When a transcript is available, the following piece will be updated to include more direct citations from the SpaceX earnings call. What follows encompasses information from the SpaceX earnings digest and our live listen of its moot with investors.
Today, after the bell, SpaceX reported its second-quarter earnings. The street expected revenue of $6.93 billion and a per-share loss of $0.26. Instead, SpaceX reported revenue of $7.81 billion (+92% YoY) and a $0.09 per-share loss. In total, the company was effectively break-even on an operating basis. After rising 9.4% during regular trading, shares of SpaceX are down around 7% in after-hours trading.
- Buy the rumor, sell the news. AI compute growth (see below) was priced in.
Before SpaceX’s second-quarter results dropped, ** CO **formulated four questions that we wanted answered. Here they are, and what we learned from the company’s earnings results regarding each query:
Space margins: Is the business healthy?
Space revenues at SpaceX grew to $962 million in the second quarter, up from $619 million in Q1 2026, and $746 million in Q2 2025. The company’s space business lost $542 million on an operating basis in the second quarter, down from $662 million in Q1 2026 and up from $369 million in Q2 2025.
Why did revenue grow so quickly compared to the first quarter? More customer launches in the second quarter (10) than in the first (7). SpaceX’s own launches (“Internal launches”) fell from 33 to 28 from the first quarter to the second. Space at SpaceX is more than only launches, however. The company also books “launch & development revenues” in the segment, which grew to $314 million in Q2 2026 compared to the first quarter.
- Space capex rose from $1.05 billion in Q1 2026 to $1.17 billion in the second.
Starship’s recent successful launch turned heads (the company claims it’s on the “precipice” of commercial Starship operations), but SpaceX’s space business today remains a relatively modest portion of its total topline. Perhaps today it’s best to consider space (launch) at SpaceX as the chariot for its profit engine:
Starlink economics: How quickly can SpaceX add subscribers, and at what cost to ARPU?
Starlink had a good quarter, adding 1.7 million subscribers to its satellite Internet business (12.0 million Q2 2026 compared to 10.3 million Q1 2026; Starlink added more subscribers in Q2 than in Q1). More importantly, the company’s average revenue per user (monthly) failed to decline, staying pat at $66. That is the same number SpaceX reported for the first quarter, and while far lower than the year-ago $85, was static; there had been concern amongst investors that falling Starlink ARPU would bend its revenue growth curve downward over time.
Total “Connectivity” revenue at SpaceX rose from $3.26 billion in Q1 2026 to $4.29 billion in the second quarter ($2.59 billion in the year-ago period). Especially buoyant at Starlink in the period were “Enterprise & government revenues,” which soared from $1.11 billion in the first quarter to $1.81 billion in the second.
All told, Starlink generated $1.66 billion worth of operating income in SpaceX’s most recent quarter, up from $1.19 billion in the first. Very good. Capex was all but flat quarter-over-quarter to boot.
- For those looking further ahead, Starlink’s future cellular connectivity business ( more here) was the topic of many analyst questions. Shares of major American cellular companies were largely flat after the call.
Compute deals: How much revenue are they driving today?
A lot! First, advertising revenues fall under the AI segment at SpaceX because X (Twitter) offers Grok access to paid subscribers; the revenue is already mixed. Regardless, advertising revenues at SpaceX were $367 million in the second quarter, up from $343 million in the first, and down from $426 million in the year-ago period.
The “AI solutions & infrastructure revenues” line item in the AI category at SpaceX was more exciting, growing from $475 million (Q1 2026) to $2.19 billion (Q2 2026). Much of that was due to SpaceX earning $1.6 billion in “incremental AI infrastructure revenues” during the second quarter, per the company. SpaceX notes that revenue in its AI group rose in part thanks to “an increase in Grok and X subscription revenues.” How much? Looking at the figures, not that much.
- For fun, SpaceXAI had “approximately 0.9 million SuperGrok and SuperGrok Heavy paid subscribers” at the end of 2025, and “approximately 1.9 million SuperGrok, SuperGrok Heavy and SuperGrok Lite paid subscribers,” per its S-1 filing at the end of the first quarter. You can interpolate from there.
In short, AI at SpaceX is mainly neocloud-ish revenues, sold to major partners (competitors, really) who wanted to use the company’s spare GPUs. Which are costly, as SpaceX notes that it racked up $1.89 billion worth of depreciation (and amortization, to be fair) in the period. Or less than its compute footprint generated in revenue terms from third-parties, if you are keeping count. Talk about a quarter-saver; if there wasn’t excess compute demand in the market, SpaceXAI would look pretty ugly at present, in economic terms.
- During its earnings call, the company said that its compute deals reduced the payback period for new hardware to under a year. The company expects $100 billion worth of annual run rate (including Cursor) by December of 2026, largely driven by its compute agreements.
- SpaceX said it added more than $6 billion in new compute deals thus far in Q3.
- How compute-limited is the company, and what does it need to do to balance internal demand versus serving customers? Musk said that he expects the share of compute reserved for internal uses to fall over time compared to what it serves to others; perhaps 10% used for Grok training, per the CEO.
All told, SpaceXAI lost $1.26 billion on an operating basis in the quarter, though if you yank out depreciation costs and add back in $516 million worth of share-based compensation expenses, it made money! Viva la adjusted EBITDA.
- Recall that SpaceX expects its acquisition of Cursor to wrap in the third (present) quarter; that deal will add revenue (and costs) to SpaceXAI’s business in its second public earnings drop.
- During its earnings call, SpaceX promised Grok 4.6 and 4.7 in the coming weeks, and Grok 5 before the end of the year, stuffed with all historical SpaceX data.
What’s the capex ramp looking like?
As noted, SpaceX’s space and Starlink capex were relatively flat in the quarter. So, what is the full story when we include AI? This:
Ah, nevertheless. It’s very expensive to add 0.4 gigawatts (SpaceX closed Q1 with 1.0 gigawatts of compute, and ended the second quarter with 1.4 gigawatts). Musk mentioned on its earnings call that it is “exclusive” with Nvidia for its compute ramp, and will reach 2.0 gigawatts this year. (The company also namechecked its upcoming ‘Starmind’ compute satellites during its call.)
A few final thoughts:
SpaceX has a massive warchest. About $100 billion when you add up cash, equivalents, and marketable securities. Yes, it has nearly $40 billion worth of “debt and finance leases,” but that still leaves it with dozens of billions of dollars worth of cash to invest. The company is capitalized very well; its IPO was a success, raising nearly $86 billion worth of capital for the company.- Not that it can continue to spend $18.4 billion per quarter on capex in perpetuity sans greater internal cash generation (+$3.47 billion H1 2026, up from +$351 million H1 2025).
- It may need more money to build towards 10 gigawatts of compute by the end of 2027, with Musk providing a 5-10 gigawatt target for next year. But those expenses won’t come sans demand, so they will presumably be easier to fund in Q2 2027 than, say, Q3 2026.
- Specifically asked about the capex picture of SpaceX, and what it might need to source for its future plans (Terrafab, etc), the company reiterated its quick payback period for certain spend (data centers), and noted that those investments are different from its outlays for Starship; Musk also said that his company’s forecasted ramp to $1 trillion in revenue was previously expected in 2031, a date that has fallen to 2030. I presume the gist was that expected revenue growth will lower any need for external capital.
Neocloud revenues will expand: During its earnings call, SpaceX said that it expects the supply/demand situation (too much of the latter, not enough of the former) to stay where it is, which means that the company can continue to wring attractive margins out of its compute deals. This is the single most important growth vector for SpaceX.Starlink enterprise > Starlink consumer: The company was incredibly bullish on its future selling satellite Internet to corporations and governments, with Musk stating that he expects the enterprise side of Starlink to surpass consumer revenues. There are, in fact, a lot of ships and planes in the world. Presumably those ARPUs will be tasty.Moon factories: During its call, the company stressed that it is going to build facilities on the Moon. I volunteer to service them so long as it comes with a spacesuit.