Built by Claude FableSPCX | How the Business Works
Deep dive pages: Overview | Business Model | Financial Trends | Thematic Exposure | Management Quality | Investor Sentiment | Concerns, Catalysts & Risks
Prepared 2026-10-29. Fiscal year = calendar year. History is recast for the xAI common control merger (closed 2026-02-02), so 2025 quarters include xAI and X. Cursor closed mid-Q3 2026 and added $608M of Q3 revenue; it is not in prior periods. Daloopa carries FY2023 onward and quarters from Q1 2025.
| Segment | What it sells | How it is priced | Q3'26 revenue ($M) | YoY | Share of Q3 |
|---|---|---|---|---|---|
| Space | Falcon and Starship launches, Dragon crew and cargo, NSSL missions; launch development contracts (NASA, DoD) | Fixed price per launch; milestone-based development | 1,767 | +6.1% | 12% |
| Connectivity | Starlink consumer broadband; enterprise and government (aviation, maritime, Starshield, mobile) | Monthly subscription plus hardware (ARPU $69/month); enterprise and government contracts | 5,248 | +69.5% | 36% |
| AI | Cloud services (rented compute), Grok subscriptions and API, Cursor, X advertising | Compute rented at "effectively the spot market rate" on short contracts (filed terms show 90-day outs); subscriptions; ads | 7,560 | +829.9% | 52% |
| Total | 14,575 | +161.4% | 100% |
The flywheel: Starlink buys most Falcon flights, which funds a launch cadence no competitor matches, which lowers the cost of deploying satellites. The AI segment borrows the engineering and power-building skills (in-house turbine blade casting, Colossus build speed) but not the launch moat; its moat is speed to energize, which erodes as grid and turbine bottlenecks ease.
FY2021 and FY2022 are not in Daloopa; the S-1 recast starts at FY2023.
| Metric | Q1'25 | Q2'25 | Q3'25 | Q4'25 | Q1'26 | Q2'26 | Q3'26 |
|---|---|---|---|---|---|---|---|
| Launch services | n/a | n/a | n/a | n/a | 330 | 648 | 1,130 |
| Launch development | n/a | n/a | n/a | n/a | 289 | 314 | 637 |
| Space | 865 | 746 | 1,666 | 809 | 619 | 962 | 1,767 |
| Space YoY | -28.4% | +29.0% | +6.1% | ||||
| Consumer | 1,492 | 1,721 | 1,986 | 2,009 | 2,148 | 2,485 | 3,121 |
| Consumer YoY | +44.0% | +44.4% | +57.1% | ||||
| Enterprise and government | 983 | 867 | 1,111 | 1,218 | 1,109 | 1,806 | 2,127 |
| Enterprise and government YoY | +12.8% | +108.3% | +91.4% | ||||
| Connectivity | 2,475 | 2,588 | 3,097 | 3,227 | 3,257 | 4,291 | 5,248 |
| Connectivity YoY | +31.6% | +65.8% | +69.5% | ||||
| AI solutions and infrastructure | 284 | 311 | 359 | 403 | 475 | 2,194 | 7,022 |
| AI solutions YoY | +67.3% | +605.5% | +1,856% (ex Cursor +1,687%) | ||||
| Advertising (X) | 443 | 426 | 454 | 521 | 343 | 367 | 538 |
| Advertising YoY | -22.6% | -13.8% | +18.5% | ||||
| AI | 727 | 737 | 813 | 924 | 818 | 2,561 | 7,560 |
| AI YoY | +12.5% | +247.5% | +829.9% |
The 2025 quarterly launch services and development split was not pulled for this run; the Q2 2026 8-K shows Q2 2025 launch services of $490M and launch development of $256M.
| Metric | Q1'25 | Q2'25 | Q3'25 | Q4'25 | Q1'26 | Q2'26 | Q3'26 |
|---|---|---|---|---|---|---|---|
| Space Adj. EBITDA | 224 | -93 | 674 | -152 | -351 | -205 | 357 |
| Space margin | 25.9% | -12.5% | 40.5% | -18.8% | -56.7% | -21.3% | 20.2% |
| Connectivity Adj. EBITDA | 1,618 | 1,583 | 1,923 | 2,044 | 2,087 | 2,597 | 3,499 |
| Connectivity margin | 65.4% | 61.2% | 62.1% | 63.3% | 64.1% | 60.5% | 66.7% |
| Connectivity EBITDA YoY | +29.0% | +64.1% | +82.0% | ||||
| AI Adj. EBITDA | -112 | -276 | -428 | -421 | -609 | 1,146 | 4,953 |
| AI margin | -15.4% | -37.4% | -52.6% | -45.6% | -74.4% | 44.7% | 65.5% |
| AI EBITDA YoY change ($M) | -497 | +1,422 | +5,381 | ||||
| Total Adj. EBITDA | 1,730 | 1,214 | 2,169 | 1,471 | 1,127 | 3,538 | 8,809 |
| Total Adj. EBITDA YoY | -34.9% | +191.4% | +306.1% |
AI now earns a Connectivity-like segment margin (65.5% vs 66.7%), but on very different terms: Connectivity is millions of monthly subscribers, AI is a few large customers on short contracts at spot prices. Space margin is falling as Starship R&D rises.
AI compute economics: revenue per watt
| Yield measure ($ revenue per W per year) | Value | Basis |
|---|---|---|
| Q3 2026 blended (AI solutions / end nameplate) | $16.5 ($15.1 ex Cursor) | $7,022M x 4 / 1.7 GW |
| Q3 2025 blended | $2.4 | $359M x 4 / 0.6 GW |
| YoY change | +591% | new cloud contracts on newly energized capacity |
| Per IT watt (nameplate / 1.4) | $23.1 | house IT basis |
| External rented capacity (about half of compute) | about $24 to $26 per nameplate W; $33 to $36 per IT W | estimate |
| Google contract (filed terms) | about $51 per IT W | research/ai-economics/debate/cloud_rev_per_mw.md |
| Anthropic contract (filed terms) | about $24 per IT W | same |
| Management range (Q2 and Q3 calls) | $30 to $50, "slightly above the high end" now | rented IT watts, newest contracts |
| Q1 2027 guide implied | about $35 per nameplate W | estimate from the $33B to $37B guide |
| Metric | Q1'25 | Q2'25 | Q3'25 | Q4'25 | Q1'26 | Q2'26 | Q3'26 |
|---|---|---|---|---|---|---|---|
| Starlink subscribers (M) | 5.0 | 6.0 | 7.2 | 8.9 | 10.3 | 12.0 | 13.6 |
| Subscribers YoY | +106% | +100% | +89% | ||||
| Starlink ARPU ($/month) | 86 | 85 | 83 | n/a | 66 | 66 | 69 |
| ARPU YoY | -23% | -22% | -17% | ||||
| Launches | 38 | 46 | 44 | 42 | 40 | 38 | 35 |
| Launches YoY | +5% | -17% | -20% | ||||
| Mass to orbit (t) | 450 | 652 | 525 | 586 | 556 | 485 | 476 |
| Mass YoY | +24% | -26% | -9% | ||||
| Nameplate compute (GW) | 0.3 | 0.4 | 0.6 | 0.8 | 1.0 | 1.4 | 1.7 |
| Nameplate YoY | +233% | +250% | +183% | ||||
| Backlog ($M) | n/a | n/a | n/a | 28,377 | 27,621 | 47,461 | 62,959 |
| Backlog within 1 year | n/a | n/a | n/a | 32% | 36% | 56% | 67% |
Subscriber growth is slowing on a larger base (13.6M), and ARPU is down 17% YoY on geographic and plan mix (international and lower-priced plans); the filings do not split international vs US subscribers. Launch counts fell YoY in Q2 and Q3 because internal Starlink Falcon flights dropped as deployment shifts to Starship V3.
Two customers were about 51% of Q3 revenue, and one customer was 27.7% of receivables at 9/30/26. Management names Google and Anthropic as cloud customers; the filing does not identify Customer B.
| Segment | SpaceX position | Main competitors | Replaceable within 12 months? | Pricing |
|---|---|---|---|---|
| Launch | About 50% of world launches, over 80% of mass, about 90% of US launches | CASC and Chinese commercial, Rocket Lab, ULA, Arianespace, Blue Origin | No | Setter |
| Satellite broadband | About two-thirds of active satellites, over 80% of satellite broadband subscribers (est) | Amazon Leo (about 400 satellites), Eutelsat OneWeb, Viasat, Chinese constellations | No in rural, maritime, aviation, government; yes in dense urban (fiber, 5G) | Setter |
| AI compute | About 7% of 2026 global AI GW adds, about #6 builder | AWS, Microsoft, Google, Oracle, Meta, CoreWeave | Yes, short contracts at spot | Taker of a tight market |
| Metric | Q1'25 | Q2'25 | Q3'25 | Q4'25 | Q1'26 | Q2'26 | Q3'26 |
|---|---|---|---|---|---|---|---|
| Space capex | 759 | 946 | 1,113 | 1,014 | 1,052 | 1,174 | 1,606 |
| Connectivity capex | 814 | 1,130 | 1,007 | 1,227 | 1,332 | 1,367 | 1,543 |
| AI capex | 2,567 | 749 | 4,760 | 4,651 | 7,723 | 15,828 | 16,444 |
| AI capex YoY | +201% | n.m. | +245% | ||||
| Total capex (PP&E) | 4,140 | 2,825 | 6,881 | 6,891 | 10,107 | 18,369 | 19,593 |
| Total capex YoY | +144% | +550% | +185% | ||||
| Operating cash flow | 727 | -376 | 3,769 | 2,665 | 1,047 | 2,419 | 9,631 |
| Buybacks | n/a | n/a | n/a | n/a | 4,346 | 80 | 0 |
| Cash and equivalents | n/a | n/a | n/a | 24,747 | 15,852 | 93,522 | 51,294 |
| Total debt | n/a | n/a | n/a | 22,049 | 29,132 | 38,433 | 38,006 |
AI took 84% of Q3 capex. Funding so far is mostly equity: the IPO raised $85.7B net, Cursor was paid with 394M Class A shares ($42.7B), and debt rose to $38.0B. The CFO's plan for the 2027 build is operating cash flow, more investment-grade debt and vendor and customer financing (Q3 call). Buybacks were pre-IPO tender activity only.
Data sourced from Daloopa.