The sector's dominant business-model migration is a convergence on vertically integrated, end-to-end service provision — the SpaceX template, minus (sometimes plus) launch.
It is driven from both sides: supply-side, SpaceX demonstrated that integration captures the margin merchant layers used to share, collapsing the economics of selling one layer to primes who now build it themselves; demand-side, the customers that matter — defense and sovereign buyers above all — increasingly buy outcomes (data, connectivity, delivery) rather than hardware. The visible result across the watchlist: companies exiting single-layer merchant positions and rebuilding as mission primes — Rocket Lab (launch + manufacturing → operator via Iridium), MDA (chip-to-bus-to-data-product), Intuitive Machines (lander hardware → lunar infrastructure services). Because the migration is mostly organic and contract-driven, the deal record in industry-consolidation captures only its loudest instrument; this dossier tracks the destination model itself. The squeeze casualty is the merchant middle — pure component and subsystem suppliers whose customer base verticalizes away.
(a) merchant pure-plays (Karman, Redwire, Comtech, Gilat as watchlist proxies) sustain revenue growth and stable gross margins through FY2027 while primes verticalize
no margin series built yet, first measurement pass pending
(b) the services/data share of new-space integrators' revenue mix stops rising for 3+ consecutive quarters across the cohort, or a verticalized entrant divests a layer it built or bought
revenue-mix series not yet built (segment work per docs/proposals/COMPANY-DOSSIER-PROPOSAL.md)
(c) major government procurements revert to disaggregated component/bus buying instead of end-to-end service awards
contract-structure series not yet built
This dossier exists because the migration was visible in the desk's corpus but had no container: it is not a deal series (the consolidation dossier's beat), not an event (the briefs' beat), and not any single company's story. The seeding evidence is what the desk already holds: the Rocket Lab-Iridium S-4 mechanics and MDA's build-out in industry-consolidation, and a transcript corpus (inbox/) in which Intuitive Machines, MDA, Rocket Lab, and Firefly management pitch end-to-end positioning quarter after quarter.
The four series defined in TRACKED SERIES: the layer-occupancy matrix baseline for the watchlist, the services-vs-hardware revenue mix from segment disclosures, the management-language frequency series from the transcript corpus, and the deal-direction reclassification of the consolidation dossier's log (three rows seeded). Until the mix series exists, this dossier's confidence stays deliberately below its prose style — see ANALYTIC STANCE.
SES is a L4/L5 operator making its own first verticalization moves down the stack — the factory and payload-level build (operator, 2026-08-25) put it at the entry of L2, though nowhere near the full-stack primes it now competes with and buys from. That makes the convergence question sharper, not moot: the moves so far are payload-deep, and this dossier tracks whether SES extends them (bus, integration) or holds at payloads while suppliers become competitors (MDA is both vendor and emerging end-to-end rival, and moving faster). The Intelsat acquisition was the scale logic, not verticalization — the factory is the first evidence SES is now playing the other logic too. IRIS² is SES's structural answer worth naming as such: a government-mandated consortium is a way to rent an end-to-end stack without owning it. Watch items feeding the "Satellite consolidation wave" thread in priors.md: supplier verticalization eroding SES's negotiating position on its own procurement, and whether the CMD narrative needs an explicit answer to "what is SES in the converged model" — the overhang board in ir-ledger.md suggests the analysts will ask it.
This dossier was operator-seeded from a stated intuition, not earned by a research pass — its job is to make the intuition falsifiable before making it confident, and until the revenue-mix series exists the prose should be read as hypothesis-shaped. Three bias risks named at birth: (1) "end-to-end" is IR marketing language before it is revenue — the `lang.endtoend` series is a narrative tracker and must never be cited as evidence of migration without the mix series moving the same direction; (2) SpaceX-template survivorship — one spectacular success with captive demand does not prove the model generalizes to companies that must buy their demand at market; (3) corpus sampling — the desk's transcript shelf over-represents companies that talk, so quiet non-migrators (and private component houses with no calls) are systematically under-observed; the EDGAR segment data is the corrective, not more transcripts.
Is the five-layer taxonomy right-grained? Ground stations and user terminals may deserve their own layer (Gilat, KVH exist there). Resolved by: friction in the first full watchlist scoring pass.
Do segment disclosures actually separate services from hardware per company, or does `revenue.services-share` need per-company proxies? Resolved by: a 10-K/20-F segment audit across the Tier 1 set.
Where does the merchant middle consolidate — monopoly niches, absorption, or decline? Resolved by: Karman/Redwire/Comtech margin and backlog series over 4+ quarters (falsifier (a)'s data).
Is demand pull or supply push dominant? Resolved by: whether contract-structure changes lead or lag the revenue-mix shifts in time.
Does the convergence eventually reach operators — i.e., is a pure L4/L5 player structurally short the trend, or its beneficiary as integration raises rivals' costs? Resolved by: relative margin trajectories of integrated vs pure-play operators, 2026-2028.
How large is the bus-supply surplus over launch capacity, quantified? Resolved by: a census series — active merchant bus/smallsat product lines and stated annual unit capacity vs the launch-slot supply the launch-bottleneck dossier already tracks; data/raw/tam/ workbook satellite-built-vs-launched sheets as the baseline.
EO as the furthest-along arena: end-to-end EO (tasking → analytics) has its own dynamics the eo-adjacency dossier owns — what belongs there vs here? Resolved by: first pass that tries to score the EO names on the layer matrix; keep this dossier cross-sector.
Is BlackSky a merchant manufacturer (L2:both)? The operator believes it has sold satellites; the mechanism would be LeoStella (the manufacturing JV with Thales Alenia, which has built for third parties such as NorthStar) — but a half-owned JV's merchant sales are not obviously BlackSky "occupying L2 merchant", and the JV's current ownership needs confirming. Resolved by: LeoStella ownership status + any satellite-sale revenue in BlackSky's segment disclosures; until then the cell stays captive rather than carrying a claim the desk cannot source.
Do Iridium and Globalstar BUILD their own terminals/devices, or buy them? Under the build-vs-buy ground rule this decides a G:captive cell for each (Iridium designs transceiver modules and handsets but contract-manufactures; Globalstar's SPOT devices similarly). The operator explicitly left Iridium open. Resolved by: equipment-segment disclosures and manufacturing arrangements in each 10-K; score only what the filings support.
Complete append-only series — 16 readings, 25 Aug 2026 → 26 Aug 2026. Rows never change once added; the time series is the evidence base. Each row number is a permanent link — click it to copy a link straight to that reading.
| # | Date | Metric | Value | Source |
|---|---|---|---|---|
| 1 | --- | --- | --- | --- |
| 2 | 2026-08-25 | deal.direction.rocketlab-iridium | vertical (manufacturer/launcher → operator) | [industry-consolidation](dossiers/industry-consolidation) |
| 3 | 2026-08-25 | deal.direction.amazon-globalstar | horizontal-scale (spectrum/subscriber mass) | [industry-consolidation](dossiers/industry-consolidation) |
| 4 | 2026-08-25 | deal.direction.ses-intelsat | horizontal-scale (same-layer operator) | [industry-consolidation](dossiers/industry-consolidation) |
| 5 | 2026-08-26 | layer.matrix.ses | L1:entering + L2:entering (payload build = L1 subsystem work; planned whole-vehicle assembly = the L2 step) | operator, 2026-08-26; boundary definition companies/README.md |
| 6 | 2026-08-26 | layer.matrix.planet | L2:both (sells satellites merchant — Pelican-derived sale to SKY Perfect JSAT, 2025) | Planet press 2025-02; operator flag 2026-08-26 |
| 7 | 2026-08-26 | layer.matrix.spire | L2:both + L4:both (Space Services builds AND operates satellites for customers) | Spire Space Services segment, company reporting; operator flag 2026-08-26 |
| 8 | 2026-08-26 | layer.matrix.spacex | L2:both (Starshield builds satellites merchant for the US government — NRO proliferated constellation) | Starshield/NRO public reporting 2024; full sweep 2026-08-26 |
| 9 | 2026-08-26 | layer.matrix.mda | L4:captive (RADARSAT-2 owned and operated today — L4 was mis-scored as merely entering) | MDA Geointelligence segment; full sweep 2026-08-26 |
| 10 | 2026-08-26 | layer.matrix.satellogic | L2:both (dedicated satellite programs sold to governments, e.g. Albania 2023) | Satellogic company reporting; full sweep 2026-08-26 |
| 11 | 2026-08-26 | layer.matrix.sidus | +L4:entering (LizzieSat constellation operating — the position was missing) | Sidus LizzieSat program; full sweep 2026-08-26 |
| 12 | 2026-08-26 | layer.matrix.ohb | +L1:merchant (subsystems/equipment sold below whole-vehicle level) | OHB subsidiaries' product lines; full sweep 2026-08-26 |
| 13 | 2026-08-26 | layer.matrix.kratos | −L2 RETRACTED (no sourced evidence Kratos ships whole satellites; L1 electronics + G OpenSpace stand) | source-it-or-retract rule, companies/README.md; full sweep 2026-08-26 |
| 14 | 2026-08-26 | layer.matrix.kvh | +L5:merchant (primarily a maritime connectivity service provider — airtime/AgilePlans — alongside terminals) | KVH segment reporting; operator flag 2026-08-26 |
| 15 | 2026-08-26 | layer.matrix.spacex | +G:captive (designs and builds Starlink terminals and gateways in-house) | operator rule 2026-08-26 (build vs buy); SpaceX terminal production, public record |
| 16 | 2026-08-26 | layer.matrix.amazon | +G:captive (Leo customer terminals designed and built in-house, own chipset) | operator rule 2026-08-26 (build vs buy); Amazon Leo terminal disclosures |
Earlier claims that evidence disproved — kept on record. 2 corrections documented.
Own-network terminals as L5 delivery, never ground. Believed 2026-08-26 (full-sweep rule) → 2026-08-26: the sweep's second scoring rule held that terminals sold to consume the company's own network count as L5 service delivery, not a ground-layer position — so SpaceX and Amazon carried no G. Overturned same day by operator: building your own terminals and ground infrastructure IS verticalization, and scoring it away loses exactly the signal the matrix exists to show. Replaced with the build-vs-buy rule (companies/README.md): G:captive for in-house ground build, G:merchant for third-party sales, no G for operating on purchased equipment. SpaceX and Amazon re-scored +G:captive; whether Iridium and Globalstar design their own devices or buy them is OPEN QUESTIONS #9, not assumed either way.
SES as pure L4/L5 ("no captive manufacturing"). Believed 2026-08-25 (seed) → 2026-08-25: the seed asserted SES occupies L4/L5 only, with no captive manufacturing. Overturned same day by operator correction: SES is verticalizing down the stack — its factory builds at least at the payload level — so SES is an entrant at L2's edge, not a pure-play operator. The IMPLICATIONS FOR SES read is corrected accordingly; the lesson for this dossier's method is that layer-matrix scoring must come from capability evidence, not from a company's historical category.