The setup
The Q2 dive left one question open: could Starlink keep paying for an AI capex bill of $15.83B a quarter? Q3 answered a different one. The AI segment, renamed Super Intelligence (SI), now out-earns Starlink on revenue, and it turned an operating profit.
Revenue was $14.58B, up 161% on the year from $5.58B and 86.5% on the quarter from $7.81B. Income from operations was $2.79B, against a $143M loss in Q2 and a $1.94B loss in Q1. Net income was $4.40B, $0.35 per diluted share, after a $541M loss in Q2. Adjusted EBITDA was $8.81B, a 60.4% margin.
Put the lead number in its series. Q3 added $6.76B of revenue on the quarter. Before Q2 2026, SpaceX never reported a full quarter larger than that: the biggest was $5.58B, in Q3 2025. SI alone added $5.00B — more than SpaceX’s entire Q1 2026 revenue of $4.69B — and went from 32.8% of group revenue in Q2 to 51.9% in Q3. Against the Street, revenue beat an LSEG consensus of $12.72B and operating income beat a $2.33B estimate, per Yahoo Finance and Fortune. What it is not: a clean read on SI profit, because the first 48 days of the Cursor deal sit inside it.
SPCX closed at $159.28 on , up 7.77% from $147.79.
The numbers
| Revenue | 14,575 · +161% YoY · +86.5% QoQ (Q2: 7,814; Q3’25: 5,576) |
| Space · Connectivity · SI | 1,767 · 5,248 · 7,560 |
| Income (loss) from operations | 2,791 (Q2: (143); Q3’25: (273)) |
| Space · Connectivity · SI | (7) · 2,367 · 431 |
| Net income · diluted EPS | 4,404 · $0.35 |
| Adjusted EBITDA (non-GAAP) | 8,809 (Q2: 3,538; Q3’25: 2,169) |
| Capex · of which SI | 19,593 · 16,444 |
| Starlink consumer subscribers · ARPU | 13.6M (Q2: 12.0M) · $69/mo (Q2: $66) |
| Nameplate compute | 1.7 GW (Q2: 1.4 GW; Q3’25: 0.6 GW) |
| Operating cash flow · free cash flow (derived) | 9,631 · (9,962) |
| Cash + marketable securities (derived) | 89,855 (Q2: 100,009) |
| SI operating income before Cursor acquisition charges (derived) | 1,656 (Q2: (1,257)) |
Reported lines from SpaceX’s Q3 2026 earnings release, Form 8-K Ex. 99.1 (). The copper rows are derived: Q3 operating cash flow is the nine-month figure less the first half from the Q2 Form 10-Q; free cash flow subtracts purchases of property, plant and equipment; the SI line adds back $1,217M of acquisition-related stock compensation and $8M of acquisition costs.
Read operating income before net income. Net income exceeded it by $1.61B, so 36.6% of the $4.40B came from below the line: interest income of $873M against interest expense of $675M, other income of $1,025M ($561M of it related party), and a $390M tax benefit where a year ago there was a $562M provision. The profit that recurs is the $2.79B. The rest is the IPO cash pile earning interest and a tax line swinging.
The rest of this dive is for paid subscribers.
The headline numbers are above. The 6 sections below carry the mechanism, the peer read, the valuation work, and the dated tests that decide it.
- What’s actually driving it
- How peers read this
- Valuation reality check
- What we’d watch from here
- The trade
- Bottom line
Already a subscriber? Sign in.
What’s actually driving it
The profit has a stock charge growing inside it. The release: SI income from operations “was $431 million, inclusive of $1.2 billion in share-based compensation from the Cursor acquisition for the quarter.” Decode the calendar. Cursor closed on , so the charge covers 48 days of a 92-day quarter. Add back $1,217M of acquisition stock compensation and $8M of acquisition costs and SI earned $1.66B before the deal, against a $1.26B loss in Q2. If the charge ran evenly, a full quarter is about $2.33B — and the filing says $23.4B of award value is attributable to post-acquisition service, subject to forfeitures. The $431M is a 48-day number. Q4 is the first full quarter of the cost.
The margin moved in the rental, not the deal. Before crediting Cursor, find the segment whose margin actually moved. Cursor brought $608M of revenue. SI adjusted EBITDA rose $3.81B on $5.00B of added revenue, a 76.2% incremental margin; give Cursor every one of its dollars as EBITDA and the rest of SI still added $3.20B. The source is in the Q2 release, which said SpaceX “entered into several Cloud Services Agreements totaling $14.1 billion in contracted sales.” Decode it against the revenue line. SI solutions and infrastructure revenue was $475M in Q1 and $2,194M in Q2; excluding Cursor it was $6.41B in Q3. Those agreements turned on this quarter.
Now divide the dollars by the units the company itself named. Annualized over 1.7 GW of nameplate compute, ex-Cursor solutions revenue is about $15.09B per GW-year, against $6.27B in Q2 on 1.4 GW. Run it back through the prospectus and the ratio fell for a year first: from $3.79B in Q1 2025 to $1.90B in Q1 2026, as GPUs went in faster than customers arrived. Q2 and Q3 are the reversal. That is two quarters — a phase until the series confirms it — and nameplate is a period-end count, so a fast-build quarter understates the average denominator.
Depreciation is the second bill, and it is the one capex writes. SI depreciation and amortization was $811M in Q2 2025, $1,493M in Q1 2026, $1,885M in Q2 and $2,473M in Q3: up $392M and then $588M a quarter. SI capex was $15.83B in Q2 and $16.44B in Q3, 83.9% of the group’s spend, and those GPUs depreciate from the day they are placed in service. So Q4 SI operating income carries two growing costs, not one. As an illustration — assuming D&A rises by Q3’s $588M, the Cursor charge runs at the $2.33B pace, and other stock compensation holds at $821M — SI needs adjusted EBITDA of about $6.21B in Q4 to stay above zero. That is 25.5% above Q3’s $4.95B, after a quarter that added $3.81B.
The capex is running ahead of the gigawatts. Nameplate rose 0.4 GW in Q2 and 0.3 GW in Q3 while SI capex rose. Divide one by the other and Q2 spent about $39.57B of SI capex per GW added; Q3 spent about $54.81B. That is not a cost per GW — capex pays for halls, power and GPUs that are not yet counted — but it says Q3 was a building quarter more than an installing one. Musk said on the call that SpaceX is on track for about 200,000 GB300s online per month for the rest of 2026, per Yahoo Finance, and a Deutsche Bank note puts management’s target at more than 2 GW exiting Q4. Cash spending records decisions already made; the purchase obligations show the ones still running. At June 30 the 10-Q carried $27.96B of unconditional purchase obligations, $22.24B of them due in 2027.
The customers are paying ahead. CFO Bret Johnsen on the call: “There are many customers asking for compute.” Decode it against the balance sheet. Current deferred revenue rose $2.81B in the quarter to $10.78B, and the cash-flow statement credits $2,753M of Q3 operating cash flow to the deferred-revenue line: 28.6% of the $9.63B. Without it, Q3 operating cash flow was $6.88B. Demand is not the debate here; the customers are prepaying for it. Prepaid growth is a customer loan in a cash-flow costume, so read the $9.63B as $6.88B of earned cash and $2.75B of customer money not yet worked off.
Starlink stopped leaking price. The prospectus, in June: “We generally expect Starlink Subscriber ARPU to continue to decline over the next few years.” Decode it against Q3. ARPU went $86, $85, $83 through the first three quarters of 2025, $81 for the year, then $66 in Q1 and Q2 2026. The release now says “Starlink consumer monthly ARPU improved 5% sequentially to $69,” the first sequential rise in the series, “driven by growth in developed markets.” Net adds were 1.4M, 1.7M and 1.6M across 2026’s three quarters, so price rose while adds slowed: the mix moved toward richer markets. It is one quarter, against management’s own guidance, so we read it as mix rather than pricing power. CFO Bret Johnsen said on the call that Connectivity gross margin included a one-time benefit of roughly 3 points, mainly tariff refunds, per Yahoo Finance.
The rest of Connectivity is enterprise. Enterprise and government revenue was $2.13B, up 91.4% on the year, against consumer revenue of $3.12B, up 57.2%. Connectivity operating margin was 35.7% in Q2 2025, 38.4% in Q3 2025, 36.5% in Q1 2026, 38.6% in Q2 and 45.1% in Q3. Most of the Q3 step is real; about 3 points of gross margin is the one-time item. Connectivity capex was $1.54B against $3.50B of segment adjusted EBITDA, so Starlink still funds itself 2.27 times over and produced 84.8% of group operating income.
Space is Starship’s budget line. Space revenue was $1.77B, up 6% on the year, on 35 launches against 44 a year ago and 12 customer launches against 15. The segment lost $7M from operations because Space R&D was $1.23B, $376M more than a year earlier, almost all of it Starship. Segment adjusted EBITDA was $357M, the first positive Space quarter of 2026. Flight 14 was the first orbital Starship flight and delivered 26 V3 Starlink satellites, which the release puts at roughly 10x the downlink capacity of a Falcon 9 Starlink launch. Musk on the call: “Success is not guaranteed on these first flights.” Decode: the Connectivity capacity plan now runs through a vehicle that has flown to orbit once.
The bill is paid from the IPO, for now. The Q2 CFO commentary ended on “$100 billion of cash, cash equivalents, and marketable securities, and $47.5 billion in backlog.” The Q3 commentary promises “a robust, investment-grade balance sheet — while moving with the velocity to capture the opportunity in front of us,” and gives neither figure. Decode the velocity. Q3 operating cash flow was $9.63B against $19.59B of capex, so free cash flow was about −$9.96B, and cash plus marketable securities fell $10.15B to $89.86B. The coverage did improve: capex was 2.22x adjusted EBITDA, against 5.19x in Q2, and the gap narrowed from $15.95B in Q2 because operating cash flow jumped. Per Yahoo Finance, management guided Q4 2026 capex to $35B–40B, close to double Q3.
Look at what sits under the cash. In the first half SpaceX took in $85.68B from the IPO and $51.81B of new debt, and repaid $39.40B. Debt and finance leases were $38.84B at September 30, of which $12.90B, or 33.2%, was owed to a related party and drew $351M of the quarter’s $675M interest expense (52.0%). The prospectus names that lender at March 31: equipment leases for AI infrastructure hardware with Valor Equity Partners, whose founder sits on SpaceX’s board, booked as failed sale-leasebacks and totalling $9.04B of debt then. On who ultimately pays for the capex, this quarter’s answer is the June IPO, the customers’ prepayments and a director’s GPU lease book. And the carrying cost is senior to the shareholder runs backwards for now: interest income of $873M covers interest expense of $675M because the cash pile is still large. Every quarter of capex at this pace shrinks the pile that earns that interest.
The balance sheet absorbed a $60B deal. Goodwill rose $36.78B in the quarter to $48.42B and intangibles reached $6.35B, against GAAP purchase consideration of $42.5B for Cursor. Net property, plant and equipment rose $23.75B to $89.48B. Shareholders’ equity is $176.77B.
Here is the bear case at full strength. 36.6% of net income came from below the line. SI’s profit carries 48 days of a charge that runs about $2.33B over a full quarter, and SI depreciation rose $588M last quarter behind $16B of quarterly capex, so on a full-quarter basis the $431M becomes a loss unless the business grows by about a quarter again. A third of the debt is a director’s lease book. Free cash flow was about −$9.96B, capex is going up, and the stock trades near 59.9x annualized Q3 adjusted EBITDA and 141.3x the trailing four quarters. On that reading the quarter is an accounting high-water mark. We land the other way, for one reason: before the acquisition charges, SI earned $1.66B, a $2.91B swing in one quarter, on compute that customers are paying for in advance, and the per-GW series turned in Q2 before Cursor existed in the numbers. That operating line has to add about $0.68B to absorb a full quarter of the charge, and more to carry the new depreciation. If it can’t, the bear is right.
How peers read this
SpaceX now reads as a compute landlord with a satellite cash engine attached. Against CoreWeave, the same model applies — demand-long, balance-sheet-short — but SpaceX covered this quarter’s gap by drawing down the cash it raised in June: cash and securities fell $10.15B to $89.86B. Against Oracle, the $2.81B rise in current deferred revenue is the same customer-funded build in a smaller dose. Fortune reports the SI segment leases compute to Anthropic and Google, among others; the release does not name customers.
For NVIDIA, $16.44B of SI capex in one quarter is one customer’s GPU and data-center bill; the per-GW series says SpaceX is now renting that hardware out at a rate no earlier disclosed quarter reached. For the telecom carriers, the 800 MHz agreement signed on — up to 14 MHz of paired low-band spectrum, closing in the first half of 2027 — adds low-band spectrum to the Starlink Mobile plan. On the call, Musk said SpaceX and Tesla have each formed special committees to consider a merger, per Yahoo Finance.
Valuation reality check
At $159.28 on 13,579M Class A and Class B shares outstanding at September 30, market value is about $2.16T. Add $38.84B of debt and finance leases, subtract $89.86B of cash and securities, and enterprise value is about $2.11T — about 59.9x annualized Q3 adjusted EBITDA, and 141.3x the $14.95B of the last four quarters. That multiple only holds if Q3 is a floor, so the 7.77% move says the market already reads it as one. The library’s rule runs the other way: the increment is not the base. The market capitalized this quarter’s increment as if it were the base.
Split the EBITDA and the dependence is plain. Annualized, Q3 Connectivity adjusted EBITDA is $14.00B and SI is $19.81B. Starlink alone, on any reasonable telecom multiple, is a fraction of a $2.11T enterprise value; the rest of the price is a bet that SI’s Q3 is the run-rate, not the peak of one agreement cycle.
What we’d watch from here
- Q4 SI income from operations vs zero. Above it, the rental business is outgrowing a full quarter of the Cursor charge. Below it, the bear case is live. This is the test the page is written to.
- Q4 SI adjusted EBITDA against about $6.21B, our illustrative breakeven bar with D&A, the Cursor charge and other stock compensation as stated above.
- The Cursor acquisition charge in Q4 against about $2.33B, the full-quarter figure at the Q3 run-rate, and SI D&A against $2.47B.
- SI revenue per GW-year against $15.09B as nameplate passes 1.7 GW. Holding it while capacity grows is the confirmation; a fall back toward $6.27B says Q3 was one agreement cycle switching on.
- Remaining performance obligations in the Q3 10-Q against $47.46B at June 30, and the purchase-obligation table against $27.96B.
- Cash plus marketable securities against $89.86B, with Q4 capex in the guided range.
- Starlink consumer ARPU against $69, net of the one-time margin benefit, and net adds against 1.6M.
The trade
Evidence-led, not a chase. At about 59.9x annualized Q3 adjusted EBITDA the stock already prices SI holding its profit through a full quarter of the Cursor charge and the depreciation behind $16B of quarterly capex. A holder who wants exposure while being paid for volatility could overwrite part of a position into the Q4 print, when the 48-day number becomes a 92-day number. We have not priced strikes here.
The risk sits in the company’s own filings: $23.4B of Cursor award value still to be expensed, $27.96B of purchase obligations at June 30, a third of the debt owed to a related party, and capex rising while the cash pile falls.
Bottom line
Q2 asked whether Starlink could pay for the compute. Q3 says the compute can pay for itself before a stock charge, that customers are prepaying for more, and that revenue per gigawatt turned after a year of falling. But the profit on the page is a 48-day number, depreciation is compounding behind it, and the cash came from the IPO. Our call: SI income from operations stays above zero in Q4 2026. Wrong if the Q4 release shows an SI operating loss.