SpaceX

Most people think SpaceX is a rocket company. The numbers say otherwise. It is three businesses sharing a name and a founder:

Starlink sells internet access from roughly ten thousand satellites in low orbit (approx.) — to homes without cable, ships, aircraft, armies and, increasingly, mobile phones. In 2025 it earned $11.4bn of revenue, 61% of the company, and $4.4bn of operating profit. In the second quarter of 2026 it had 12.0 million subscribers, twice as many as a year earlier.

The rockets — Falcon 9, Falcon Heavy, Dragon and the giant Starship still in testing — launch almost everything America sends to orbit, including Starlink itself. SpaceX flew 78 rockets in the first half of 2026. The segment loses money because it is spending roughly $2bn a half-year developing Starship.

The AI business arrived in February 2026, when SpaceX absorbed Elon Musk's xAI — the maker of the Grok chatbot, which also owns the social network X — at a valuation of $125bn. It rents computing capacity to other AI companies, sells Grok subscriptions and advertising on X, and has agreed to buy the coding tool Cursor for $60bn.

A ten-year-old would say: SpaceX sells internet from space, which works; flies rockets better than anyone, which is astonishing; and is now trying to build artificial intelligence, which is very expensive. The price you pay today is mostly for the third thing, and for things that have not happened yet.

The IPO. SpaceX listed on Nasdaq on 12 June 2026 at $135, selling 639 million shares for about $85.7bn of net proceeds — the largest share sale in history. The shares closed the first day at $160.95, peaked at $225.64 on 16 June, fell to about $107 in late July as enthusiasm cooled and a Starship booster misbehaved, and stand at $148.03. Two weeks after the IPO the company also sold $25bn of bonds at an average of 5.86%.

⚠️ A note on history. We have one quarter of results as a public company and three years of audited figures from the prospectus. For most companies on this board we judge the future partly by a decade of the past. Here, almost everything that matters to the price is in the future, which is why most of this analysis is about plans and dates.

When a company is priced mostly on what it has not yet done, three questions replace the usual ones.

First: what must be true? Not "is this a great company" — SpaceX plainly is — but "what does the company have to become for today's price to be a fair one?" We answer that with arithmetic in Part XI, and the answer is startling: to earn a 10% annual return from here to 2035, SpaceX must by then be earning roughly $235bn a year of profit. Microsoft, the most profitable software company in history, earned $128.8bn in its last fiscal year.

Second: by when? Value delayed is value lost. A Moon base that pays in 2035 is worth far less today than one that pays in 2030, and with the ten-year Treasury now above 5% the penalty for waiting is higher than it has been in two decades.

Third: how reliable is the calendar? This is where SpaceX is unusual. Almost every date that matters has been set by one man, publicly, many times. So we can do something rare: compare what was promised with what happened, and use the pattern to discount what is still promised. That is the promise ledger in Part IV, and it is the most important table in this report.

★ Buffett's warning about transformative industries is exactly on point. He has written that the automobile and the aeroplane changed the world, and that investors in both lost fortunes: thousands of car makers came and went, and he has described the airline industry's cumulative return to its owners, over its first century, as close to zero. A technology can be real, world-changing and poor for its shareholders at the same time — because the price already assumed the change, or because the change enriched customers rather than owners.

None of that means SpaceX will disappoint. It means the burden of proof sits with the price, and at $1.96 trillion it is very heavy.

Starlink is the business Buffett would want to own. Revenue +66% in the second quarter to $4.3bn; operating profit +79% to $1.7bn; 12.0 million subscribers, adding 1.7 million a quarter. Enterprise and government revenue more than doubled, and SpaceX won over $6bn of multi-year Space Force contracts for Starshield, its military version, in one quarter. American, Southwest, Virgin Atlantic and others have signed for in-flight service. The one soft number is price: average revenue per user fell 22% to $66 a month as growth came from poorer countries and cheaper plans. Starlink is buying volume with price, which is what a low-cost leader should do.

The rockets are the moat and a cost centre. Space revenue was $962m in the second quarter, +29%, but the segment lost $542m because Starship research cost $1.08bn. Note one quiet fact: SpaceX launched 78 times in the first half of 2026 against 84 a year earlier, and put 1,041 tonnes in orbit against 1,102. The Falcon machine is near its natural ceiling; the next leap needs Starship.

The AI business is where the money goes. Its second quarter looked like a turnaround — revenue +247% to $2.56bn and its first positive adjusted EBITDA, $1.15bn — because SpaceX signed agreements to rent computing capacity to other AI companies, worth $14.1bn of contracted sales. But in that quarter the AI segment spent $15.8bn on equipment, took the company's capital spending to $18.4bn, and still lost $1.26bn at the operating line. ⚠️ Press reports on the prospectus describe the largest compute contracts — with Anthropic and Google — as carrying 90-day termination clauses. "Contracted sales" in SpaceX's own definition covers only the non-cancellable period.

★ Put simply: the best business in the group is being used to finance the most speculative one. In six months Starlink earned $2.8bn of operating profit and the AI business spent $23.6bn on equipment — eight times as much. Free cash flow for the half was about −$25bn. The IPO and the bond sale raised $111bn to pay for it. That is not a scandal; it is a strategy. But it is the strategy of a venture fund, and it should be priced like one.

Here is the vision, in the company's own terms: make life multiplanetary. Everything else — Starlink, the rockets, and now AI — is described as the means of paying for it. Musk's pay package makes the vision literal: in January the board granted him 1 billion additional shares in 15 tranches, vesting only as SpaceX's market value climbs from $500bn to $7.5 trillion and SpaceX establishes a permanent human colony on Mars of at least one million people.

We take the vision seriously. We also take the calendar seriously, which is a different thing. Before the table, the calibration.

★ SpaceX's record is late, not fake. Almost everything it has promised, it has eventually done: landed and re-flown orbital boosters, carried astronauts to the space station, caught a Super Heavy booster in the launch tower's arms, built the largest satellite network in history. But almost nothing arrived on the date first given. Crew Dragon was meant to carry astronauts in 2017; it did so on 30 May 2020. Our working rule for the dates below is the one the record supports: add two to three years to anything not yet demonstrated, and more to anything that depends on something else not yet demonstrated.

Three things stand out from the ledger.

One: the proven business is not what the price is for. Starlink is delivering on schedule, or close to it, and it is worth a great deal — Part XI puts it at $300–500bn. But the company is valued at nearly $2 trillion. The difference rests on the rows coloured amber and red.

Two: nearly every red row depends on one amber row. The Moon, Mars, cheap orbital data centres and Starlink's next generation all require Starship to fly often, cheaply and with propellant transferred between ships in orbit. That demonstration has slipped twice and is not yet scheduled. It is the single most important event for SpaceX's long-term value, and when it happens — probably 2027 — it will deserve more attention than any quarterly result.

Three: the vision has already changed once this year. In February the priority moved from Mars to the Moon. Twelve months earlier the Moon was "a distraction". We do not criticise that — the Moon is closer, launch windows come every ten days rather than every twenty-six months, and it is the sensible first step. But an investor should notice that the plan the price depends on is revised by one person, in public, as he goes.

Type: cost advantage through vertical integration, the rarest and most durable kind. SpaceX builds its own engines, rockets and satellites, launches its own satellites on its own rockets, and sells the service directly. Every competitor in launch or broadband is paying a margin to someone that SpaceX keeps. We score the moat 8 rather than 9 or 10 only because the third business — now the largest use of capital — has none.

Governance — the part we must dwell on. SpaceX is the first major American IPO to adopt mandatory shareholder arbitration under the SEC's 2025 policy change: most claims, including securities-fraud claims, must be brought individually in arbitration, with class actions and jury trials waived. CalPERS and the New York State and City comptrollers wrote jointly to object. Combine that with ten-vote Class B shares, Texas incorporation, a chief executive who also runs Tesla and whose companies trade with each other, and an outside shareholder has almost no remedy if things go wrong. Buffett's first test is integrity, and we have no evidence of dishonesty here. But he also asks whether owners are treated as partners, and these documents answer that question plainly.

Capital allocation is where we score 4. The engineering capital allocation is superb — Falcon and Starlink are among the best investments of the century. The corporate capital allocation of 2026 is another matter: xAI absorbed at $125bn in a transaction between two Musk-controlled companies, $60bn for Cursor, and $23.6bn of AI capex in six months, from a company whose only profitable division earned $2.8bn in the same period.

Balance sheet at 30 June 2026: cash and marketable securities of $100bn, debt of $39.4bn (including the $25bn June bond at a weighted 5.86%), equity of $127bn and backlog of $47.5bn. Net cash of about $60bn — before the $60bn Cursor purchase, expected to close in the third quarter, on terms whose cash component we have not seen confirmed. Figures for 2023–2025 include xAI on the combined basis presented in the prospectus.

SpaceX pays no dividend and, on its current plans, cannot. Free cash flow was −$14.0bn in 2025 and about −$25bn in the first half of 2026; the spending is financed by the IPO and bonds. There is no cover to test, no trend to measure and nothing that could be cut. For an income investor, the answer is simply that this is not an income security, and will not be one this decade.

★ The one dividend-adjacent point worth making is about Starlink. On its own, a business with $10bn of annualised adjusted EBITDA growing 60% a year would, once the constellation matured, be a formidable cash distributor. Inside SpaceX, that cash will fund Starship, the Moon and AI for as long as Musk runs it — which the pay package is designed to make a very long time.

SpaceX pays no dividend, so a shareholder's entire return must come from a rising share price. Suppose you want 10% a year — roughly what the stock market has paid over long periods. From ~$1.96tn today, that means a company worth about $4.7 trillion at the end of 2035. If by then SpaceX is a mature, highly profitable company valued at a generous 20–25× earnings, it must be earning $190–235bn a year.

For scale: Microsoft earned $128.8bn in its last fiscal year. SpaceX lost $4.9bn in 2025 and consensus has it earning perhaps $12bn in 2028. The price requires SpaceX to become, within a decade, one and a half to two times as profitable as the most profitable software company in history — while also funding the Moon and Mars. That is not impossible. It is the base case the price has already assumed, and there is no margin for the calendar slipping.

★ Our playbook's new rule does not apply here, and it is worth saying why. After Microsoft we committed never to demand a bargain for a wonderful business trading inside our own estimate of value. SpaceX's core businesses are wonderful. But the price is not inside our range — it is almost half again above the top of it — and our quality score is dragged down by governance and capital allocation. So "wait" is the honest answer, and the zone we name is where the knowable businesses pay for the price, not where everything becomes a bargain.