

Worked Example: the delta-v and payload cost of coming home.

Recall the Chapter 3 result: expended, this Falcon 9 with a $15\ \text{t}$ payload produces about $\Delta v_1 + \Delta v_2 = 3{,}862 + 6{,}019 = 9{,}881\ \text{m/s}$ — comfortably above the $\sim 9{,}400\ \text{m/s}$ that ascent to LEO demands (orbital speed plus the gravity and drag losses of Chapter 4).

Now recover the booster. It must reserve propellant for the boostback, entry, and landing burns — take a droneship recovery reserving about $10\%$ of the first stage's propellant, roughly $39{,}600\ \text{kg}$ (Tier 3, illustrative). That reserved propellant is carried up but not burned on ascent, so it stays with the stage and becomes part of its burnout mass. The first stage's final mass rises from $136{,}600\ \text{kg}$ to $$m_{f,1}' = 136{,}600 + 39{,}600 = 176{,}200\ \text{kg},$$ and its ascent delta-v falls to $$\Delta v_1' = 2{,}840 \times \ln\!\left(\frac{532{,}300}{176{,}200}\right) = 2{,}840 \times \ln(3.021) > = 2{,}840 \times 1.106 = 3{,}140\ \text{m/s}.$$ The booster gave up $3{,}862 - 3{,}140 = \mathbf{722\ \text{m/s}}$ of ascent delta-v to keep enough fuel to fly home. The second stage is unchanged ($\Delta v_2 = 6{,}019\ \text{m/s}$), so the new total is $$\Delta v_{\text{total}}' = 3{,}140 + 6{,}019 = 9{,}159\ \text{m/s}.$$ That is now below the $\sim 9{,}400\ \text{m/s}$ needed for orbit. The same rocket that reached orbit with $15\ \text{t}$ expended cannot deliver $15\ \text{t}$ and still come home. To close the $\sim 240\ \text{m/s}$ gap you must offload payload, trading mass off the top of the stack until the total climbs back to $9{,}400$.

Sanity check against reality. Recovering the booster on a droneship drops Falcon 9's LEO payload from roughly $22.8\ \text{t}$ (expended) to about $17\ \text{t}$ (recovered) — a cut of about $25\%$ (Appendix H). Our $722\ \text{m/s}$ ascent deficit, converted back into payload through the stack's mass ratios, implies a comparable multi-tonne offload. The two views agree: reuse costs roughly a quarter of the payload.

The size of that bite depends on how you come home, and the difference is again pure delta-v budgeting. A return to launch site (RTLS) requires a large boostback burn to cancel the booster's downrange velocity and fly it all the way back — expensive in reserved propellant, so it costs the most payload. A droneship landing far downrange skips most or all of the boostback (the booster mostly keeps going the way it was already headed), reserving less propellant and costing less payload. That is why heavy or high-energy missions land on the ship while lighter missions with margin to spare can afford to bring the booster all the way home. The choice of recovery mode is, at bottom, a line item in a delta-v budget — exactly the framing Chapter 22 promised.

⚠️ Common Misconception: "Reuse is free because the rocket comes back." Recovery is emphatically not free — it costs a quarter of your payload before you count a dollar. The point of reuse is not that it is costless but that the thing you buy back (an entire first stage: nine engines, tanks, avionics, worth tens of millions) is worth vastly more than the payload you gave up (a few tonnes of capacity you can often spare, or sell to a customer who doesn't need the maximum). Reuse is a trade, and 38.4 shows why the trade wins — but a trade is not a free lunch, and anyone who tells you reuse "just" saves money has skipped the mass-ratio bill.

🔗 Connection: the whole return is physics you already own. Nothing in the landing is new physics — it is a synthesis. The entry burn slows the booster before the air grows dense, holding peak heating and dynamic pressure within what the aluminum-lithium skin can take, so no heavy heat shield is needed (Chapter 7; Chapter 5). The grid fins steer with aerodynamic lift. The landing burn is timed by the guidance loop of Chapter 27 to bring velocity to zero exactly at the ground — the hoverslam, forced by the deep-throttle-but-still-too-strong Merlin (Chapter 17). Reusability is not a bolt-on subsystem; it reaches back through re-entry physics, aerodynamics, guidance, and the engine cycle all at once.

🔄 Check Your Understanding 1. Why does reserving landing propellant reduce ascent delta-v, mechanically, in the rocket equation? 2. Why does an RTLS recovery cost more payload than a downrange droneship landing?

Answers

1. Reserved propellant is carried during ascent but not burned, so it stays attached to the stage and raises its burnout (final) mass $m_{f,1}$. A larger $m_{f,1}$ means a smaller mass ratio $m_{0,1}/m_{f,1}$, and since $\Delta v = v_e\ln(m_0/m_f)$, a smaller ratio means less ascent delta-v. 2. RTLS needs a large boostback burn to cancel the booster's downrange velocity and send it back to the pad; that burn consumes extra reserved propellant, raising the burnout mass further and cutting ascent delta-v (hence payload) more than a droneship landing, which keeps the booster roughly on its existing downrange path and reserves less.

38.4 Driving down cost

Here is the theme this whole book has been building toward, made quantitative. The physics of landing a booster is settled; the question that decides whether it mattered is economic. We now name the discipline that answers it.

Definition (reusability economics). Reusability economics is the analysis of how recovering and reflying launch hardware changes the cost per flight and the cost per kilogram to orbit: amortizing a vehicle's manufacturing cost over many flights, accounting for refurbishment and the fixed costs reuse cannot recover, and identifying the flight rate and cost floor at which reuse pays. It is the economic counterpart to the mass-ratio bill of 38.3 — the ledger on which the payload you spent buys the vehicle you keep.

Start where the money actually goes, a fact from Chapter 22 worth repeating because it is the crux: a Falcon 9's propellant costs only a few hundred thousand dollars, well under one percent of a launch. Almost the entire cost is hardware — engines, tanks, avionics — historically thrown away after one flight. Reuse is the proposal to stop doing that. Chapter 22 built the model; recall it: $$C_{\text{reuse}}(N) = \frac{M}{N} + R + F,$$ where $M$ is the reusable hardware's build cost amortized over $N$ flights, $R$ is per-flight refurbishment, and $F$ is the fixed per-flight cost reuse never recovers — the expended upper stage, propellant, range fees, and operations. As $N$ grows, $M/N$ shrinks and the cost per flight falls toward a floor of $R + F$. Now let us anchor the whole thing with the number that made the world pay attention: cost per kilogram to orbit, Shuttle versus Falcon 9.

Worked Example: cost per kilogram, then and now. Cost per kilogram is the softest, most-abused number in the business (Appendix H warns at length), so treat every figure here as Tier 2, order-of-magnitude, and watch the ratio, not the digits.

The Space Shuttle, fully accounted, cost on the order of $\$1.5\ \text{billion}$ per flight (the entire program's cost divided by its 135 flights) and could carry about $27.5\ \text{t}$ to LEO. That is $$\frac{\$1.5\times10^{9}}{27{,}500\ \text{kg}} \approx \$54{,}500\ \text{per kg}.$$ Falcon 9, at a list price around $\$62\ \text{million}$ and up to about $22.8\ \text{t}$ to LEO, comes to $$\frac{\$62\times10^{6}}{22{,}800\ \text{kg}} \approx \$2{,}700\ \text{per kg}.$$ The ratio is about twenty to one. Reusable, mass-produced, iteratively refined launch is roughly an order of magnitude cheaper per kilogram than the vehicle that was supposed to make spaceflight routine. That twenty-fold drop — not any single mission — is the revolution.

Three honest caveats keep this from becoming hype. First, price is not cost. The $\$62\ \text{M}$ is what a customer pays; SpaceX's internal cost per flight is reportedly lower, and its price includes margin. Second, per-kilogram flatters the big rocket: it assumes you fill the payload, and a half-empty launch costs the same but delivers a worse per-kg number (our $\$2{,}700$ used near-maximum payload; the $17\ \text{t}$ recovered figure gives closer to $\$3{,}600/\text{kg}$). Third, the Shuttle's $\$54{,}500$ is a full-program accounting including development and standing army; a marginal-flight number is lower. None of these caveats change the headline: the gap is roughly a factor of twenty, a true order of magnitude.

What drove it? Four things, each traceable to something earlier in this book:

• Amortization through reuse (Chapter 22's $M/N$ term): flying one booster many times spreads its build cost over many flights.
• A high flight rate spreads the fixed costs ($F$, the launch site, the workforce) over more launches, and only a reusable fleet can fly often enough to matter.
• Mass production of engines and vehicles: one Merlin production line building hundreds of identical, deliberately un-exotic engines (Chapter 17) is far cheaper per engine than a handful of hand-built masterpieces.
• Minimal refurbishment (a small $R$): Block 5's design-for-reuse is what keeps turnaround cheap — the exact lesson the Shuttle taught in the negative, where refurbishing the tiles and engines between flights devoured the savings reuse was supposed to bring (Chapter 37).

🐛 Find the Error. A commentator argues: "Falcon 9 reuses its first stage, which is most of the rocket, so its cost per flight should fall essentially to the cost of propellant — a few hundred thousand dollars. The rest is just SpaceX overcharging." Where is the reasoning wrong?

Answer

It ignores the fixed cost $F$ that reuse never recovers. In the model $C_{\text{reuse}}(N) \to R + F$ as $N \to \infty$, a floor far above propellant cost — set by the expended second stage (built new every flight), plus refurbishment, range fees, and operations. Propellant is cheap, but you keep buying a fresh upper stage and paying the standing costs on every flight. Recovering the first stage removes the first-stage manufacturing term, not the whole cost. Getting near the propellant-cost limit requires reusing every stage and driving operations down — which is precisely why Falcon 9's success sets up Starship rather than ending the story.

That last point is the bridge to everything that follows. If reusing the first stage alone drops cost to a floor set by the expendable second stage, then the only way past the floor is to reuse the second stage too — and to attack the operations cost with fast turnaround and few people. That is exactly Starship's bet.

🚪 Threshold Concept. For the entire history of rocketry, launch was a product you bought once — you commissioned a rocket, it flew, it was gone. Reusability turns launch into a service you buy repeatedly, priced like a seat on an aircraft rather than the aircraft itself. That single reclassification resets the economics of everything downstream. A twenty-fold cheaper launch does not merely make old missions cheaper; it makes new activities feasible that were unthinkable at the old price — thousand-satellite constellations (Chapter 33), propellant depots, routine crewed flight, and missions that can now afford generous margins instead of shaving every gram. Once you see launch as a commodity service, the map of what is affordable in space redraws itself — and we are at the beginning of that redrawing, not the end.

38.5 Starship: fully reusable super-heavy

Everything so far — iterate fast, block upgrades, propulsive landing, the cost floor set by the thrown-away upper stage — converges on one vehicle. This is the systems-engineering case study the whole book has pointed toward since we first met "Starship" as a running example in Chapter 17. We will define it, and then do the thing the book keeps promising: trace every strange choice back to the physics and economics you already understand.

Definition (Starship). Starship is SpaceX's fully reusable, two-stage, super-heavy-lift launch system: a first stage — Super Heavy — powered by 33 Raptor engines, and a second stage, confusingly also called Starship (or "the Ship"), powered by a mix of sea-level and vacuum Raptors. Both stages are built of stainless steel, burn methalox (liquid methane + liquid oxygen), and are designed to be recovered and reflown — the booster caught by its launch tower, the Ship landing propulsively. Its target payload is roughly $100$–$150\ \text{t}$ to LEO, fully reusable, and it reaches higher-energy destinations through on-orbit refueling rather than a fixed upper-stage burn. As a vehicle still in development, every one of these numbers is a design goal (Tier 2/3), not a demonstrated capability — read the whole definition as intent.

Now the case study. Four choices define Starship, and each looks bizarre until you trace it.

Choice 1 — Stainless steel, not aluminum or carbon fiber. Launch vehicles are almost always built of lightweight aluminum-lithium or carbon-fiber composite, because mass is the enemy (Chapter 23) and steel is about three times denser. Starship is steel anyway, and the reasoning is a beautiful lesson in optimizing the whole vehicle over its whole life rather than a single number.

Worked Example: why a heavier metal can win. Compare three structural materials on the properties that matter for a reusable vehicle that must survive re-entry heating many times (Tier 2, representative):

Material	Density (kg/m³)	Useful up to ~	Rough cost/kg	Re-entry verdict
Aluminum-lithium	~2,700	~400 K (softens); melts ~800 K	~$$$ | needs a full heat shield everywhere | > | Carbon composite | ~1,600 | ~450–600 K (resin degrades) | ~$$$$ | needs a full heat shield; hard to inspect/repair | > | Stainless steel (301) | ~7,900 | ~1,100+ K (retains strength) | ~$ (≈1/50 of composite) | survives with far less thermal protection | > > Steel is three-to-five times denser, so a steel tank wall carrying the same load is heavier — a real > penalty the rocket equation charges. But look at the temperature column. At re-entry, steel keeps its > strength to well over $1{,}000\ \text{K}$, where aluminum has melted and composite resin has charred. So > a steel vehicle needs *far less* thermal protection — over much of its surface, the structure *is* the > heat shield ([Chapter 7](../../part-01-the-physics-of-spaceflight/chapter-07-atmospheric-reentry/index.md); > [Chapter 24](../../part-04-spacecraft-systems/chapter-24-thermal-control/index.md)). The heat-shield mass > you *save* can outweigh the structural mass you *spend*, especially for a vehicle designed to re-enter > hundreds of times. Add that steel costs roughly one-fiftieth of composite per kilogram, can be welded in > the open air rather than cured in a giant autoclave, and stays strong at cryogenic temperatures (it > actually gets *stronger* cold), and the choice that looks like heresy becomes, in Musk's own assessment, > the best design decision on the vehicle. **Choice 2 — Methane, not kerosene or hydrogen.** We laid the groundwork in [Chapters 17](../../part-03-propulsion/chapter-17-chemical-rocket-engines/index.md) and [18](../../part-03-propulsion/chapter-18-combustion-and-propellants/index.md): methalox sits between hydrogen and kerosene in density, burns cleanly (no coking, so engines reuse without soot-clogging), is only mildly cryogenic (close enough to liquid oxygen's temperature to share tank and cooling architecture), and — the decisive systems fact — **can be manufactured on Mars** from atmospheric $\text{CO}_2$ and subsurface water via the Sabatier reaction. A vehicle that intends to refuel on another planet ([Chapter 34](../../part-05-mission-design-and-operations/chapter-34-mission-to-mars/index.md)) cannot burn a fuel it cannot make there. Methane's clean-burning reusability and its Mars-makeability are not two separate perks; they are the same choice seen from Earth and from Mars. **Choice 3 — Full reuse, and the refueling it forces.** Falcon 9 hit a cost floor set by its expendable upper stage (38.4). Starship crosses that floor by recovering *both* stages. But full reuse of the upper stage is brutally expensive in mass ratio: the Ship must carry a heat shield, control flaps, and landing propellant all the way to orbit and back, dead mass that a normal expendable upper stage never hauls. The rocket equation punishes this immediately. > **Worked Example: full reuse eats the mass ratio, so you refuel in orbit.** Model the Ship as an upper > stage with dry mass $\sim 120\ \text{t}$ (structure + heat shield + flaps + engines — Tier 3), > propellant capacity $\sim 1{,}200\ \text{t}$, and a vacuum-Raptor exhaust velocity > $v_e \approx 3{,}700\ \text{m/s}$ ($I_{sp}\approx 375\ \text{s}$). Fully fueled *in orbit* with a > $100\ \text{t}$ payload: > $$m_0 = 120 + 1{,}200 + 100 = 1{,}420\ \text{t}, \qquad m_f = 120 + 100 = 220\ \text{t},$$ > $$\Delta v = 3{,}700 \times \ln!\left(\frac{1{,}420}{220}\right) = 3{,}700 \times \ln(6.45)
= 3{,}700 \times 1.864 = 6{,}900\ \text{m/s}.$$ > That $6.9\ \text{km/s}$ is enough for trans-Mars injection ($\sim 3.6\ \text{km/s}$ from LEO) or a lunar > landing, with margin. But notice the catch buried in the phrase *fully fueled in orbit*: the Ship arrived > in LEO nearly **empty**, having spent almost all its propellant just getting there. To do anything beyond > LEO it must be **refueled in orbit** — by a series of tanker Starships that launch, rendezvous, and > transfer propellant. Needing $\sim 1{,}200\ \text{t}$ of methalox, and with each tanker delivering perhaps > $100$–$150\ \text{t}$, a single Mars-bound Ship may require on the order of a dozen tanker flights (Tier 3, > widely cited as "roughly 8–16"). Such an architecture is only *conceivable* because each launch is cheap > and reusable — orbital refueling is a direct child of the reuse economics of 38.4, and the enabling step > for the Mars mission of [Chapter 34](../../part-05-mission-design-and-operations/chapter-34-mission-to-mars/index.md). **Choice 4 — Catch the booster.** Falcon 9 lands on legs; Super Heavy returns to its launch tower and is *caught* in mid-air by two giant arms — the "chopsticks," first achieved in October 2024 (a feat [Chapter 22](../../part-03-propulsion/chapter-22-staging-and-reusability/index.md) foreshadowed). Why omit the legs? Because legs are dead mass exactly where the rocket equation is stingiest — on the booster you must accelerate — and because a caught booster can be re-stacked and reflown faster, driving down the operations cost $F$ that sets the floor. "The best part is no part": remove the legs, let the tower catch the vehicle, and you save both mass and turnaround time. It is audacious, and it is the same logic as steel and methane — optimize the whole reusable system, not the single flight. The re-entry itself is novel and worth picturing. The Ship does not come down engines-first like a Falcon booster; it falls **belly-first**, flat to the airflow like a skydiver, deliberately presenting a huge area to maximize drag and shed energy high in the atmosphere where the air is thin — a low ballistic-coefficient re-entry, the gentle end of the spectrum from [Chapter 7](../../part-01-the-physics-of-spaceflight/chapter-07-atmospheric-reentry/index.md). Only in the final seconds does it use its flaps to flip upright and light its engines for the landing burn — the "belly-flop and flip" that the SN-series prototypes spent 2020–2021 learning to survive, crash by instructive crash, in full public view. That very public campaign of exploding prototypes was 38.1's iterate-fast philosophy applied at super-heavy scale. > **🚪 Threshold Concept.** Partial reuse makes a rocket cheaper; *full* reuse changes what a rocket *is*. > When both stages come home and fly again with little work, a launch vehicle stops being a consumable — > like a bullet, spent once — and becomes genuine transportation infrastructure, like an aircraft that > flies a route thousands of times. At that point the marginal cost of a flight approaches the cost of > propellant and operations, and the binding constraint on space activity shifts from *can we afford to > launch?* to *what do we want to do up there?* Every choice in this section — steel, methane, catch, > refueling — is subordinate to that one reclassification. Whether Starship achieves it is still being > tested; that the achievement would be categorical, not incremental, is already clear. > **🔧 Engineering Reality: none of these numbers are settled.** Starship is a vehicle under active > development, and its dry mass, payload, propellant load, tanker count, and cost are all *design goals in > flux*, revised with nearly every flight test. Everything in this section is Tier 2 or Tier 3 and should > be read as the *shape* of the argument, not a specification. What is durable is the *reasoning* — steel > for reusable thermal robustness, methane for clean reuse and Mars, full reuse to cross the cost floor, > catch and refuel to save mass and enable beyond-LEO. The specific tonnages will be outdated before this > ink dries; the logic will not. > **🔄 Check Your Understanding** > 1. Steel is three times denser than aluminum. Give the two reasons Starship uses it anyway. > 2. Why does a *fully reusable* upper stage need orbital refueling to go beyond LEO, when an expendable > upper stage of similar size would not? > > Answers > > 1. (i) Steel keeps its strength to far higher temperature (~1,100+ K vs aluminum's ~400–800 K), so the > structure can double as the heat shield and the vehicle needs far less thermal protection — saving mass > that offsets steel's density. (ii) Steel is vastly cheaper (~1/50 the cost of composite per kg), weldable > in the open, and actually strengthens at cryogenic temperature — ideal for a mass-produced, reusable > vehicle. 2. Full reuse forces the upper stage to carry heat shield, flaps, and landing propellant to > orbit and back — dead mass that ruins its mass ratio, leaving it nearly empty on reaching LEO. An > expendable upper stage sheds none of its performance to recovery, so it keeps enough propellant for a > beyond-LEO burn. The reusable Ship must instead be refueled in orbit to restore the mass ratio it spent > on coming home. > --- ## 38.6 Raptor and full-flow staged combustion analyzed At the heart of Starship — 33 in the booster, 6 in the Ship — sits the engine this book has been pointing toward since [Chapter 17](../../part-03-propulsion/chapter-17-chemical-rocket-engines/index.md). We defined its cycle there; here we read the engine as the place where propulsion and reuse become the same problem. > **Definition (Raptor).** **Raptor** is SpaceX's methalox, **full-flow staged combustion** engine, the > powerplant of Starship and Super Heavy. In 2019 it became the *first full-flow staged-combustion engine > ever to fly.* It runs at roughly $300\ \text{bar}$ chamber pressure — among the highest of any operational > engine — producing about $2{,}300\ \text{kN}$ of sea-level thrust (Raptor 2; Tier 2, and climbing with > each version). It is designed above all to be **mass-produced and rapidly reused.** We do not redefine [full-flow staged combustion](../../part-03-propulsion/chapter-17-chemical-rocket-engines/index.md) — recall its scheme from Chapter 17: *two* preburners, one fuel-rich and one oxidizer-rich, pass *all* the fuel and *all* the oxidizer through their turbines before both streams meet as hot gases in the main chamber. What we do here is see *why that exotic cycle is the right one for a reusable engine* — because reuse is exactly what it buys: 1. **Gas–gas injection burns clean and stable.** Both propellants arrive already vaporized, so the injector mixes gas with gas rather than atomizing liquids — more complete, more stable combustion, taming the instability demon of Chapter 17. Clean, stable burns mean less wear, which means an engine you can fly again. 2. **No interpropellant seal.** Each turbine handles only one propellant type (fuel-rich *or* oxidizer-rich), so there is no need for the delicate seal that keeps fuel-side and oxidizer-side gases apart on a shared shaft — historically a prime failure point. Deleting a failure mode is deleting a reason to inspect and refurbish. 3. **Cooler turbines at very high pressure.** Passing the *full* flow through each turbine lets it make its power at a lower temperature. The combination — high chamber pressure for performance, moderate turbine temperatures for longevity — is precisely what a rapidly reused engine needs. High performance *and* long life, the two usually trade against each other; full-flow relaxes the trade. There is a fourth payoff that reaches out of the engine and into the whole vehicle, straight from [Chapter 22](../../part-03-propulsion/chapter-22-staging-and-reusability/index.md): Raptor pressurizes its tanks **autogenously**. It taps its own methane and oxygen, vaporizes them with engine heat, and feeds the gas back to pressurize the tanks — gaseous methane over the liquid methane, gaseous oxygen over the liquid oxygen — eliminating the helium and heavy composite-overwrapped bottles that a conventional vehicle needs. No helium means one less consumable, one less failure mode (recall the 2016 helium-bottle pad failure of Chapter 22), and — decisively — nothing that must be shipped from Earth to refuel on Mars. The engine's cycle and the vehicle's Mars ambition are the same decision. The 33-engine booster cluster is not brute-force excess; it is the Merlin philosophy of 38.2 taken to its limit, and it earns its keep three ways at once. > **Worked Example: why 33 engines.** Take a fully fueled Starship stack at liftoff of roughly > $5{,}000\ \text{t}$ (Tier 3), so it weighs > $$W = 5{,}000{,}000\ \text{kg} \times 9.81\ \text{m/s}^2 = 4.905\times10^{7}\ \text{N} = 49{,}050\ \text{kN}.$$ > With 33 Raptors at $\sim 2{,}300\ \text{kN}$ each, liftoff thrust is > $$T = 33 \times 2{,}300 = 75{,}900\ \text{kN}, \qquad T/W = \frac{75{,}900}{49{,}050} = 1.55,$$
a healthy liftoff thrust-to-weight (Chapter 16 wants roughly $1.2$–$1.5$). Lose one engine and
$T = 32 \times 2{,}300 = 73{,}600\ \text{kN}$, giving $T/W = 1.50$ — barely dented: **engine-out
capability** by design (Chapter 22).
The same 33 engines also mean a single production line stamping out identical units (mass production
drives the per-engine cost down, feeding 38.4's economics), and they let the booster land on just a few
deeply throttled engines. One choice — many small engines — buys thrust, redundancy, cheapness, and
landing control together, exactly as it did for Falcon 9's nine.

📜 From History: the summit that stayed on the test stand. Full-flow staged combustion was not a new idea in 2019 — it was a fifty-year-old one that had defeated everyone who tried it. The Soviet RD-270 of the 1960s and an American hydrogen demonstrator in the 2000s were tested but never flew; the cycle's two high-pressure hot-gas turbines and fearsome plumbing kept it on the ground. Raptor flew it not by out- analyzing those predecessors but by iterating on it relentlessly — building, firing, failing, and refining engine after engine on the test stand until it worked, then flying it. The hardest engine cycle ever devised reached flight because SpaceX applied 38.1's method to propulsion: history's "impossible" is often just "nobody has iterated on it enough yet."

🔗 Connection: the anchor closes. Trace the thread. In Chapter 3 we computed Falcon 9's delta-v from the rocket equation. In Chapters 16–19 we found where its exhaust velocity comes from. In Chapter 17 we climbed the ladder of engine cycles to full-flow staged combustion and named Raptor at the top. In Chapter 22 we staged and landed the booster and did the economics of reuse. In Chapter 23 we chose steel. Every one of those threads is a strand of the same rope, and Starship is where they are braided together: a stainless-steel, methane-burning, full-flow-powered, fully reusable vehicle, each choice a response to the tyranny of the rocket equation and the economics of throwing things away. The book's anchors — Tsiolkovsky × Falcon 9 and Starship-as-case-study — meet here. This is what they were built for.

🔄 Check Your Understanding 1. Name two features of the full-flow cycle that specifically make Raptor easier to reuse, as opposed to merely more efficient. 2. What does autogenous pressurization delete from the vehicle, and why does that matter for a Mars mission?

Answers

1. Any two of: gas–gas injection gives cleaner, more stable combustion (less wear); each turbine handles one propellant type, so there is no interpropellant seal to fail or inspect; passing the full flow lets the turbines run cooler for the same power, extending engine life. All three reduce the wear and failure modes that would otherwise force refurbishment between flights. 2. It deletes the helium pressurant and its heavy high-pressure bottles — the engine vaporizes its own methane and oxygen to pressurize the tanks. This removes a consumable and a failure mode, and, critically, requires nothing that must be shipped from Earth: a vehicle that pressurizes itself from its own propellants can refuel on Mars, where there is no helium supply.

Mission Design Checkpoint: your mission's economics, under reuse

Every chapter has advanced the mission you have been designing since Chapter 1. This chapter does not add a maneuver or a subsystem — it changes the price of everything, so your checkpoint is a reflection with a calculation behind it: how does reusable launch change your mission's economics?

The reflection. Open your Mission Design Review and add a short "launch economics" note. Answer three questions for your track:

• Does cheaper launch change your architecture? At $\$50$–$60\text{k}/\text{kg}$ (Shuttle era) you shaved every gram and flew rarely. At a few thousand dollars per kilogram, mass margin is cheap — you can afford a heavier, more robust, less optimized spacecraft, or fly more often. A Track-A comsat operator might now launch a constellation of many cheaper satellites (Chapter 33) instead of one exquisite bird; a Track-C Mars orbiter might buy margin instead of risk.
• Does reuse fit your flight rate? Recall 38.3–38.4: reuse costs payload and only pays off across many flights. A high-cadence mission loves it; a one-off deep-space probe may fly so rarely that the payload penalty of recovery outweighs any amortization — for a single flight, just expend the booster.
• Where does your launch cost land? Estimate it, and feed the number to your Chapter 40 capstone.

The calculation. Reuse the Chapter 22 cost model and turn it into a cost-per-kilogram your mission can actually use. This small helper builds on mission.py (the module you begin in Chapter 29) and feeds the launch-vehicle selection of Chapter 30:

# astrotools/mission.py -- Chapter 38 increment: launch economics under reuse.
# Builds on the reuse cost model of Chapter 22. Illustrative (Tier 3); never executed.

def reuse_cost_per_flight(M, R, F, N):
    """Cost per flight of a reusable stage flown N times (Chapter 22 model):
    build cost M amortized over N flights, plus refurb R and fixed cost F (all $M)."""
    return M / N + R + F

def cost_per_kg(cost_per_flight_musd, payload_kg):
    """Launch cost per kilogram, in dollars, from cost per flight ($M) and payload (kg)."""
    return cost_per_flight_musd * 1e6 / payload_kg

# Falcon 9-class: booster build M=$30M, refurb R=$1M, expendable-upper+ops F=$18M.
# Fly the booster once (expendable-style) vs ten times (mature reuse); payload ~17 t recovered.
for N in (1, 10):
    c = reuse_cost_per_flight(30, 1, 18, N)
    print(N, "flights ->", "$%.0fM/flight," % c, "$%.0f/kg" % cost_per_kg(c, 17000))
# Expected output:
# 1 flights -> $49M/flight, $2882/kg
# 10 flights -> $22M/flight, $1294/kg

Hand-trace it: at $N=1$, cost $= 30/1 + 1 + 18 = \$49\text{M}$, so $49\times10^{6}/17{,}000 \approx \$2{,}882/\text{kg}$; at $N=10$, cost $= 30/10 + 1 + 18 = \$22\text{M}$, so $22\times10^{6}/17{,}000 \approx \$1{,}294/\text{kg}$. Reuse more than halves the per-kilogram cost — and notice it never reaches zero, bottoming out at the floor $R + F = \$19\text{M}$ set by the expended upper stage and operations (38.4). Point this at your mission's payload and expected flight count, write down the cost per kilogram, and carry it to the capstone: for the first time in your design, launch is a line item you can drive down, not a fixed toll.

Summary

SpaceX's contribution was not a new equation but a new economics, wrung from old physics by iterating on reusable hardware. Carry these forward:

Numbers worth remembering (all Tier 2): Falcon 9 ~$2,700/kg vs Shuttle ~$54,500/kg (≈20× drop); reuse costs ~25% of payload; Raptor ~300 bar chamber pressure; Super Heavy liftoff $T/W \approx 1.5$ with 33 engines; Starship target ~100–150 t to LEO (a design goal, not a record).

Spaced Review

Retrieval strengthens memory. Answer from memory before checking, then look back — this set revisits Chapter 22 (staging, landing, reuse economics) and Chapter 37 (the Shuttle's reusability).

1. (§38.3, Ch. 22) Chapter 22 proved a Falcon 9 booster "cannot hover." State the reason in one sentence, and say what landing technique that forces.
2. (§38.4, Ch. 22) In the reuse cost model $C_{\text{reuse}}(N)=M/N+R+F$, what is the cost floor as $N\to\infty$, and what physically sets it for Falcon 9?
3. (§38.1, §38.4, Ch. 37) The Space Shuttle was reusable, yet it did not make launch cheap. Give the two reasons from Chapter 37 that this chapter builds on.
4. (§38.6, Ch. 22) Autogenous pressurization replaces what, and name one advantage Chapter 22 attributed to it.
5. (§38.5, Ch. 37) The Shuttle carried a crew on its very first flight; Starship exploded prototype after prototype uncrewed. Explain how each choice reflects a different answer to "how do you make a reusable vehicle safe?"

Answers

1. Even at minimum throttle, a single Merlin's thrust exceeds the nearly empty booster's weight ($T/W>1$), so the engine cannot balance gravity to hover; this forces the "hoverslam"/suicide burn — igniting so that deceleration brings velocity to zero exactly at the ground. 2. The floor is $R+F$ (refurbishment plus fixed per-flight cost); for Falcon 9 it is set mainly by the expended second stage built new each flight, plus operations and range costs — which is why full reuse is needed to go lower. 3. Its refurbishment was enormously labor-intensive (tiles inspected/replaced, engines torn down between flights), and its fixed program costs and standing army were huge and spread over a low flight rate — so reuse never translated into low cost per flight. 4. It replaces stored helium pressurant (and its heavy COPV bottles); advantages include mass and plumbing simplicity, one fewer consumable/failure mode, and no reliance on a gas that cannot be produced off-Earth (enabling Mars refueling). 5. The Shuttle bet on making a precious, crewed vehicle safe through exhaustive up-front analysis (and could not test it to destruction); Starship bets on iterating a cheap, uncrewed vehicle to destruction first to discover and fix failure modes, then flying crew only after reliability is earned — safety through iteration rather than through analysis alone.

What's Next

We have watched reusable rockets cut the price of orbit by an order of magnitude, and seen a vehicle designed to cut it by another. That falling price is not the end of the story — it is the enabling condition for everything that comes next. When launch is cheap and frequent, the questions change from "can we afford to go?" to "where shall we go, and to do what?" In Chapter 39 we turn from the past and present to the future the reusability revolution makes plausible: the return to the Moon under Artemis, the long-promised human mission to Mars that Starship's steel and methane and refueling were designed for, commercial space stations, orbital tourism, asteroid mining, and the shape of a space economy to 2040 and beyond. We will try to do there what this chapter did here — separate what the physics permits from what the economics will actually pay for, and tell the possible from the merely dreamed. The rocket equation still rules; but for the first time in sixty years, the ledger it is weighed against has changed.