Start by killing the number that has been circulating since Tuesday evening: SpaceX did not grow profits by 92%. Revenue grew 92%, to $7.81 billion. The company still lost money — a net loss of $541 million, per the results filed with the SEC on 4 August 2026. Adjusted EBITDA nearly tripled to $3.54 billion, revenue beat the FactSet consensus of roughly $6.9 billion by almost a billion dollars, and Starlink subscribers doubled to 12 million. And SPCX still fell 7.6% in after-hours trading, extending to a 10.9% decline by Wednesday’s pre-market at $111.70. The reason sits in one line of the release: capital expenditure of $18.37 billion for a single quarter, against $10.11 billion in the prior quarter. SpaceX spent 2.4 times its revenue building things. That is the whole story, and it is neither a growth story nor a profit story. It is a timing story.
Here is the framing that the coverage has missed. The question facing SPCX is not whether SpaceX grows — it obviously does — nor whether the loss narrows, which it already has, by $467 million year on year. The question is whether the capex curve converges before the cash does. And when I rebuilt the segment tables from the 8-K rather than reading the headline, the answer turns out to be hiding in plain sight: of the $14.83 billion gap between what SpaceX spent on capex and what it earned in adjusted EBITDA last quarter, $14.68 billion — 99.0% of it — came from one segment. The AI business. Connectivity, meanwhile, generated $2.60 billion of adjusted EBITDA against $1.37 billion of capex, throwing off $1.23 billion of surplus cash in the quarter. One segment has already converged. One is nowhere near. The bull case and the bear case are the same arithmetic read at different speeds, and that is why this stock can be worth $190 or $76 depending on a single variable.
Key facts: SpaceX Q2 2026 at a glance
- Revenue $7.81bn, up 92% from $4.07bn — beating consensus of ~$6.9bn — SpaceX Q2 2026 results, SEC Form 8-K EX-99.1, 4 Aug 2026
- Net loss $541m (−$0.09 per share), narrowed from a $1.01bn loss a year earlier — SEC Form 8-K EX-99.1, 4 Aug 2026
- Adjusted EBITDA $3.54bn, up 191%; loss from operations $143m — SEC Form 8-K EX-99.1, 4 Aug 2026
- Capex $18.37bn in Q2 vs $10.11bn in Q1 and $2.83bn a year earlier; $15.83bn of it in AI — SEC Form 8-K EX-99.1, 4 Aug 2026
- Connectivity is the only profitable segment: operating income $1.66bn, up 79%; Space lost $542m and AI lost $1.26bn — SEC Form 8-K EX-99.1, 4 Aug 2026
- Starlink subscribers 12.0m, doubled year on year, but ARPU fell 22.4% to $66/month from $85 — SEC Form 8-K EX-99.1, 4 Aug 2026
- $100.0bn of cash and marketable securities against $39.4bn of debt; backlog $47.5bn — SEC Form 8-K EX-99.1, 4 Aug 2026
What actually happened, and why a sell-off followed a blowout
Think of SpaceX right now as a utility that owns a construction company. The utility — Starlink, filed under the Connectivity segment — is mature, profitable and compounding. The construction company is building an AI compute estate at a rate that would embarrass a national grid operator. Both sit inside one income statement, and last quarter the construction company won.
The bridge from a $3.54 billion adjusted EBITDA to a $541 million net loss is worth walking, because it explains why “EBITDA nearly tripled” and “the company lost money” are both true. Depreciation and amortisation of $2.85 billion consumed 80.5% of adjusted EBITDA on its own. Share-based compensation of $831 million took another 23.5%. Together they account for 104% of EBITDA, which is exactly how you arrive at a $143 million operating loss. Below that line, net interest expense of $289 million, $86 million of other expense and $23 million of tax complete the walk to $541 million. Every one of those figures reconciles precisely against the filing.
That D&A line is the crux, because it is the delayed echo of capex. Rockets, satellites and GPUs get bought once and expensed over years. AI segment depreciation has already climbed from $811 million to $1.885 billion year on year, a 132% increase — and that reflects assets bought before this quarter. The $15.83 billion SpaceX spent on AI infrastructure in Q2 alone has barely begun depreciating. On a five-year life, that single quarter of spending adds roughly $791 million per quarter of new depreciation once the kit is in service; on a four-year life, closer to $989 million.
Set that against what the AI segment actually earns. It delivered its first positive adjusted EBITDA quarter at $1.146 billion. Its depreciation on assets already in service is $1.885 billion. The segment therefore covers only 0.61 times its own depreciation today, and Q2’s build alone pushes the bar to roughly $2.68 billion — meaning AI EBITDA must rise 2.3 times simply to break even at the operating line on what has already been bought. Management guided Q3 and Q4 capex to be similar to Q2. Two more quarters at that pace would add another $1.6 billion per quarter of depreciation on top.
None of which was framed as a problem by the company. “Revenue growth accelerated across all our business segments and we delivered strong operating leverage, with significant margin expansion led by our new AI compute agreements,” wrote chief financial officer Bret Johnsen in the commentary accompanying the results. He is not wrong — operating leverage did improve. The market’s objection is that the denominator is growing faster.
What SpaceX and the Street are actually doing about it
SpaceX’s response to a capex-driven sell-off was, remarkably, to announce more capex. On 5 August the company unveiled a partnership with Nvidia to build “Starmind” orbital AI data centre satellites, each carrying Rubin GPUs and Vera CPUs and drawing roughly 120 kilowatts of sustained compute power at altitudes between 500 and 2,000 kilometres. As FinanceFeeds reported on the Starmind announcement, SpaceX has filed with the FCC for authority covering up to one million such satellites. At 120kW each, a fully built constellation would represent about 120 gigawatts of compute — roughly 86 times the 1.4GW of nameplate capacity SpaceX operates today.
Read one way, that is a company doubling down the morning after being punished. Read another, it is the only coherent answer to the depreciation problem: terrestrial AI data centres are constrained by power and cooling, and orbit offers continuous solar and radiative cooling. The bull and the bear will cite the same press release.
The sell-side had, to its credit, named the tripwire in advance. Morgan Stanley’s Adam Jonas carried an Overweight rating and a $300 target into the print while warning that capex materially above his roughly $50 billion 2026 estimate would pressure profitability. First-half capex is already $28.48 billion. If the second half matches guidance of roughly two more Q2-sized quarters, full-year capex lands near $65.2 billion — about 30% above the level Jonas flagged. The most bullish framework on the Street contained its own disconfirmation trigger, and SpaceX tripped it.
The broader analyst picture has not yet caught up. MarketBeat’s compilation of 39 analysts shows a consensus target of $230.50, with a high of $800 and a low of $115 — and SPCX in Wednesday’s pre-market was trading below every published target on the Street, including the two $115 lows from HSBC’s Nicolas Cote-Colisson and CFRA’s Keith Snyder. That is less a signal of value than a signal that targets are stale. Expect revisions.
Operationally, management leaned on the segment that works. “We had an exceptional second quarter,” said president and chief operating officer Gwynne Shotwell on the call, noting the company “added net more than 1.7 million Starlink subscribers globally, consumer.” That is the part of SpaceX nobody is arguing about.
The number that decides this: capex intensity, by segment
Capex intensity — capital spending as a percentage of segment revenue — is the cleanest way to see convergence happening or not happening. Connectivity’s has fallen from 43.7% a year ago to 40.9% in Q1 to 31.9% last quarter, while its revenue grew 66% year on year. Capex rose just 2.6% sequentially, from $1.332bn to $1.367bn, while revenue rose 31.7%. That is a converged capex curve, and it is not a theory: SpaceX has already done this once, with a satellite constellation, which is not an obviously easier thing to build than a data centre.
AI’s capex intensity is 618%. It is spending 19.4 times as much capital per dollar of revenue as Connectivity does. Connectivity earns 2.06 times its own depreciation; AI earns 0.61 times. The entire investment debate reduces to how quickly the second number travels toward the first.
| The bull case | The bear case |
|---|---|
| Connectivity proves SpaceX converges capex curves: intensity fell 43.7% → 31.9% in a year | Connectivity took a decade to get there; AI is at 618% and rising |
| AI hit its first positive adjusted EBITDA quarter (+$1.15bn, from −$609m in Q1) | AI still covers only 0.61× its own depreciation, before Q2’s $15.83bn starts depreciating |
| $14.1bn of new Cloud Services Agreements signed; $47.5bn total backlog | Only $1.6bn of that was recognised in Q2 — growth from here needs new contracts, not this one |
| $100.0bn of cash and securities; $85.7bn IPO proceeds and a $25bn bond already banked | Net cash of $60.6bn funds roughly four more quarters at Q2’s $14.83bn gap |
| Starlink subscribers doubled to 12m; enterprise and government revenue up 108% | ARPU fell 22.4% to $66 — the profitable business monetises each user less every year |
That cash figure deserves care, because it is the most abused number in the bull case. SpaceX has $100.0 billion of cash and marketable securities, which sounds unassailable. Net of $39.4 billion of debt, it is $60.6 billion. At Q2’s gap between capex and EBITDA, that funds about 4.1 quarters; gross cash funds about 6.7. This is emphatically not a solvency question — operating cash flow was a positive $3.47 billion in the first half, and a company with SpaceX’s access can raise more or simply slow the build. It is a dilution-and-leverage question, and it has a date attached. FinanceFeeds has been tracking the same tension since the stock first slipped below $115 in late July.
The supply problem arriving on Thursday
There is a second, entirely mechanical force acting on SPCX this week, and it is arguably larger than the earnings themselves. On 6 August, up to 911.5 million shares become eligible to trade as the first tranche of the post-IPO lock-up releases — timed, as FinanceFeeds noted, to land 48 hours after the first earnings report. That is 6.9% of the 13.18 billion shares outstanding, worth roughly $102 billion at Wednesday’s pre-market price.
The number that matters more is the comparison nobody is making. SPCX’s tradeable float is roughly 646 million shares — the IPO sold 638.9 million. The unlock is therefore 1.41 times the entire current float. And because this tranche represents only 20% of the standard lock-up pool, the full pool implies roughly 4.56 billion shares, or 34.6% of the company — about seven times the float — releasing in stages.
Eligibility is not selling; employees and early investors decide individually, and most do not liquidate into a 50% drawdown. But the float mathematics mean even modest participation moves price disproportionately, and it explains how a stock can be simultaneously cheap against consensus and heavy against supply. This is the structural tension in the SPCX story: the same lock-up conventions that protect an orderly IPO create a mechanical overhang precisely when sentiment is weakest. SpaceX now sits roughly 50% below the $225.64 all-time high it reached in June, and 17.3% below its $135 IPO price.
SPCX price prediction: bull $190, base $135, bear $76
Because SpaceX’s current earnings cannot support its valuation on any conventional multiple — enterprise value is about 45 times annualised Q2 revenue — the honest way to set levels is against the company’s own forward guidance. Johnsen told the call that a $100 billion annualised revenue run-rate is within reach by December, helped by cloud contracts and the pending $60 billion Cursor acquisition. Every level below is simply a multiple of that guided run-rate applied to 13.18 billion shares. At $111.70, the market pays 14.7 times.
| Case | Level | Move | Implied multiple of guided run-rate | Reasoning |
|---|---|---|---|---|
| Bull | $190 | +70% | 25× | AI capex intensity starts falling as Connectivity’s did, and Q3 brings new Cloud Services Agreements comparable to Q2’s $14.1bn |
| Base | $135 | +21% | 17.8× | A round trip to the IPO price: the unlock is absorbed, capex plateaus rather than falls, and the December run-rate is met |
| Bear | $76 | −32% | 10× | Capex stays near $18bn while AI revenue growth decelerates; the market stops paying for 2030 and pays for December 2026 |
Note where this sits relative to the Street: our bull case of $190 is 18% below the $230.50 consensus. That is deliberate. Consensus targets were largely set before anyone saw an $18.37 billion capex quarter, and Morgan Stanley’s $300 framework reportedly attributes $152 of it — more than half — to the enterprise AI business, which is precisely the segment that just consumed 99% of the group’s cash gap and still loses $1.26 billion a quarter at the operating line.
The bear case is not a bet against SpaceX. It is a bet that the market re-rates a company from a 2030 story to a 2026 story for two or three quarters, which is something equity markets do routinely to capital-intensive compounders. Musk’s own answer to the monetisation worry is characteristically wide-angle: “Even if our monetization per bit dropped by a factor of 10, that would still mean a 10x increase in the revenue of Starlink,” he told the earnings call, per Fortune’s account. He also said internal projections for reaching $1 trillion in revenue “have moved up from 2031 to 2030,” with “a non-zero chance of that being in 2029.” If that proves right, $190 is far too low. The bear case simply observes that shareholders have to survive the interval.
What would prove each case wrong
The bull case fails if: Q3 contracted sales come in materially below the $14.1 billion of Cloud Services Agreements signed in Q2, revealing that quarter as a one-off rather than a run-rate — only $1.6 billion of the $14.1 billion was recognised in Q2, so the remaining ~$12.5 billion recognised over time is flat revenue, not growth. It also fails if Connectivity capex intensity climbs back above 40% of revenue as V3 satellite production scales, or if Starlink net additions fall below roughly 1.2 million in a quarter after Q2’s 1.7 million.
The bear case fails if: Q3 capex comes in meaningfully below $15.8 billion in the AI segment despite guidance for “similar” spending — a downshift would signal discipline and reset the entire cash-gap arithmetic. It also fails if the AI operating loss narrows below roughly $600 million from $1.26 billion, or if the 6 August unlock passes with SPCX holding above the $104.83 post-IPO low on normal volume, which would show the float can absorb supply.
The base case fails in either direction the moment one of the above triggers fires — most likely at the Q3 report, the first print in which depreciation from this quarter’s $15.83 billion of AI capex begins showing up in the numbers.
Frequently asked questions
Did SpaceX’s profits rise 92% in Q2 2026?
No. This is the most common error in circulation. Revenue rose 92% to $7.81 billion. SpaceX reported a net loss of $541 million for the quarter, narrowed from a $1.01 billion loss a year earlier. Adjusted EBITDA — a non-GAAP measure that excludes depreciation, share-based compensation, interest and tax — rose 191% to $3.54 billion. The company was not profitable on a GAAP basis.
Why did SPCX stock fall if revenue beat expectations by $1 billion?
Capital expenditure. SpaceX spent $18.37 billion in Q2, up from $10.11 billion in Q1 and against analyst forecasts nearer $13 billion. Of that, $15.83 billion went into AI compute infrastructure. Investors read an annualised capex run-rate of roughly $73.5 billion against annualised adjusted EBITDA of about $14.2 billion and repriced the timeline to free cash flow, not the growth rate.
What is the SpaceX share price forecast for 2026?
Our levels are $190 bull, $135 base and $76 bear, derived as 25×, 17.8× and 10× the company’s guided $100 billion December revenue run-rate across 13.18 billion shares. Street consensus is higher at $230.50 across 39 analysts, though those targets largely predate the Q2 capex figure. SPCX traded at $111.70 in pre-market on 5 August 2026. None of this is investment advice.
How big is the SpaceX share lock-up expiry?
Up to 911.5 million shares become eligible to trade from 6 August 2026, worth roughly $102 billion at current prices and equal to about 1.41 times the existing float of roughly 646 million shares. That tranche is 20% of the standard lock-up pool, implying a full pool near 4.56 billion shares. Eligibility to sell is not the same as selling.
Is Starlink profitable on its own?
Yes, and it is the only SpaceX segment that is. Connectivity generated $1.656 billion of operating income on $4.291 billion of revenue in Q2 2026, a 38.6% operating margin, up 79% year on year. It also generated $2.597 billion of adjusted EBITDA against just $1.367 billion of capex — a surplus of $1.23 billion in the quarter. Space lost $542 million and AI lost $1.257 billion at the operating line.
What should investors watch in SpaceX’s Q3 2026 results?
Three lines, in order: AI segment capital expenditure (guided “similar” to Q2’s $15.83 billion — any downshift is the single most bullish possible datapoint), new contracted sales beyond the $14.1 billion of Cloud Services Agreements, and AI segment depreciation, which will begin absorbing this quarter’s build. Starlink net additions and ARPU are the check on whether the profitable segment stays profitable.
This article is analysis and information, not investment advice. Price levels are the author’s estimates derived from company filings and are not forecasts of actual returns. Figures are drawn from SpaceX’s Q2 2026 results filed with the SEC on 4 August 2026, with price data from Investing.com and StockAnalysis, and consensus targets from MarketBeat, all accessed 5 August 2026.