Starlink (SpaceX LEO) — Scope & Funding Ledger
The evidence base behind starlink.json. Reconstructed by reading the SpaceX primary record chronologically — Farrar/TMF's July-2026 Starlink profile (the backbone, itself quoting the S-1, FCC filings and SpaceX progress reports), the Aug-4-2026 Q2 earnings-call transcript, Farrar's Aug-4 note on the Q2 print, and the spacex-financial-architecture dossier's 10-Q series. Every figure carries a dated quote and a HARD/SOFT tag — HARD = SpaceX official / S-1 / 10-Q / call / FCC order; SOFT = Farrar (TMF) or analyst estimate. Unsourced figures are flagged, not adopted.
Last reconstructed: 2026-08-14 · Currency: USD throughout (no cross-currency issue, unlike Lightspeed).
1. Generation / fleet evolution — reconciled
The satellite design (bus/mass) is distinct from the constellation (FCC license) it flies into — §1 tracks the bus, §2 the license.
| Gen (bus) | Mass | First flight | Per-sat capacity | Tag |
|---|---|---|---|---|
| V0.9 | simplified | May 2019 | none (never commercial) | HARD |
| V1.0 | 260 kg (app said 386) | Nov 2019 | 17–23 Gbps claimed / 8–10 realized | mass HARD, realized SOFT |
| V1.5 | ~300 kg (adds crosslinks) | 2022 | ~8–10 Gbps realized | HARD |
| V2 mini | ~800 kg → cut to ~575 kg (spring 2025) | Feb 2023 | 96 Gbps nominal / ~50 realized | HARD nominal |
| V3 | up to 2,000 kg (Starship) | late 2026 / operational Flight 14 | 1 Tbps down / 160 Gbps up | HARD design figure |
| Mobile / Direct-to-Cell | dedicated bus, 350 km | Jan 2024 test; 650-sat set complete Jun 2025 | D2D | HARD |
Current in-orbit (2026-06-30), HARD (Shotwell, Aug-4 call): "our constellation … grew to roughly 10,200 with our 9,600 broadband satellites delivering roughly 800 terabits per second of total downlink capacity." Implied ~600 mobile/DTC (10,200 − 9,600) — implied, not stated; flagged. Farrar's model (SOFT) puts service sats at 10,813 (2025) → 12,286 (2026) — a projection, not the HARD count.
2. FCC authorization evolution — the notional ceiling
| Date | System | Scope | Tag |
|---|---|---|---|
| Nov 2016 | Gen1 Ku/Ka | 4,425 (1,600 initial) | HARD |
| Mar 2017 | V-band | 7,518 (later subsumed as Gen2 payloads) | HARD |
| May 2020 | Gen2 applied | 30,000 | HARD |
| Dec 2022 | Gen2 partial grant | 7,500 @ ~500 km | HARD |
| Jan 2026 | Gen2 expanded | 15,000, altitude 530→480 km (DA-26-36) | HARD |
| Jul 6 2026 | Gen3 filed | up to 100,000 | HARD (a filing, not a grant) |
| Jan 2026 | AI / "Starmind" filed | up to 1,000,000 orbital-data-center sats | HARD (a filing) |
Ceiling target (SOFT): Shotwell (Mar 2026) named 15K–20K V3 as the practical maximum. The 100,000 / 1,000,000 numbers are filings — see §6.
3. Capacity — nominal ≫ saleable ≫ carried
Headline capacity falls off sharply from nominal to saleable to carried — the 800 Tbps figure is gross launched capacity, not deliverable revenue capacity.
Per-gen capacity: nominal figures far exceed realized, and V3's 1 Tbps is a design figure not yet demonstrated at scale.
4. Capex / revenue / profitability timeline
SpaceX group (excl. xAI), S-1 — HARD ($M):
| 2023 | 2024 | 2025 | |
|---|---|---|---|
| Revenue | 7,426 | 11,395 | 15,473 |
| Capex | 3,952 | 5,530 | 8,010 |
| Profit | 468 | 2,027 | 3,766 |
Starlink is the dominant share of SpaceX group capex.
Starlink revenue growth and FY25 operating profit.
Q2 2026 Connectivity segment is the only profitable segment, funding Space and AI losses.
5. Launch — vehicles, packing, cadence
Falcon 9: 650+ launches; commercial price raised to $74M (Mar 2026); 165 launches in 2025. Starlink-dedicated launches/yr (Farrar): 63 (2023) → 90 (2024) → 122 (2025). H1 2026 (HARD, Shotwell): "78 total launches and 1,041 tons … primarily … internal Starlink." Packing: up to 56 V1.5 → 20–23 V2 mini (800 kg) → ~29 (575 kg); V3 on Starship (Flight 13 = 20 test units). V3-at-scale threshold (Musk): "~1,000 … approximately second quarter next year [2027]."
6. FUNDED / operational vs NOTIONAL
FUNDED / in-service — committed, operational programs, all HARD.
NOTIONAL — filed or aspired, no committed program.
7. Honest gaps — flagged, not adopted
Enterprise/gov/mobile split is not disclosed — Farrar's finer split is his estimate.
~600 mobile sats is implied (10,200 − 9,600), not stated.
800 Tbps is nominal, not saleable — saleable ~40–60 Tbps; carried ~3% of nominal.
V3's 1 Tbps is a design figure, not demonstrated in operational orbit.
Aug-11 "~11,000 sats / 13M subs" is an internal all-hands video, unfiled.
Starlink standalone capex/FCF beyond 2023–24 is TMF-modeled.
What this maps to in starlink.json
The fixture's verdict is disclosed free cash flow — Q2 Connectivity adj. EBITDA ($2,597M) minus segment capex ($1,367M) = +$1.23B, self-funding, published, needing no assumption. The modeled IRR rides an estimated cumulative capex, so it stays withheld. Cost/sat (~$1.4M) and the internal marginal launch $/kg are Farrar/un-anchorable — consistent with §4/§7 here.