SpaceX Is Now an AI Company That Owns Rockets — and Last Quarter, the Rockets Lost Money
x/money · 
SpaceX Is Now an AI Company That Owns Rockets — and Last Quarter, the Rockets Lost Money
Published: October 21, 2026 | Reading Time: ~11 minutes | Channel: Money
SpaceX just posted its first meaningful quarter as a public company: $14.58 billion in revenue, up 161% year over year, crushing the $12.72 billion consensus, with net income of $4.4 billion against a $1.05 billion loss a year ago.¹ ² The stock jumped more than 8% premarket.¹ ³ Cue the confetti.
Now the fine print. The segment that built this company's legend — rockets — generated $1.77 billion of revenue and posted a $7 million operating loss.¹ ² The segment that actually out-earned everything else combined is an AI compute business SpaceX quietly rebranded as "Super Intelligence," which did $7.56 billion in quarterly revenue, up 830%.¹ ² Read that again: at the world's most famous space company, selling compute now out-earns selling space — $7.56 billion versus $7.02 billion for launch and Starlink combined. The rocket company is, by revenue, an AI company with a very expensive rocket collection attached.
This matters beyond one ticker. It's the cleanest window yet into where the AI trade's money actually flows, what "profitable AI" really looks like on a GAAP statement, and — if you're holding anything adjacent to compute — how fragile the whole structure is at a 5.12% ten-year yield.⁴ Let's dig in.
The Quarter That Changed What SpaceX Is
First, the basics. SpaceX IPO'd on June 12, 2026, opening at $150 a share. Tuesday's close: $147.79, down 0.11% — yes, the stock that just beat the Street by 25 cents is still trading below its IPO open four months later, despite a market cap of roughly $2 trillion.¹ ³ After hours, it rallied as high as $159.85, up 8.16%.³ ⁴
Here's the scoreboard against estimates:
| Metric | Actual | Consensus | Result |
|---|---|---|---|
| Revenue | $14.58B | $12.72B (LSEG) | Beat by ~15%¹ |
| EPS (GAAP, diluted) | $0.35 | $0.10–$0.15 (provider-dependent)² ⁴ | Beat |
| Adjusted EPS | $0.45 | $0.13 (LSEG) | Beat by ~3.5x¹ |
| Adjusted EBITDA | $8.8B (+306%) | $6.59B (Bloomberg consensus)⁴ | Beat |
| Operating income | $2.8B | — | vs. −$273M a year ago² |
| Net income | $4.4B | — | vs. −$1.05B a year ago¹ ² |
| Capex | $19.59B | $19.6B est. | In line — relief rally fuel⁴ |
One quarter ago this was a loss-making private curiosity. Now it's printing $4.4 billion of quarterly net income. The bulls will tell you this proves the AI trade's economics are real. They're half right, and the half they're wrong about is where your money is at risk. But first, the segment math — because that's the real story.
By the Numbers: Three Companies in One Ticker
| Segment | Q3 2026 Revenue | YoY | Operating Income | The Fine Print |
|---|---|---|---|---|
| Super Intelligence (AI) | $7.56B | +830% (from $813M)¹ ² | +$431M (vs −$1.7B)² | Incl. $1.2B Cursor stock-comp; ~5.7% op margin² |
| Connectivity (Starlink) | $5.25B | +69% (from $3.1B)¹ | ~$2.37B (roughly doubled)¹ | ~85% of total operating income |
| Space (Launch) | $1.77B | +6% (from $1.67B)¹ | −$7M (vs +$259M)¹ ² | Yes, a loss. Starship is expensive. |
| Total | $14.58B | +161% | $2.8B | Net income flattered by $1.0B other income + $390M tax benefit² |
Five things the segment table tells you that the headline doesn't:
1. AI is now 52% of the company. Super Intelligence went from 15% of revenue a year ago to 51.8% this quarter — driven by cloud services agreements, the $60 billion Cursor acquisition (closed August 14, contributing $608 million in its first six weeks), and Grok adoption.² SpaceX is now renting compute capacity to Google, Anthropic, and others.¹ Your rocket company's biggest customer category: other AI labs.
2. Starlink is the profit engine, not the story. Connectivity threw off roughly $2.37 billion of operating income on $5.25 billion of revenue — a ~45% segment margin, funding everything else.¹ Subscribers hit 13.6 million (from 12 million), ARPU rose to $69 from $66, and SpaceX added 1.6 million consumer subscribers in a single quarter across 174 markets.¹ ⁴ The boring satellite internet business is the only thing reliably making money here.
3. The rockets lost money. The Space segment swung from a $259 million operating profit to a $7 million loss while revenue grew 6%.¹ ² Flight 14 finally put Starship in orbit and deployed 26 Starlink V3 satellites — 26 terabits per second, the equivalent of ten Falcon 9 launches in one go — but "revolutionary" and "profitable" are different words, and last quarter only one of them applied.¹ ⁴
4. Net income is softer than it looks. Of that $4.4 billion headline: $1.0 billion was "other income" and $390 million was an income tax benefit — roughly 32% of the bottom line arrived without the business operating.² Pre-tax income was $4.0 billion. Real money, genuinely impressive money — but not the number the press release wants you to marinate in.
5. The capex curve is vertical. Q3 capex was $19.6 billion — up from $6.9 billion a year ago, with $16.4 billion going to AI.² And CFO Bret Johnsen guided Q4 capex to $35–40 billion.⁴ That's a 79–104% sequential increase. Debt and finance leases were already $36.8 billion as of Q2, up from $22 billion the quarter before.¹ Nameplate compute reached 1.7 gigawatts at quarter-end, with management targeting >2 GW exiting Q4 and 200,000 Nvidia GB300 systems coming online per month through year-end.⁴ ⁵
Why "SpaceX Won the Quarter" Is the Wrong Takeaway
The consensus narrative writes itself: legendary company, blowout beat, AI demands are infinite, buy everything with a launchpad. Here's what that story airbrushes out.
The guidance is where the wheels come off the narrative. Johnsen guided Q1 2027 revenue to $33–37 billion against a $22.2 billion consensus — that's 49% to 67% above the Street.¹ That's not guidance; that's a dare. And Musk went further, claiming SpaceX's annual recurring revenue will hit $500 billion to $1 trillion by the end of 2027.¹ ⁴ Meanwhile, actual analysts — the people whose models get marked to reality every quarter — project full-year 2027 revenue of about $110 billion.¹ Musk's low end is 4.5x the consensus. His high end is 9x. When a CEO's target is an order of magnitude above the entire analyst community, one of two things is true: he sees something nobody else sees, or he's anchoring the market's imagination so far above the numbers that $110 billion eventually reads as a disappointment. History generously offers both explanations.
The AI trade's financing math got hostile — and SpaceX just went all-in anyway. The ten-year Treasury sits at 5.12%.⁴ Look down the neocloud food chain: CoreWeave and Nebius are down 28% in October, IREN has plunged about 35%, and Australian operator Firmus withdrew its IPO citing market conditions.¹ SpaceX is "far less reliant" on third-party cloud revenue than those firms, as CNBC put it — true. But it's about to spend $35–40 billion in one quarter building compute infrastructure partly financed with $36.8 billion of debt and finance leases, at the exact moment the bond market is repricing every long-duration cash flow on Earth. The company holding $51.3 billion in cash can survive this math.² The valuation attached to it might not.
The lockup clock is ringing this week. Per Evercore ISI's Kutgun Maral, 1.30 billion shares unlock roughly two days after earnings (October 22), another 329 million on October 24, with the remaining 857 million tranche of the 180-day group thereafter — and the company moved its earnings date up with unusual haste, which Maral flagged as possibly timed "ahead of further financing arrangements" or to let the market digest before those releases.⁴ Against roughly 13.5 billion shares outstanding (a $2 trillion market cap at $147.79), this week's two unlocks alone represent on the order of 12% of the float arriving in four trading days. That's not a technicality. That's a supply event.
The merger chatter is the tell, not the thesis. Musk said a Tesla–SpaceX merger would be "a good idea," confirmed both boards have formed special committees of independent directors (he's excluded as a conflicted party), and noted Tesla's side needs approval from 50% of ALL outstanding shares — not just votes cast — which he called "quite a high hurdle."¹ ⁴ Cathie Wood's Ark Invest projects a combined entity would do $3.8 trillion in sales and $2.7 trillion in operating profit by 2031.⁴ Against Street estimates of roughly $43 billion (SpaceX FY26) and $107 billion (Tesla FY26) in revenue, Ark's five-year number requires a compounding rate I'd describe as "aggressive" only if I were being paid by the word. Note what a merger actually solves for: Tesla's cash flows help fund SpaceX's orbital data center ambitions — the bullish case for the merger is the bearish case for the standalone capex burden, printed on letterhead.
What This Means For You
You don't need to trade SPCX to learn from it. Five moves, in order of urgency:
1. If you own SPCX, respect the calendar, not the premarket. The stock is up ~8% premarket on a beat — into a week where 1.6+ billion shares unlock by Friday.⁴ Classic playbook: if you're a holder with gains, the unlock window is when supply arrives; decide now what you'll do if the stock retests the August low of $104.83³ rather than improvising at 10:47 AM. If you're a buyer, waiting for the unlock to clear costs you a few points of upside and removes the single largest known supply overhang. Patience is a position.
2. Watch two numbers tonight that aren't SpaceX's. Tesla reports after Wednesday's close (consensus ~$27.9 billion revenue, LSEG) and Meta reports too — with Musk's merger comments live, a Tesla number that wobbles drags the whole Musk-complex sentiment trade with it.¹ If you own any of TSLA, SPCX, or Musk-adjacent names, tonight is a correlated-risk event, not three independent events. Size accordingly.
3. Steal the KPI stack, ignore the ARR. The verifiable metrics that will actually mark SpaceX's story to market: GW of compute deployed (>2 GW exiting Q4; Deutsche Bank's Edison Yu models ~6 GW exiting 2027 at the low end of management's 5–10 GW range), Starship flight cadence (weekly targeted H1 2027), and whether Q4 capex lands at $35–40 billion or blows through it.⁴ ⁵ When management offers a $500B–$1T ARR target, mentally file it under marketing and track the gigawatts instead. This is also a free lesson for any operator building AI infrastructure: Johnsen's disclosure that compute is pricing slightly above the high end of $30–50 per watt with sub-one-year paybacks⁴ is the single best public benchmark for your own build-vs-buy math.
4. Audit your own portfolio for "SpaceX risk" — heavy capex financed at 2022-era assumptions. Any company whose story depends on spending billions now for revenue later is repricing daily at a 5.12% ten-year. CoreWeave, Nebius, and IREN didn't fall 28–35% in a month because AI demand died; they fell because their cost of capital didn't.⁴ Check the duration of what you own: rising-rate sensitivity hides in capex-heavy balance sheets long before it shows up in revenue.
5. Don't confuse segment mix with moat — yet. The contrarian bull case is actually the Starlink case: 13.6 million subscribers, $69 ARPU growing, ~45% segment margins, spectrum bought, an FCC-approved 15,000-satellite direct-to-cell constellation, and 3,800+ aircraft wired for American Airlines-class contracts.¹ That's a moat. The AI compute business has demand (real, contracted, accelerating) but competes in a capital arms race against hyperscalers with triple-A balance sheets. If you're underwriting SPCX, underwrite the internet business; treat the compute business as an option you're getting for free — or not.
⚠️ The Risks Nobody's Talking About
- Financing risk at scale. Q4 capex guidance of $35–40 billion against $36.8 billion of existing debt and finance leases, with the ten-year at 5.12%, means either equity issuance (dilution on top of lockup supply), more debt (margin compression the moment AI pricing cracks), or miracle-level cash conversion. The neocloud unwind — CoreWeave −28%, IREN −35%, Firmus IPO pulled — is the market stress-testing exactly this structure.⁴
- Guidance credibility risk. A Q1 2027 guide 49–67% above consensus and a $500B–$1T ARR claim leave zero room for softness: if Q1 lands at, say, $30 billion — which would be spectacular growth — it still reads as a miss against the company's own framing. Management has anchored expectations where only perfection pays.¹ ⁴
- Supply and governance overhang. ~1.63 billion shares unlock by October 24, with 857 million more in the 180-day tranche — into a stock still below its June IPO open.¹ ³ ⁴ Add a potential Tesla merger requiring 50% of all outstanding Tesla shares (months of headline risk, committee leaks, and arbitrage churn) and the tape has two structural noise machines running through year-end.¹ ⁴
- Regulatory and geopolitical exposure. A federal judge let the NAACP's Clean Air Act suit proceed over the methane turbines powering SpaceX's Memphis "Colossus" data centers; the Wall Street Journal reports Musk personally ordered Starshield speed restrictions that impeded Ukrainian military operations; and Starlink's expansion is hostage to regulatory goodwill in 174 markets.¹ When one founder is simultaneously the largest shareholder, the key salesman to governments, and a political actor, the company's risk register is a personality profile.
🎯 The Bottom Line
SpaceX just proved a hybrid model can work: a cash-generative satellite ISP funding a compute land-grab, with legendary rockets as the brand and (for now) the loss leader. That's real — the beat was real, the Starlink margins are real, the contracted AI demand is real. But the stock's own price action is telling you the market hasn't decided whether this is the next megacap or the most elaborate financing story ever launched: four months post-IPO, it still trades below its opening print, with a tenth of its float arriving this week. Watch the gigawatts, honor the calendar, and never — ever — underwrite a CEO's ARR the way you'd underwrite a CFO's revenue line.
📚 Verified Sources
- CNBC — "SpaceX (SPCX) Q3 2026 earnings report: Live updates" (Lora Kolodny, Ari Levy): consensus beats, segment revenue and operating income, merger committee comments, NAACP ruling, Terafab timeline, lockup history. https://www.cnbc.com/2026/10/20/spacex-spcx-earnings-live-updates-q3-2026html.html
- Quartz — "SpaceX reported $4.4 billion net income as revenue rose 161% to $14.6 billion" (Anthony Lopopolo): GAAP detail — EPS $0.35 vs −$0.36, other income, tax benefit, capex, 1.7 GW compute, $51.3B cash. https://qz.com/spacex-earnings-revenue-net-income-super-intelligence-starlink-102026
- MarketBeat — "SpaceX (SPCX) Q3 2026 Earnings Results & Call Transcript" (Quartr transcript): EPS $0.35 vs $0.10 consensus, +161.4% YoY, Q4 capex guide $35–40B, Musk remarks on GB300 ramp, GW targets, and mass-to-orbit share; after-hours pricing. https://www.marketbeat.com/earnings/reports/2026-10-20-spacex-stock/
- Yahoo Finance — "SpaceX earnings Live: SpaceX beats on earnings, capex comes in less than expected" (Pras Subramanian, Brian Sozzi): Bloomberg consensus, Johnsen compute pricing and payback quotes, Evercore ISI lockup schedule, Deutsche Bank GW targets, Ark Invest merger note, Tesla consensus, market data panel (10-yr 5.12%, futures). https://finance.yahoo.com/markets/stocks/live/spacex-earnings-live-updates-133218212.html
- SEC EDGAR — SpaceX 8-K earnings release, filed October 20, 2026 (filing existence and date confirmed via Finnhub SEC filings feed; SEC blocks automated scraping of the document itself). https://www.sec.gov/Archives/edgar/data/1181412/000162828026066482/spcx-20261020.htm
All claims verified against Gold-tier (SEC EDGAR filing confirmation) and Silver-tier (CNBC, Quartz, Yahoo Finance, MarketBeat/Quartr transcript) sources. Each source URL was scraped and confirmed accessible on October 21, 2026; figures marked with provider names (LSEG, Bloomberg, StreetAccount) reflect each outlet's cited consensus. Derived calculations (segment shares, margins, float percentages) are labeled as analysis. Last verified: October 21, 2026.
The rockets were never the business. They were the demo. 🎯