Amazon Leo (Kuiper) — Scope & Funding Ledger

The evidence base behind kuiper.json. Reconstructed by reading the primary record chronologically — FCC filings (DA 26-553; Kuiper Systems' 2026-07-24 D2D application + Exhibit C; the 2020/2024 Gen1 orders), Globalstar SEC filings, and the dossier's FCC-cited tracked series — plus Farrar's June-2-2026 analyst report for the capacity/cost/launch economics. Amazon does not break Kuiper/Leo out in earnings, so there is no company revenue, subscriber, ARPU, or bus-cost figure anywhere in the record. Every number is tagged [FCC], [COMPANY], [DERIVED], or [ANALYST]. Figures that cannot be sourced are flagged, not adopted.

Last reconstructed: 2026-08-14 · Sources: FCC DA 26-553; Kuiper Systems D2D application SAT-LOA-20260601-00224 (2026-07-24); Globalstar 8-K/10-Q/S-4; Farrar/TMF (2026-06-02); planet4589 GCAT — via inbox/, dossiers/amazon-leo-kuiper.md, and the cited FCC/SEC primaries.

1. Authorization & count evolution — reconciled

Four separately-licensed systems, four milestone clocks. Do not sum them into one "constellation."

Count	System / meaning	Primary source
3,236	Gen1, as originally authorized (2020)	[FCC] Order and Authorization, 35 FCC Rcd 8324 (2020), per DA 26-553 n.3
3,232	Gen1, as modified (reduced by 4, 2024)	[FCC] DA 26-553 n.4: Gen1 Orbital Parameters Modification, 39 FCC Rcd 2180 (SB 2024)
3,212	Gen2, authorized	[FCC] DA 26-553 n.8; grant Feb 10 2026 (Call Sign S3105/S3051)
1,292	Polar, authorized (V-band added to Gen1 same date)	[FCC] DA 26-553 n.8
~7,736	Total across the three FSS systems	[DERIVED] 3,232 + 3,212 + 1,292
5,105	D2D/MSS system — the fourth sibling	[FCC/COMPANY] D2D application SAT-LOA-20260601-00224 (2026-07-24), "up to 5,105 low Earth orbit satellites" — contingent on the Globalstar close

Gen1 milestone structure [FCC DA 26-553 ¶3], verbatim: "1,616 satellites by July 30, 2026" (= 50% of 3,232) "and 3,232 satellites by July 30, 2029." The interim bar is 1,616, not 1,618; the total is 3,232, not 3,236 — the 1,618/3,236 pair still common in trade press is the pre-2024 authorization.

The waiver (¶¶18–23) — NOT the extension Amazon asked for. The Jan-30-2026 application sought to move the 50% milestone to Jul 30 2028. The Bureau instead waived the 50%-by-deadline requirement in place: launching may continue, but post-deadline satellites lose spectrum priority until Mar 30 2028 (or Oct 30 2027 if ≥50% are "constructed with launches procured"), whichever first. The surety bond was not waived; a miss is a §25.164(b)(1) violation. Sole opponent: SpaceX. Gen2/Polar clocks [n.8]: 50% by Feb 10 2032 — years out, not milestone-active.

2. Deployment — the launch-bound thesis

Gen1 50% milestone — missedsats

24.5%396 in orbit of 1,616 due Jul 30 2026

1,220 short; ~27 Vulcan flights to close. Bureau waived the deadline in place — post-deadline sats lose spectrum priority.as of 2026-07-27FCC DA 26-553 (milestone, HARD); GCAT (in-orbit count)

Date	Metric	Value	Source
2026-01-30	Amazon self-projection at waiver filing	~700 by Jul 30 2026 (≈21%) — itself below the 50% bar	[FCC] DA 26-553
2026-07-02	Final Atlas V; retired from Kuiper manifest	396 in orbit	[COMPANY]
2026-07-27	Launched vs in orbit	398 launched / 396 in orbit (2 prototypes deorbited)	[COMPANY/DERIVED] GCAT
2026-07-27	Milestone shortfall	396 / 1,616 = 24.5%; 1,220 short; ~27 Vulcan flights to close	[DERIVED]
2026-07-30+	Deadline lapse	Missed; no FCC order/enforcement naming compliance found	[FCC-negative]

The binding constraint is launch, not capital, manufacturing, or demand. Amazon's own ~700 self-projection was below the bar, then beaten to the downside (396). Consumer service slipped "mid-2026" → "later in 2026" [COMPANY].

3. Launch manifest by vehicle — the grounded-vehicle problem

Launch manifest — capacity vs commitmentSats per flight against missions booked — the biggest bets sit on grounded vehicles

Show the numbers

Launch manifest — capacity vs commitment	Sats per flight	Missions booked
Vulcan · grounded	45	38
New Glenn · NET-2026	48	12
Ariane 6	36	18
Falcon 9	24	13
Atlas V · retired	29	9

Area (sats/flight × missions) ≈ satellites delivered. Vulcan (45 × 38 ≈ 1,710) is the backbone — and it's grounded; New Glenn, the densest per flight (48), is NET-2026. Only Ariane and Falcon 9 are flying, at lower capacity. Muted = grounded or retired: the multi-sourcing hedge failed because the failures correlated.as of 2026-07orbitalradar + dossiers/launch-bottleneck.md (SECONDARY); ECFS '92+ secured'

Vehicle	Missions	Sats/flight	Status
ULA Vulcan Centaur	38	45	Grounded (SRM separation, Feb 2026)
Ariane 6	18	36	Flying, low cadence
Blue Origin New Glenn	12+	48	NET-2026 (GS1 static-fire explosion)
SpaceX Falcon 9	13+	~24	Flying — bought from the competitor Amazon is racing
ULA Atlas V	9	~29	Retired 2026-07-02

Sources: [COMPANY/secondary] orbitalradar + dossiers/launch-bottleneck.md; an ECFS comment cites "92+ launches secured / $10B+ launch contracts" [FCC/ECFS]. The multi-sourcing hedge failed because the failures correlated — the two high-capacity vehicles (Vulcan, New Glenn) are both down. Farrar [ANALYST]: "nearly $70M per Falcon 9 … over $100M for each launch on Atlas V, Vulcan, Ariane 6 and New Glenn."

4. Economics — the Quilty analyst anchor (still not disclosed)

Everything here is estimate. Amazon has disclosed no bus cost, launch total, capacity/sat, terminal cost, ARPU, subscriber count, or Leo-segment revenue. What changed: the Quilty Space deep dive now gives a sized analyst view of the cost where there was only a hole — an estimate, not a disclosure, but a corroborated one.

Quilty's Gen1 cost estimate — $16.5–20B$Bvs Amazon's original $10B pledge — a 65–100% overrun; launch is >half

Gateways

$1–2B

Manufacturing

$5–6.5B

Launch

$10.5–11.5B

Excludes user terminals and fleet replenishment. Midpoints shown; total $16.5–20B.as of ~2025Quilty Space, 'Amazon Project Kuiper: A Quilty Space Deep Dive' (INBOX) — Quilty estimate

Quilty estimates the satellite bus cost at $1.5–2M each — roughly 3–4× Amazon's original target.

targetCOMPANY

Amazon originally aimed for a $500k per satellite cost.

estimateANALYST · Quilty

"our estimates suggest it's closer to $1.5 to $2M each. Enhanced features like laser crosslinks, onboard processors, and a high-power design have pushed the original target out of reach."

cross-checkDERIVED

$1.5–2M × 3,232 ≈ $5–6.5B manufacturing — matches Quilty's own manufacturing bucket.

→ maps to kuiper.json · busCostPerSat = $1.75M (Quilty midpoint), movable $0.5–2M — anchors cost/sat as estimated

Quilty puts total Gen1 at $16.5–20B, a 65–100% overrun on Amazon's $10B pledge — and launch is more than half of it.

pledgeCOMPANY

Amazon (2020): "invest more than $10 billion in Project Kuiper."

totalANALYST · Quilty

"true costs of standing up the network ranging somewhere between $16.5 and $20B" — buckets: gateways $1–2B, manufacturing $5–6.5B, launch $10.5–11.5B.

launchANALYST · Quilty

"more than half of Kuiper's internal spending (excluding user terminals) is allocated to launch expenses alone." $7.4B Blue Origin+ULA is corporate-disclosed; Ariane 6 $2–3B, Atlas 5s ~$1B, Falcon 9s ~$200M are Quilty estimates.

scopeFLAG

Excludes user terminals and fleet replenishment — the $16.5–20B is build-only.

Capacity and launch estimates corroborate across analysts.

capacityANALYST · Farrar

~117 Tbps nominal from 3,232 Gen1 = ~36 Gbps/sat; a ~2.3M-residential-subscriber global ceiling on full Gen1; a Gen1 sat is "only about 40% of the capacity of the Starlink V2 mini." Quilty publishes no Kuiper Tbps figure ("n.a.").

launchANALYST · Farrar

"~$10B for the launch of its 3,200 satellites (~$3M each)" — consistent with Quilty's $10.5–11.5B launch bucket.

Terminals and consumer price remain soft.

terminalCOMPANY

Standard terminal cut to a $400 materials cost (2023, volume unstated); Quilty notes Starlink started at $3,000 — "7.5× Kuiper's cost at the jump."

terminalFLAG · ANALYST

Farrar judges a $400 cost "very tenuous" at Amazon's likely 10–20%-of-Starlink volume; consumer monthly price unannounced.

The one company hard number remains a delta, not a level — CFO Olsavsky, Q1 2026: a ~$1B YoY satellite manufacturing + launch cost increase hit Q2 operating income.

5. D2D / spectrum — the bought layer

Date	Item	Value	Source
2026-04-13	Globalstar acquisition announced	~$11.6B; $90/share; ≤$110M milestone adjustment	[COMPANY] Globalstar 8-K + Ex. 99.1
2026-04-13	Apple anchor transfers to Amazon	prepayment ~$468M → ~$1.58B aggregate	[COMPANY]
2026-04-16	MSS spectrum	~8.725 MHz L-band + 16.5 MHz S-band; ~28 ground stations; 24 in-orbit sats	[ANALYST] bands confirmed in D2D filing
2026-07-24	D2D application	5,105 sats, 5 shells; MSS 1.6/2.4 GHz; US SCS not requested; contingent on Globalstar close	[FCC/COMPANY] SAT-LOA-20260601-00224
2028	Next-gen D2D deployment	Amazon to deploy "from 2028"	[COMPANY, forward-looking]

Transfer-of-control review: GN Docket 26-134 (corrected from the "26-129" the dossier previously carried). Globalstar close guided to 2027, gated on HIBLEO-4 launches and S-4 effectiveness; termination fees $419.8M / $592.1M.

6. Demand book — split by conversion horizon

Near-term demand (2026) is government-mediated.

NBN CoCOMPANY

Australia 300,000+ premises from mid-2026.

US BEADPRIMARY

Colorado $25.3M / 42,252 locations.

footprintFLAG · ANALYST

a national ~219,000-location / ~19-state footprint is a New Street estimate, not a government tally — flagged, not adopted.

Forward demand (2027–2028) is anchored by aviation and telco backhaul.

DeltaCOMPANY

500 aircraft from 2028 (won vs Starlink).

JetBlueCOMPANY

~75 aircraft from 2027; Herotel/"evry" from 2027; telco backhaul (Verizon, Vodafone).

A disconfirmer breaks the clean "Amazon wins aviation" read.

AmericanANALYST

American Airlines chose Starlink over Amazon Leo "despite … much lower prices from Amazon Leo."

7. Funded vs notional — do not conflate

FUNDED (being built): the Gen1 build — but company funding evidence is a cost step, not a level.

Gen1COMPANY

3,232 authorized; 396 in orbit.

cost stepCOMPANY

CFO Olsavsky, Q1 2026: a ~$1B YoY satellite manufacturing + launch cost increase hit Q2 operating income — a delta, not a total.

launchFCC · ECFS

launch contracts "$10B+" contracted.

sizingANALYST · Quilty

the only sized totals are analyst: Quilty's $16.5–20B (Farrar's ~$20B in the same range).

Authorized, not milestone-active: Gen2 and Polar are licensed but their clock is years out.

Gen2 + PolarFCC

Gen2 (3,212) and Polar (1,292) licensed; the 50% clock starts Feb 10 2032. No capex traceable.

Notional / contingent: the D2D system is filed but not committed.

D2DNOTIONAL

5,105 sats — filed, contingent on the Globalstar close, "from 2028."

Acquisition (funded on close, not yet closed): the Globalstar deal.

GlobalstarCOMPANY

$11.6B deal + Apple ~$1.58B prepayment — guided to a 2027 close.

8. How much is still un-anchorable — the explicit caveat

The cost side now has an analyst anchor; the revenue side does not — and that is what still blocks a verdict.

The cost side is now anchored (analyst), but the revenue side remains fully undisclosed.

costANALYST · Quilty

bus $1.5–2M/sat, total $16.5–20B — sized, corroborated, estimated (not disclosed).

revenueGAP

Amazon reports no Leo segment: no hard subscriber, ARPU, or Leo-segment-revenue figure exists. Quilty's only revenue math is a hypothetical ("if Amazon can reach 100M subs × $30 ARPU → $36B") — explicitly not a projection.

verdictDERIVED

So the IRR is suppressed on the revenue hole, not the cost hole: at Quilty's cost the replenishment treadmill exceeds net revenue at any disclosed-scale scenario. The balance-sheet-depth thesis is supportable; the unit-economics verdict is not, on disclosed evidence.

What this maps to in `kuiper.json`

The fixture models Gen1 only (3,232 authorized, 396 deployed). Capacity and launch $ sit at the estimated tier (Farrar); the bus cost is now anchored to Quilty's $1.5–2M estimate (movable slider, default $1.75M), so cost/sat (~$4.84M) and the replenishment treadmill (~$3.1B/yr) compute — reliable, but not verdict-publishable because they rest on analyst estimates. The IRR verdict stays suppressed, but for a sharper reason than before: with cost anchored, no IRR exists at any disclosed-scale revenue (the treadmill outruns net revenue), so it can only be made to close by assuming revenue Amazon has never disclosed. Gen2/Polar/D2D are out of the modelled scope (authorized/notional).

Amazon Leo (Kuiper)