SpaceX Q2 2026: revenue grew 92%, and capex was 2.4 times revenue

Twenty-four years after it was founded and seven weeks after it went public, SpaceX filed its first quarterly report as a listed company. It beat both of the estimates CNBC published from LSEG.

	Q2 2026	LSEG estimate (via CNBC)
Revenue	$7.81B	$6.93B	beat
Loss per share	$(0.09)	$(0.26)	beat
Adjusted EBITDA	$3.54B	—	was $1.21B a year earlier
Net loss	$541M	—	Q1 2026: $4.28B (our arithmetic)

Revenue grew 92% year over year, from $4.07 billion. Adjusted EBITDA grew 191%. The net loss narrowed by about $3.7 billion from the first quarter (our arithmetic: the six-month net loss of $4,817 million less Q2’s $541 million puts Q1 at $4,276 million).

The line that matters most is not in the table above. It is further down, in the cash flow statement.

Did SpaceX issue guidance for Q2 2026?

No — no Q2 2026 guidance appears in any SEC filing. Its IPO prospectus (Form 424B4, filed June 12, 2026) contains no revenue or earnings outlook, and neither the Q2 results release (Exhibit 99.1 to the Form 8-K filed Aug. 4, 2026) nor the Q2 Form 10-Q contains a guidance table or a revenue outlook; the forward-looking remarks quoted further down were made on the call and are attributed to the outlets that reported them. Against the year-ago quarter, revenue rose 92% to $7,814 million from $4,071 million, the net loss narrowed to $541 million from $1,008 million, and the loss per share narrowed to $(0.09) from $(0.34).

$18.37 billion

That is what SpaceX spent on capital expenditure in the three months to 30 June — against $7.81 billion of revenue. Capital expenditure was 2.4 times everything the company sold (our arithmetic). Of it, $15.83 billion went to the AI segment: data centres, power, compute.

Put the halves of the year side by side and the shape is clearer still.

Six months to 30 June 2026
Net cash provided by operating activities	$3.47B
Purchases of property, plant and equipment	$28.48B
Implied free cash flow	roughly −$25.0B (our arithmetic)

The company ended the quarter with $100 billion of cash, cash equivalents and marketable securities. Most of it came from two June financings: an IPO that brought net proceeds of about $85.7 billion, and a $25 billion investment-grade bond issue that closed on 26 June.

What comes next on capex is not in the filings. CNN reported that analysts expect SpaceX to top $45 billion of capex for the year. With $28.48 billion already spent in the first half, a full-year figure just above $45 billion would leave roughly $16.5 billion for the entire second half (our arithmetic) — less than Q2 alone. Either the outside expectation assumes a sharp slowdown from the Q2 pace, or it is low. The Q3 capex line will show which.

That is the tension in the quarter: a beat on both published estimates, alongside the cost of buying that growth.

Adjusted EBITDA is the wrong metric for this company

$3.54 billion of adjusted EBITDA, up 191%, is the headline management leads with, and in most businesses it would be the right one.

Not here. EBITDA excludes depreciation and amortisation — the accounting recognition of capital already spent — and it excludes capital expenditure entirely. For a software company that owns almost nothing, that exclusion is harmless. For a company whose defining financial characteristic is that it spent $18.37 billion on physical assets in ninety days, EBITDA excludes precisely the item that determines whether the business works.

The relationship worth carrying instead:

• Adjusted EBITDA of $3.54 billion covers about 19% of the quarter’s capital expenditure (our arithmetic). The rest comes from the balance sheet.
• Every dollar of revenue was accompanied by $2.35 of capital spending (our arithmetic).
• The net loss of $541 million, impressive as the improvement is, is a poor guide to cash. Quarterly capital spending was about 34 times the net loss (our arithmetic), because most of the spending was capitalised rather than expensed.

None of this makes the spending wrong. Building out compute — 1.4 GW of nameplate capacity at quarter-end, up from 0.4 GW a year earlier, per the release — and a next-generation satellite constellation is capital-intensive by definition, and a company that just raised $85.7 billion is supposed to deploy it. But an investor reading “net loss narrowed to $541 million” and concluding that SpaceX is approaching break-even has read the wrong statement. On a cash basis this was an extraordinarily expensive quarter.

One segment pays for the other two

This is the structure of the business, and it is not the structure the narrative implies.

Segment	Revenue	YoY	Operating income
Connectivity (Starlink)	$4.29B	+66%	+$1.66B
AI	$2.56B	+247%	−$1.26B
Space (launch)	$962M	+29%	−$542M

Connectivity is 55% of revenue (our arithmetic) and the entire profit engine. Starlink — a satellite broadband subscription business — earns $1.66 billion of operating income and funds a combined $1.8 billion of losses in the two segments that generate the headlines.

Some perspective on each:

Connectivity is genuinely strong. 12.0 million subscribers, double the 6.0 million of a year ago, with 1.7 million net additions in the quarter. Operating margin rose to about 38.6% from about 35.7% a year earlier (our arithmetic). The one soft line: average revenue per user was $66 a month, flat on Q1 but down from $85 a year earlier. If SpaceX were only this, it would be a straightforward high-quality growth business.

AI is growing extraordinarily and losing money at scale. Revenue up 247% to $2.56 billion, mostly AI solutions and infrastructure revenue from new cloud services agreements; operating loss of $1.257 billion, though that loss narrowed 49% from the first quarter and segment adjusted EBITDA turned positive at $1.146 billion. Against that segment’s $15.83 billion of quarterly capex, the loss narrowing is the encouraging part and the ratio is the sobering part. CFO Bret Johnsen said on the call that AI compute capital is being deployed at “less than a one-year payback”, per CNBC — a claim, not a disclosure, and one that the segment’s current economics do not yet demonstrate.

Space is the smallest and it loses money. $962 million of revenue, up 29%, and a $542 million operating loss — the release attributes the higher costs to accelerated Starship research and development, expensed against a launch business that is not large enough to absorb it. The rocket company is the smallest part of the rocket company.

The practical consequence: anyone who bought SPCX for spaceflight owns, in revenue terms, mostly a broadband ISP and, in capital terms, mostly a data centre developer. That is not a criticism of the strategy. It is a description of what the revenue and the capital are.

The targets on the record

The filings contain no targets. On the call, management made several forward claims, which the outlets below reported. They are specific, and specific is checkable.

Target	What was said	Reported by
Revenue run rate	On pace for $100 billion of annualized recurring revenue by the end of 2026, assuming a contribution from Cursor (CFO Bret Johnsen)	CNBC
Contracted, post-quarter	An additional $6.7 billion of cloud services revenue contracted in the first weeks of Q3, “over a six month period that begins ramping starting in October” (Johnsen)	CNBC
Compute capacity	Over 2 gigawatts by end-2026, closer to 10 gigawatts by end-2027 (Musk)	CNBC
AI hardware	Data centres built exclusively on Nvidia chips (Musk)	CNBC
Starlink V3	”An order of magnitude more capable” than the current satellite (Musk)	Fortune
$1 trillion revenue	Internal projection moved up “from 2031 to 2030”, with “a non-zero chance of that being in 2029” (Musk)	CNN, Fortune

What the release itself does disclose on this front: in Q2, SpaceX entered into cloud services agreements totaling $14.1 billion in contracted sales, which produced $1.6 billion of incremental AI infrastructure revenue in the quarter, and it agreed to acquire Cursor for $60 billion, expected to close in the third quarter.

Two of the call claims deserve to be held up to the light.

The $100 billion annualized run rate by December. Q2 revenue of $7.81 billion annualizes to about $31.3 billion (our arithmetic). Reaching a $100 billion run rate by December means roughly tripling the quarterly revenue line in two quarters. The stated bridge is the cloud contracts — the filed $14.1 billion and the $6.7 billion the CFO described — plus Cursor. That is a very large gap to close, and it is the most immediately falsifiable claim management made: the Q4 report will settle it.

The $1 trillion by 2030. From roughly $31 billion annualized, that is about 32 times in four and a half years (our arithmetic). For context, TradingKey reported that Morgan Stanley’s analysis put SpaceX’s 2030 revenue at about $330 billion and Goldman Sachs’s at about $474 billion — between a third and a half of the company’s own projection. Musk’s figure is roughly double Goldman’s (our arithmetic). The distance between Musk’s number and the sell side’s is, in effect, the entire investment debate.

Neither observation says the targets are wrong. Both say they are extraordinary, and extraordinary targets deserve to be tracked against quarterly evidence rather than restated as facts.

Backlog, and what it does and does not cover

Backlog finished the quarter at $47.5 billion ($47,461 million in the 10-Q). Against an annualized revenue rate near $31 billion, that is roughly a year and a half of revenue (our arithmetic). The 10-Q gives the timing: about 56% is expected to be recognized within one year, about 34% in one to three years, and the remaining 10% after that.

It is a modest figure for a company targeting a thirty-fold revenue increase. Backlog underwrites the next several quarters; it does not underwrite 2030. The gap between $47.5 billion of signed work and a $1 trillion ambition has to be filled by demand that does not exist on paper yet, principally in AI compute — which is the same demand every other hyperscaler is building for simultaneously.

The $14.1 billion of cloud services agreements signed in Q2, and the further $6.7 billion the CFO described after quarter-end, are the most encouraging evidence that the filling is happening. The 10-Q also states the limit: these agreements generally run on monthly fees and, after an initial ramp period, may be terminated by either party on 90 days’ notice.

The IPO and the 6 August lock-up

SpaceX priced its IPO at $135 a share, offering 555,555,555 shares for net proceeds of about $74.4 billion, or $85.7 billion with the underwriters’ option exercised in full, per the Form 424B4. Trading began on 12 June; the release states that the offering closed on 15 June with 638,888,888 shares sold, for net proceeds of about $85.7 billion.

One further item sits immediately ahead. Per the prospectus, up to 911.5 million shares become eligible for sale on the second full trading day after the first earnings release — 6 August. It is not the last or the largest step: the same schedule releases up to 1.3 billion shares after the Q3 results. Lock-up expiries do not change what a business is worth, but they change how many shares can be sold in a given week.

The bear case, stated fairly

Not “the growth is fake.” It plainly is not. The narrower version:

1. Cash consumption is the whole story. Roughly $25 billion of negative free cash flow in six months (our arithmetic). The $100 billion cash pile is a finite runway with a calculable end, and part of it is borrowed.
2. The profitable segment is not the exciting one. Starlink connectivity carries the company; AI and Space both lose money.
3. The sub-one-year AI payback is an assertion. The segment lost $1.26 billion this quarter on $15.83 billion of quarterly capital deployment. The claim may prove correct — it is not yet demonstrated in the reported numbers.
4. The $100 billion December run rate requires roughly tripling revenue in two quarters. It rests on cloud contract ramps and one acquisition.
5. The 2030 target is roughly double Goldman Sachs’s published estimate, as reported by TradingKey.
6. Backlog is front-loaded. About 56% of the $47.5 billion is expected to be recognized within a year, and the cloud agreements behind much of the AI growth can be terminated on 90 days’ notice after their ramp period.
7. The lock-up schedule releases up to 911.5 million shares on 6 August and up to 1.3 billion more after the Q3 report.

Against that, the bull case is not complicated and should be stated too: 92% revenue growth with 191% EBITDA growth is real operating leverage; connectivity margins are expanding while subscribers double; the AI loss narrowed 49% sequentially and AI segment adjusted EBITDA turned positive; and the company has $100 billion of cash against $39.4 billion of debt and finance leases (our arithmetic: current plus non-current), so no obvious financing pressure over the next several quarters.

What to check at the Q3 report

In rough order of importance:

• Capital expenditure against Q2’s $18.37 billion. If it comes in materially below that, management has chosen discipline over the compute target. If it comes in at or above, the runway shortens — and the analyst expectation CNN reported for the year looks low. This is the number that decides the story.
• The gap between operating cash flow and capex. Free cash flow is the metric this business should be judged on; adjusted EBITDA is not.
• Revenue against the $100 billion annualized December path. Q3 revenue needs to show a visible step change for that claim to survive to the Q4 report.
• AI segment operating loss. It narrowed 49% sequentially. Whether that continues, as $15.83 billion of quarterly investment starts depreciating, tells you whether the sub-one-year payback claim is being borne out.
• Cloud contracts. Whether contracted sales grow beyond the $14.1 billion signed in Q2, and what management says about the $6.7 billion ramp due to start in October — after Q3 closes, so it belongs in the Q4 numbers.
• Starlink net additions. 1.7 million this quarter. Connectivity funds everything else, so its subscriber momentum is the company’s real margin of safety.
• Backlog. Whether $47.5 billion grows faster than revenue, which would mean demand is being signed ahead of delivery, or lags it.
• Depreciation and amortization. It was $2.85 billion in Q2, up from $1.53 billion a year earlier, and $28.48 billion of first-half capex has yet to flow through it in full. The net loss can get worse before the cash flow gets better, and that is arithmetic rather than deterioration.

The read

SpaceX’s first quarterly report as a public company beat both LSEG estimates CNBC published, and the income statement improved dramatically. The cash flow statement showed a company spending 2.4 times its revenue on physical assets in a single quarter.

Those are not contradictory facts. They are the same fact seen from two statements: SpaceX is converting an enormous pile of IPO and bond capital into compute and satellites as fast as it can, on a bet that demand arrives to fill it. If it does, the current numbers will look like a rounding error. If it arrives later or smaller than planned, the company will have committed most of its raised capital before finding out.

For anyone holding SPCX, that reframes what to watch. The revenue beat is not the signal. The signal is the distance between operating cash flow and capital expenditure, quarter after quarter, and whether the AI segment’s losses close at the pace the payback claim requires.

---

Sources and method. Company figures are from SpaceX’s Q2 2026 results release, Exhibit 99.1 to the Form 8-K filed 2026-08-04 (accession 0001628280-26-052515), and the Form 10-Q for the quarter ended 2026-06-30 (accession 0001628280-26-052535), read directly from SEC EDGAR: revenue and its 92% growth, the net loss and loss per share, adjusted EBITDA, segment revenue, operating income and capex, Starlink subscribers, ARPU and net additions, nameplate compute, the $14.1 billion of contracted cloud services sales, the Cursor agreement, cash and marketable securities, the bond issue, operating cash flow, capex, depreciation and amortization, debt, and backlog with its recognition schedule. IPO terms and the lock-up schedule are from the Form 424B4 prospectus filed 2026-06-12 (accession 0001628280-26-042639). None of these filings contains revenue or earnings guidance, so the guidance verdict judges the quarter only against the year-ago quarter; the third-party estimates below are context, not the basis of that verdict.

Not company filings, and attributed accordingly: the revenue and per-share estimates are LSEG figures as published by CNBC; the analyst expectation of more than $45 billion of 2026 capex is from CNN; the CFO’s $100 billion run-rate, $6.7 billion post-quarter contract and payback remarks, and Musk’s compute and Nvidia remarks, are as reported by CNBC from the call; Musk’s Starlink V3 remark is as reported by Fortune, and his $1 trillion remark by CNN and Fortune; the Goldman Sachs and Morgan Stanley 2030 revenue estimates are as reported by TradingKey. This piece makes no claim about the share-price reaction.

Internal consistency was checked against the filing: the three segment revenue lines sum to the reported $7,814 million, and the segment operating results sum to the reported consolidated operating loss of $143 million. Derived figures are labelled “(our arithmetic)”: the roughly −$25.0 billion of six-month free cash flow is operating cash flow less capital expenditure; the 2.4× and 19% ratios, the Q1 net loss, the annualized run rate and the ~32× required growth to $1 trillion are arithmetic on the filed figures. Those are our calculations, not company statements, and none of this is a forecast.

Related: AMD Q2 2026, the chip side of the same buildout — and note that Musk said on the call, per CNBC, that SpaceX’s AI data centres will run exclusively on Nvidia. Also Microsoft Q4 FY2026, where a company spent $41 billion in a quarter on AI.

Tradune shows what you actually hold across every account, and whether a position has beaten a plain index fund since you bought it. Educational analysis, not investment advice.