SpaceX's Earnings Debut: Who's Actually Paying the Bills

The obvious question about SpaceX's first public earnings report - due August 4 - is whether the numbers will justify a trillion-dollar-plus valuation. That is the wrong question. The right one is simpler and harder: which part of the company is actually running a business, and which parts are running an option?

Most people think SpaceX is a rocket company. The filings make it clear that it is something else entirely. It is a satellite ISP with a rocket division and an AI experiment attached. And the satellite ISP is the only thing that prints money.

Here is what the S-1 filing showed before the IPO, and what it means now that the stock has fallen well below its offering price.

In 2025, Starlink - the connectivity segment - generated $11.4 billion in revenue and $4.4 billion in operating income. That is a 39% operating margin. The Space segment brought in $4.1 billion and lost $657 million. The AI segment, which includes the merged xAI operation and the X platform, produced $3.2 billion and lost $6.4 billion. The AI unit spent more than double what it earned.

So the profitable part is the satellite internet business. The part everyone is excited about - AI - is a cash furnace. And the part everyone thinks the company is built on - rockets - is barely covering itself.

The per-user economics in the AI segment make the loss even stranger to look at. Derived figures from the S-1 put blended revenue at roughly $27 per annual Grok user, against a fully loaded cost of about $191. The math does not close. The segment survives because of a compute lease to Anthropic - $1.25 billion per month through May 2029 - which is the single largest revenue line in that division. Without it, the hole gets much worse. And the deal has a 90-day termination clause.

That is the kind of detail that gets lost when you are thinking about SpaceX as one company. It is three companies with very different unit economics, consolidated onto one balance sheet.

Starlink works because it has the classic software-like leverage of a network. Once the satellites are in orbit - roughly 9,600 as of the filing - each new subscriber adds revenue at near-zero marginal cost. The company ended March with 10.3 million subscribers, up from 8.9 million at the end of 2025. Revenue grew 50% in 2025 while operating income grew 120%. That is what happens when a fixed cost base spreads over a growing customer base.

The Space segment is an option. SpaceX launches more than 80% of all mass to orbit globally. Falcon 9 boosters have been reflown 34 times. Starship is still in development. The segment loses money because of Starship R&D - roughly $3 billion in 2025, another $930 million in Q1 2026 - and will probably keep losing money until Starship becomes operational. The payoff could be enormous if SpaceX achieves its goal of fully reusable heavy-lift launches at aviation-grade turnaround times. But it is an option, not a cash flow.

AI is the biggest bet and the most uncertain. Capital expenditure there accelerated from $12.7 billion in 2025 to $7.7 billion in Q1 2026 alone, an annualized pace of roughly $31 billion. The operating margin deteriorated from -60% in 2024 to -200% in 2025, and the Q1 2026 run-rate implies roughly -302%. Scaling is making the economics worse, not better.

This is the contradiction the stock now has to explain. SpaceX went public at $135 per share in June, climbed to $225.64, and has since fallen below $118. The stock dropped under its IPO price in mid-July and continues to slide. Two trading days after the August 4 results, up to 20% of insider and employee shares unlock. That could more than quadruple the freely tradable float. The earnings report will arrive just before what could be the largest wave of selling pressure since the offering.

I suspect the market's impatience is partly justified. The company posted a $2.6 billion operating loss in 2025 and a $1.9 billion operating loss in Q1 2026 alone. Cumulative losses since founding in 2002 stand at roughly $41 billion. But the composition of those losses matters. Most of them are going into things that could produce superlinear returns: a satellite constellation that compounds, a rocket system that could reduce launch costs by two orders of magnitude, and an AI infrastructure play that is still early.

The problem is that superlinear returns require time, and public markets do not like waiting. The AI segment in particular is burning capital at a rate that no amount of Starlink profit can sustain if it does not improve. The Anthropic lease is a bridge, not a destination.

What the August 4 numbers need to show is not profitability. They need to show trajectory. Is Starlink growth holding its pace - or is subscriber growth slowing after doubling in 2025? Is the AI loss widening at a manageable rate, or is the capex spiral getting out of control? Is the Space segment on a credible path to Starship deliveries in the second half of 2026?

The test for investors is not whether the stock bounces off these earnings. It is whether you can separate the cash engine from the options and value them differently. Starlink is a business. Starship is an option. xAI is an experiment that might become a business if the Anthropic-style model scales to other customers. Treating them as one thing leads to confusion. Treating them as three things lets you decide which one you are actually betting on.

The thing about betting on options is that most of them expire worthless. The thing about Starlink is that it is making money today. The earnings report will not tell you which bet wins. It will tell you whether the cash engine is strong enough to keep funding the experiments. That is the question worth asking before the stock does the rest of the talking.