SpaceX's $100B ARR Target Is Booked Compute, Not Banked Revenue

SpaceX's CFO just announced another AI compute deal worth about $1.1 billion a month and used it to push the company's biggest number: $100 billion in annual recurring revenue by year-end. On its face that sounds like a rocket ship. It isn't what it sounds like. And the gap between what the number means and what the market is paying for it is the whole investment case.

Start with what the company actually reported, because the market didn't like it. In its first report as a public company, for the quarter ended June 30, SpaceX posted $7.8 billion in revenue, up 92% from a year earlier, with $3.5 billion in adjusted EBITDA. That is, on its own, a $31 billion run rate — Deutsche Bank's Edison Yu put the June-quarter pace at $31 billion and called the $100 billion goal "likely very achievable" on neocloud and Cursor contribution.

But the same report showed something the market punished: capital expenditures of $18.4 billion in a single quarter, more than double the revenue it earned, much of it poured into AI compute. SpaceX's AI segment lost $1.26 billion on operations in that quarter, and lost $6.35 billion across the whole of 2025. That is the story that took the stock from a post-IPO pop above $170 down to an all-time low in August, below its $150 IPO price.

Here is the disconnect the market keeps tripping over, and it is the thing to understand.

The $100 billion is booked capacity, not banked revenue

"Annual recurring revenue" in SpaceX's telling is not money showing up in the income statement today. It is the annualized value of compute contracts the company has signed. The new deal — with an unnamed customer, worth about $13 billion a year — doesn't start generating revenue until December 1. It started at all because the company built the capacity first and is now renting out what it hasn't consumed itself, alongside deals with Anthropic at $1.25 billion a month through May 2029 and Alphabet at $920 million a month.

So when the CFO, Bret Johnsen, says SpaceX can reach $100 billion in annual recurring revenue by year-end, he is describing the contracts in hand, not what the P&L books in the next fiscal year. Musk framed the same target on the earnings call as a baseline the company could hit "if we basically did nothing" — meaning the signatures exist and the revenue is a matter of when, not whether. That is the bull reading, and it is real.

Here is the counterweight, stated fairly because it could change the conclusion: Johnsen also said that "almost all, if not all" of the compute deals let either party walk away after a few months. The contracts are real but cancellable, and they live inside a segment that currently loses money at scale — $1.26 billion in a single quarter while capex runs at 2.4 times quarterly revenue. If a customer that signed for $1.1 billion a month decides its own models don't need the capacity, the "locked-in" revenue disappears.

The scarcity test is the actual question

Strip away the booking mechanics and the real debate is whether SpaceX's compute is the scarce asset of this buildout or just more capacity in a market drowning in it. The abundance-scarcity frame says intelligence is becoming abundant, so what becomes scarce is the thing that generates it — power, chips, data-center capacity, and launch. SpaceX is the only company that controls all four at once: Colossus on the ground, a vertical launch monopoly in Starship, and a plan to put compute in orbit, with first orbital compute satellites targeted for 2027.

That is the asymmetry. The market has been pricing the part that loses money today — the $18.4 billion capex and the $1.26 billion AI loss — and underpricing the part that shows up only in signed contracts and a 2.4-times-revenue capex bill. A company spending more than double its quarterly revenue to build orbital and ground compute is not a company betting on nothing; it is a company betting that compute is scarce enough to monetize, and it is front-running that bet with its own balance sheet.

You still have to pay for the valuation while the bet resolves. At roughly $2 trillion in market value — back in the "$2 trillion club" in early September — and a $100 billion ARR target, the shares trade around 20 times the booked run rate if the target lands, but that multiple sits on an AI segment that loses money and a capital bill that keeps growing. That spread is why the price targets span $115 from CFRA on the bear side to $250 and $401 from Oppenheimer and Arete on the bull side, with New Street's cautious $165 target and the stock trading below it around $148 as of September 10.

A 20-times-ARR multiple on a loss-making, capex-heavy compute business is not cheap. It is only worth paying if the scarcity thesis converts today's booked, cancellable capacity into tomorrow's high-margin, durable revenue — and if the capex that is scaring the market today turns out to be the moat, not the hole. The CFO's $1.1 billion-a-month deal is evidence for that thesis, not proof of it. Watch whether the customers stay past the escape clauses and whether the AI segment's loss narrows as the booked capacity comes online. The market sold the losses; the question is whether it has priced the asymmetry, and that gap is where the work is.