SpaceX Beat the Earnings. The Capex Number Tells a Different Story.

SpaceX reported its first earnings as a public company on Tuesday, and the headline numbers were better than anyone expected. Revenue hit $7.8 billion, roughly 92% higher than a year ago, against a $6.8 billion Street estimate. Adjusted EBITDA - earnings before interest, taxes, depreciation, and amortization, a rough proxy for cash earnings - came in at $3.5 billion, 75% above the $2 billion consensus. The net loss narrowed from $4.3 billion in the first quarter to $541 million in the second.

The stock fell as much as 8.8% in after-hours trading, then recovered to trade up about 9% as of Tuesday evening.

The initial reaction wasn't to the beat. It was to one number that dominated the conference call: $18.4 billion in capital expenditures for the quarter.

That was roughly $5 billion more than the $13 billion analysts expected. And management said Q3 and Q4 capex would remain at a similar cadence. That annualized run rate far exceeds the $45 billion analysts had expected for the full year, on a company whose trailing twelve-month revenue is a fraction of that annualized run rate. When your capex exceeds your revenue, free cash flow is not a near-term question. It's a structural one.

So what does the earnings actually show, and what is the market supposed to make of it?

The revenue engine is accelerating

All three segments beat. Starlink - the satellite internet service and still SpaceX's only profitable division - brought in $4.29 billion in revenue against a $3.83 billion estimate, generating $1.66 billion in operating income. Subscriber count reached 12 million at the end of Q2, up 1.7 million from Q1. Management also introduced a new $10 monthly terminal rental fee, which will flow through Q3 and Q4. That's a clean margin-expansion mechanic on a base that's already growing.

The AI infrastructure segment produced $2.56 billion in revenue, up from $818 million in Q1. That's a 312% sequential jump in a single quarter. The operating loss was $1.26 billion - roughly $1.1 billion better than the $2.39 billion analysts had projected. The Space segment (launch services) came in at $962 million, against an $835 million estimate.

The AI beat matters because it de-risks the biggest open question in the company's investment case. Three compute leasing contracts anchor the second-half story. Anthropic agreed in May to pay $1.25 billion per month for exclusive access to a Memphis data center housing between 220,000 and 325,000 Nvidia GPUs - a three-year deal running through 2029. Q2 covered two months of that contract, and the $2.56 billion AI revenue implies near-full-rate recognition. Google signed a separate deal at roughly $920 million per month for about 110,000 GPUs, starting in October 2026. Reflection AI added $150 million monthly beginning July 1. Combined, those three contracts represent more than $2.3 billion per month in contracted AI infrastructure revenue by Q4, or roughly $27 billion annualized from that segment alone.

Before Tuesday's report, that math was unsettled. Now it has a confirmed quarter behind it.

The capex problem

Here's where the story gets harder. Most of that $18.4 billion in Q2 capex went to building out the AI infrastructure those contracts are based on. Roughly $16 billion was devoted to the xAI business alone. The company ended the quarter with $93.5 billion in cash and cash equivalents and $65.6 billion in total debt, which gives it $60.7 billion of net cash. That's a fortress balance sheet. But at a run rate of $18 billion per quarter, the cash pile shrinks by $13-14 billion per quarter once you factor in operating costs. Trailing twelve-month free cash flow sits at -$33.4 billion.

The market's question isn't whether the AI revenue will materialize. The contracts are signed. The question is whether the spending cadence will eventually decelerate as those data centers reach utilization and the capital-heavy build-out phase ends. Management didn't give a clear answer on when that inflection arrives. They said the cadence of AI development would "improve dramatically" - but that's an operational claim, not a financial one.

For context, the stock trades at a market cap of roughly $1.65 trillion and an enterprise value of $1.66 trillion. That's 85 times trailing sales. The valuation is priced for a future where capex eventually moderates and free cash flow turns positive. If the $18 billion quarterly run rate persists for much longer than another year or two, that future gets harder to justify at current levels.

The lock-up overhang

Compounding the near-term picture is a lock-up expiration on August 6 - two days after this earnings release. As many as 911.5 million previously restricted shares will become eligible for sale, representing roughly $100 billion in potential supply. Analysts have called it the largest first-tranche insider lock-up release in stock market history. Whether a genuine earnings beat can absorb that kind of supply over a few sessions is an open question. Index-fund investors who own the stock through the Nasdaq-100 will find themselves holding positions they didn't actively choose, which tends to make supply shocks stickier.

Where this points

The old story going into earnings was simple: SpaceX is a growth machine with enormous ambition and enormous costs. The numbers from this quarter confirm the growth part more convincingly than the Street expected - the 92% revenue increase, the EBITDA beat of 75%, the near-tripling of AI segment revenue sequentially. The loss narrowing from $4.3 billion to $541 million in a single quarter is the kind of operating leverage move that gets people excited.

But the new story that emerges from these results isn't a simple beat narrative. It's a tension between revenue that's clearly accelerating and capital spending that has no visible end date. The AI contracts are real and they're large. But the machine that prints them requires an annualized capital spend that far exceeds the $45 billion analysts had expected for the year.

The inflection thesis here would require capex to eventually decelerate as data centers reach full utilization and the company moves from build-out mode to harvest mode. If that happens within a reasonable timeframe - say, over the next two quarters or three - the revenue base supporting that future valuation starts looking credible. If capex stays at this run rate while AI losses persist, the $1.65 trillion market cap is pricing a future that the cash-flow path hasn't confirmed yet.

The lock-up expiration this week is a separate mechanical overhang that doesn't change the operating story. But it does mean the tape is going to be messy regardless of what the fundamentals say.

What would break the setup on the downside: capex guidance that extends the current run rate into the next fiscal year, or any sign that the AI contracts face delay, renegotiation, or slower-than-expected utilization. On the upside, the first quarter where capex meaningfully decelerates while AI revenue stays above $2 billion would be the signal that the transition from build-out to harvest is real. Until that inflection shows up in the numbers, the story remains unproven - even if the growth is undeniable.