Starship: The $15 Billion Hole That Buys the Whole Moat
Starship is the single largest money sink in SpaceX's S-1 — $15 billion-plus of cumulative capex through 2025 with no commercial revenue yet — and simultaneously the reason the launch chain has no credible challenger. The same vehicle that drags on near-term earnings is the only one that can fly NASA's astronauts to the Moon and the only one big enough to carry the next-generation Starlink constellation. Remove it and you strand both. That is not a contradiction; that is the moat. Figures from the SpaceX Form S-1 (SEC EDGAR CIK 0001181412) and the sources below.
Starship is a fully-reusable, two-stage super-heavy-lift stack: the Super Heavy booster (33 Raptor engines, caught at the tower by the “chopstick” arms) and the Starship upper stage, both designed to fly, return, and fly again. Falcon 9 only recovers its first stage; the second stage is thrown away every flight. That single architectural difference — recovering the upper stage — is the entire cost argument, and it is also the hardest engineering problem SpaceX has ever taken on.
The cumulative bill through 2025 is $15 billion-plus per the S-1, the largest capex sink in the company. What that money is supposed to buy is a step-change in $/kg-to-orbit. Falcon 9's public list price works out to roughly $2,720/kg to LEO at ~22.8 t reusable payload; SpaceX's internal cost is far lower — reporting in early 2026 pegged it near $300/lb (~$629/kg) through vertical integration. Starship's target is under $100/kg, with credible near-term projections in the $78–$94/kg range as reuse matures, and the often-quoted ~$2M marginal-launch-cost figure is Musk's, not a printed S-1 number — treat it as aspiration. Payload to LEO for the V3 vehicle is advertised at 100+ metric tons reusable, an order of magnitude over Falcon 9.
The engine is the lever. Raptor is a full-flow staged-combustion methalox engine — a cycle only Starship flies operationally — and the V3 vehicle moves to Raptor 3, redesigned for lower part count and integrated heat-shielding so it can survive reentry and be reflown with minimal refurbishment. The whole cost model assumes the engines come back hot and fly again next week. That is the difference between Falcon 9's economics, where a new expendable upper stage and its single Merlin Vacuum engine are written off every flight, and Starship's intended economics, where nothing is thrown away. Propellant — liquid methane and liquid oxygen — is a rounding error against hardware; on a fully-reused flight the marginal cost converges toward fuel, range, and refurbishment labor rather than a new airframe.
Here is the honest read: those $/kg figures are targets, not realized economics. As of mid-2026 no Starship has flown a paying commercial payload. The $15B is sunk; the payoff is contingent on a reuse cadence that does not yet exist. But the comparison set is what makes it matter — there is no competitor within a factor of ten. Blue Origin's New Glenn is partially reusable and roughly a third of Starship's payload class; China's reusable programs and Rocket Lab's Neutron are years behind and smaller still. Even if Starship lands only halfway to its targets — say $300–$500/kg at moderate cadence rather than sub-$100 — it still resets the floor for the entire industry and leaves rivals pricing against a vehicle they cannot match on either payload mass or per-kilogram cost. That asymmetry is the asset the $15B is really buying.
Starship $/kg is a target, not realized economics — no commercial payload has flown as of mid-2026.
Do not take the marketing cadence at face value. As of late May 2026, Starship had flown 12 integrated flight tests, 7 successful and 5 failures. The vehicle is mid-transition from Block 2 (V2) to Block 3 (V3), and the transition has been rough.
Flight 11 (Oct 13, 2025) was the high-water mark for the V2 generation: a clean suborbital flight, both stages to pinpoint splashdowns, an in-space engine relight, and eight dummy payloads deployed. The booster was not tower-caught on that flight — it did a soft splashdown in the Gulf and was expended. Flight 12 (May 22, 2026) debuted the V3 vehicle (Ship 39, Booster 19, Raptor 3) and stumbled: Booster 19 was lost after an off-nominal boostback burn, and Ship 39 took an ascent engine-out and could not perform its planned in-space relight. So the current state of play is blunt: tower-catch of the booster has been demonstrated on earlier flights, routine catch-and-reuse of both stages is not yet operational, and the V3 vehicle — the one that actually carries 100+ tons and deploys Starlink V3 — has one troubled flight on the board.
The cadence story is the gap between ambition and telemetry. SpaceX talks about dozens of flights a year and 100+ flights per vehicle. The 2026 reality is roughly a flight a month with a 58% success rate and no commercial payload flown. That gap is exactly why Starship is still a capex line and not yet a revenue line — and why the bull case rests on future cadence, not demonstrated cadence.
The Block-2-to-Block-3 transition deserves its own caution. SpaceX retired the V2 vehicle on a high (Flight 11's clean dual splashdown) and then immediately introduced a larger, more complex V3 stack that failed on its debut. That is the normal SpaceX development signature — fly, break, fix, repeat — and the company has historically converted exactly this kind of early-flight carnage into reliability (Falcon 9 went from a 2015–2016 run of failures to the most-flown, most-reliable orbital rocket in history). But for an investor reading the S-1 in June 2026, the relevant fact is that the reliability and reuse cadence the entire $/kg thesis depends on is still in front of the program, not behind it. Two things must happen before Starship stops being a money pit: routine tower-catch and rapid reflight of both stages, and a first operational payload flight. The orbital-refueling demonstration Artemis requires — multiple tanker flights to top off a depot in orbit — has also not been shown end to end, and it is on the critical path for the Moon.
This is where Starship stops being a cost center on paper and becomes the structural reason the launch chain cannot be unwound. Two independent, high-stakes programs are bolted to it, and no alternative vehicle exists for either. NASA's return to the Moon: SpaceX holds the Human Landing System contract — a $2.89B firm-fixed-price award (2021) for the Artemis III lander plus a demo, expanded by a $1.15B Option B for Artemis IV, roughly $4.3B total. NASA has no second vehicle ready to land humans on schedule; the Feb 27, 2026 replan makes Artemis IV (targeted 2028) the first crewed landing. Kill Starship and you kill the timeline for an American crewed return.
SpaceX's own next-gen constellation:Starlink V3 satellites weigh ~1,760 kg and are physically too large for Falcon 9's fairing. They can only fly on Starship, and each Starship-deployed batch adds 20x+ the network capacity of a Falcon 9 V2 mission. Starlink funds the build — $11.4B, 61% of FY25 revenue — so the cash engine paying for Starship is the same engine that needs Starship to keep growing. Mars is the long-tail option value on top.
That is the launch leg of the unremovability thesis, and it connects directly to the broader dependency graph — see /dependencies/. NASA's lunar program and SpaceX's revenue flywheel both terminate at one vehicle with no substitute. Starship is the biggest in-progress money-loser in the S-1 and the biggest moat-widener in it — the same fact viewed from two sides of the ledger. The financials show the capex drag; the Starlink engine shows what the capex is for.
Q1. How much has SpaceX spent on Starship, and is it making money yet?
SpaceX's Form S-1 (SEC EDGAR, CIK 0001181412) reports cumulative Starship/Super Heavy capital expenditure of $15 billion-plus through 2025 — the single largest capex sink in the company. As of mid-2026, Starship has generated essentially no commercial launch revenue: it has not yet flown a paying payload. That is what makes it a drag on near-term GAAP earnings even as the consolidated company posts $18.7B in FY25 revenue. The $4.9B FY25 net loss is driven primarily by xAI's operating loss (about -$6.355B), not Starship directly, but Starship's capex is the reason the launch chain's free cash flow is thinner than Starlink's segment economics alone would suggest. The payoff is contingent: it only converts to profit once reusable cadence drives $/kg down and Starlink V3 plus paying customers start flying at scale. Until then it is sunk cost with optionality. The honest framing for an investor is that you are underwriting a $15B-plus option whose strike price is "routine full reuse" — an option Starlink's $11.4B engine is paying the premium on every quarter it stays unexercised.
Q2. What is the realistic cost per kilogram to orbit on Starship versus Falcon 9?
Falcon 9's public list price implies roughly $2,720/kg to low Earth orbit at ~22.8 t reusable payload; SpaceX's internal cost is far lower, reported near $300/lb (~$629/kg) in early 2026. Starship targets under $100/kg, with credible near-term projections of $78–$94/kg as reuse matures, and a long-run aspiration (Musk's, not an S-1 figure) of single-digit-dollar $/kg at very high flight counts. Two caveats matter. First, those Starship numbers are targets — no Starship has flown a commercial payload as of mid-2026, so the realized cost is unproven. Second, the gap to competitors is enormous: nothing else in service or near-service is within a factor of ten of Starship's target on a $/kg basis. The economic thesis is real but forward-looking; the moat is the absence of any rival positioned to chase the same curve. The distinction that matters is list price versus internal cost: Starlink launches at SpaceX's internal cost, not the public sticker, which is precisely why no rival buying launch on the open market can match its connectivity unit economics.
Q3. How many Starship test flights have happened, and have any succeeded?
As of late May 2026, Starship has flown 12 integrated flight tests: 7 successes and 5 failures. Flight 11 (Oct 13, 2025) was the standout V2-generation flight — both stages to controlled splashdowns, an in-space engine relight, and eight dummy payloads deployed — though the booster was expended, not tower-caught, on that flight. Flight 12 (May 22, 2026) introduced the larger V3 vehicle and went poorly: the Super Heavy booster was lost after an off-nominal boostback, and the upper stage suffered an ascent engine-out. SpaceX has demonstrated tower-catch of the booster on earlier flights, but routine catch-and-reuse of both stages is not yet operational, and the V3 vehicle that actually carries 100+ tons has just one troubled flight on record. Treat any "dozens of flights a year" framing as a target the program has not yet hit. The test record is the honest counterweight to the cost-per-kilogram story: every $/kg projection above assumes reliable full reuse, and reliable full reuse is exactly the milestone these 12 flights have not yet delivered.
Q4. Is Starship required for NASA's Artemis Moon landings?
Yes — and there is no backup. SpaceX holds NASA's Human Landing System contract, worth about $2.89B for the Artemis III lander and demo (2021), plus a $1.15B Option B for Artemis IV, roughly $4.3B total. A Starship variant is the crewed lunar lander. NASA has no alternative vehicle ready to put astronauts on the Moon on the current schedule. On Feb 27, 2026, NASA revised the plan: Artemis III will conduct Earth-orbit rendezvous and docking tests with the commercial landers (SpaceX's Starship HLS and Blue Origin's Blue Moon), and Artemis IV — targeted for 2028 — becomes the first crewed lunar landing. The timeline has slipped, but the structural dependency is complete: there is no scenario where America returns crew to the lunar surface on this program without Starship working. That is the unremovability thesis on the government side — NASA cannot cancel the contract without canceling the program, which is exactly why the $4.3B HLS award is better read as leverage than as ordinary procurement revenue.
Q5. Why can't Starlink V3 satellites just launch on Falcon 9?
Because they physically do not fit and are too heavy to fly economically on Falcon 9. Each Starlink V3 satellite weighs about 1,760 kg and is roughly 7 meters long — too large for Falcon 9's fairing. They can only launch on Starship. The capacity math is the point: a single Starship deploying a batch of V3 satellites adds on the order of 20x+ the network capacity of a Falcon 9 mission carrying V2 satellites. This creates a tight loop in the S-1's economics. Starlink is $11.4B, 61% of FY25 revenue — it is the cash engine funding Starship's $15B-plus build. But future Starlink growth depends on V3, and V3 depends on Starship flying reliably and often. As of mid-2026 Starship has deployed only dummy V3-sized payloads in testing; at-scale V3 deployment cannot begin until Starship reaches routine, reliable flight. That circular dependency is the whole investment case in miniature: the cash engine funds the rocket, and the rocket is the only thing that can keep the cash engine growing.
- SpaceX Form S-1 / S-1A — SEC EDGAR · CIK 0001181412 (Starship cumulative capex, FY25 financials)
- List of Starship launches — flight count, success/failure record, V2/V3 history
- SpaceX Starship — architecture, Raptor, V3 100+ t spec, Flight 12 outcome
- NASA & Starship HLS — $2.89B base + $1.15B Option B (~$4.3B total); Artemis IV 2028 replan
- Launch-cost analyses (SpaceNexus, Orbital Radar, NextBigFuture) — Falcon 9 ~$2,720/kg list, ~$629/kg internal est.; Starship target $78–94/kg
Informational analysis, not financial advice. $/kg figures are targets/estimates as flagged; verify against EDGAR before acting.