Singapore Space AgencyRead the full analysis →

01Institutional Design Study · Hypothetical Venture Blueprint · July 2026

The Third Pole
in Orbit

Orbital compute is consolidating into a G2 game. The middle powers' answer must be mechanisms, not communiqués. OneWeb's soul, DayOne's method: a Singapore-founded operating company for Indo-Pacific sovereign orbital compute — multi-sovereign capital that does not control it, supplier nations that compete for it, and a constellation sized by contracts, not coverage vanity.

Dylan · Singapore Space Agency (independent research platform)

Independent analytical design study. Not a company, not a solicitation, not investment advice. Named funds, agencies and companies appear as archetypes and analytical references only. Every load-bearing figure is labelled: verified fact / company claim / modeled estimate / author inference / scenario assumption. Full report: singapore-orbital-compute-venture-2026.

Singapore Space AgencyReport date 2026-07-12

02The Strategic Gap

A two-horse race — neither can sell legal independence from its own state

US pole · verifiedSpaceX AI1

$1.77T

IPO June 12, 2026. AI1: 150 kW/satellite, 110 m² radiators, FCC filing up to 1M spacecraft. Sold as a US product on US terms — silicon, spectrum, kill-switch under one flag.

China pole · verifiedState mobilisation

Apr 2026

SASTIND feasibility study + CAICT industry committee. In orbit: Three-Body, 12 sats × 744 TOPS. A G2G offer on Beijing's terms — structurally unacceptable to half the region's capital.

The unserved middleIndo-Pacific demand

5 states

Five ASEAN sovereign-AI programmes with funded budget lines. Johor: ~3.8 GW committed demand vs a 2.6 GW grid; Malaysia's non-AI data-centre freeze is ~2 years old. Data can't leave; compute is conditional on Washington.

Terrestrially, the middle path exists — it's called sovereign cloud, and DayOne is its purest capital-markets expression. In orbit, the middle path does not exist. That absence is the venture. Europe is the hope and the counterexample: middle-power capital willed OneWeb into existence; mutual veto, juste-retour and regulatory maximalism turned it into IRIS². Keep the first lesson, dodge the second — in Asia. And a compute constellation's optical data plane sits outside the ITU radio regime — the spectrum moat that blocks new comms entrants does not bind here.

Design study — not a solicitation02 / 17

03Demand Validation — the Question We Cannot Dodge

Why orbit, and not just build the data centre on the ground?

§2's demand facts prove a policy problem is real. They do not by themselves prove anyone will pay a premium for the orbital fix. Score the honest alternative set against the same requirement list a real ministry would use:

RequirementNational sovereign cloudDedicated national DCHyperscaler sovereign regionOrbital enclave
Data stays in-jurisdictionStrongStrongestMediumDepends on gateway design
Power / water availabilityWeak — same gridWeak — same gridWeak — same gridStrong — the one row orbit wins
Upgradeability / refreshStrongStrongStrongWeak — 4–6 yr relaunch cycle
RepairabilityStrongStrongStrongEffectively none
Cost per effective compute-hourLowestLow–mediumLow–mediumHighest, 1.1–11× by year
Export-control exposureStill exists (chips)Still exists (chips)Reduced if allied-flagStill exists on performance stack

Orbit wins decisively on exactly one row. That is not a reason to abandon the thesis — it is the honest scope of it. Everywhere else, the ground wins on price and serviceability.

Infrastructure comparison — the demand answer is on the next slide03 / 17

04Demand Validation II — The Narrow Market Where Orbit Wins

Power-constrained · jurisdiction-bound · latency-tolerant

All three conditions simultaneously, or the ground wins — that sentence is the company's commercial filter for every deal. Three workload classes survive it; "general sovereign LLM inference" notably does not.

Revenue floorEO/ISR in-orbit preprocessing

Data is generated in orbit — the one workload with a physics-native edge no ground facility can copy. Cuts raw downlink ~85–90%. Closes its own economics without any sovereignty premium (slide 12).

Premium upsideSovereign batch inference

Bounded jobs on non-exportable data — viable only where the grid is capped AND the data can't cross borders: the matrix's single winning row, priced at the enclave premium.

Insurance optionResilience / continuity reserve

Survives grid, cable and facility disruption; jurisdictionally diverse by construction. Customers historically fund insurance at low utilisation.

ProvenWhat the evidence actually supports

No single country feeds its own launch/compute tier (India, Japan teardowns). Terrestrial capacity is genuinely power-gated in the target markets. Five ASEAN governments fund sovereign-AI budget lines. US export-control gravity is tightening.

Not provenWhat no public evidence yet shows

That any government will pay a 1.5–4× premium for orbital delivery specifically. That the premium clears against the real cheapest alternative. That "sovereign" and "orbital" are complements, not substitutes, in a buyer's mind.

Premium comparable: national air-gapped data centres already run at ~2–5× commodity cloud — against that substitute, 1.5–3× orbital is not exotic. But willingness-to-pay ≠ 100% capture: this deck's financials assume ~45–55% capture; at 30%, every revenue line sits at its bottom quartile. Stress this input first.

This is the load-bearing slide — everything after it is conditional on this04 / 17

05The OneWeb Autopsy

Keep the soul. Kill the disease.

Multi-sovereign capital financed a constellation no market would fund — twice — and preserved first-round spectrum through bankruptcy. It also produced mutual veto, golden-share repricing, political procurement, and a rescuer (France) that became the controller. Eutelsat-OneWeb, mid-2026: shares ~€2.5, capex cut, waiting on IRIS² (initial 2029, full ~2032).

Bought (keep)Cost (kill)Design consequence
Existence — financing no market suppliesMutual veto; strategy = treaty intersectionSovereigns: LP economics + offtake, zero operational vetoes
Spectrum priority preserved through crisisGolden share repriced all junior capitalNo golden shares; narrow collective negative controls only
Market access via flagsPolitical procurement, no cost disciplineSuppliers compete for priced work packages; supplier ≠ shareholder
Survival bridges across the valleyRescuer became controller (France / IRIS²)Hard caps: state bloc ≤20%, sovereign aggregate <49%
Legitimacy with sovereign buyersCapital patience mistaken for demandOfftake contracts precede sovereign equity
Sources: filings, SpaceNews, Eutelsat H1 FY25-2605 / 17

06The DayOne Inversion

Multi-sovereign money, without multi-sovereign control

DayOne proved the pattern under the hardest test (China-origin capital): one operating company, real assets, parent diluted to ~19.9%, Series C closed at $4.5B with a sovereign LP (INA) aboard. Applied here: sovereigns sit below control; industrial nations sit outside the cap table as competing vendors.

Decision rightsWho decides what

  • Board/management decide: strategy, architecture, procurement, budget, pricing
  • Sovereign LPs get: capped economics, information rights, offtake priority, capability-transfer programmes
  • Supplier nations get: work packages — by winning them, annually re-competed
  • Reserved matters (75%, four only, company-wide): HQ/control leaving Singapore · sale of control to any single-state acquirer · weapons payloads · exclusion-list breach
  • Per-LP enclave rights (individual, non-aggregating): data-boundary veto · sanctions/misuse veto · audit rights · non-resale protection · continuity step-in — narrow, real, cannot combine to block another LP or the company

Cap table · Phase-1 design targets (modeled)Hard caps, every round

Founders + team
18–28%
Deep-tech VC
18–25%
Gulf sovereign (≤10% each)
12–18%
Singapore anchors
10–15%
Japan strategic
6–10%
Korea strategic
6–10%
India strategic
4–8%
Taiwan (supplier only)
0%
China-linked
0%

Sovereign aggregate <49% · no single non-founder ≥20% · no golden shares · every sovereign ticket priced with a signed offtake term sheet. Caps are shareholders'-agreement covenants, not statute — closed with an anti-coalition standstill clause (no aligned voting between sovereign-linked holders outside the collective reserved-matter vote).

Archetypes only — no party's intention implied · founder voting sunsets at Phase-1 completion or IPO · non-voting observer seats offerable06 / 17

07The Regional Resource Map

Every major industrial precursor exists in the region — the integrated, space-qualified system does not

IndiaLaunch volume + cost gravity

NSIL PSLV/SSLV/LVM3 (OneWeb-proven at constellation scale); Skyroot at $1.1B pre-money, round co-led by GIC. FDI: 74% satellites / 49% launch. Wants work-share and tech absorption — deliverable without equity control.

JapanQuality layer + ¥1T chequebook

H3-30 return-to-flight (Jun 2026), ~¥5B/flight target; MELCO/IHI thermal + optical heritage; Space Strategy Fund can subsidise Japanese bids without touching governance (inference). Invests and flies.

KoreaThe memory nobody else has

SK hynix ≈ two-thirds of Nvidia HBM4. Nuri private-led era under Hanwha; KAI/Hanwha buses. A Korean LP + HBM-continuity side letter hedges AI hardware's most concentrated input.

Taiwan · 0% equityMass manufacturing option

Foxconn PEARL satellites, TASA $81M LEO industrialisation, world's densest electronics base. Custom TSMC leading-edge: no — not before 2030. Sells and finances trade; holds no equity.

GulfThe anchor chequebook

PIF/NSG, Mubadala/Space42, MGX-class comfort with non-controlling trust structures. Only regional pool writing $200–500M single-asset tickets. Wants enclaves + capability transfer, not management.

SingaporeThe control layer

NSAS (Apr 2026), MAS/SGX Global Listing Board, GIC/Temasek anchors, SIAC law, IMDA filings (plan, flagged). HQ, mission control, integration. Honest limit: no launch, no fab — and shouldn't pretend.

Suppliers compete; sovereigns anchor; one company decides07 / 17

08The Silicon Question

Dual-stack: accept US strings on one stack, minimise them on the other

StackSiliconPerformanceRegulatory postureServes
PerformanceMerchant Nvidia-class (H100 flown in orbit; B200 manifested Oct 2026) + HBM (Korean supply)N / N−1Full EAR compliance — the US is a disclosed regulatory stakeholder in this stackCustomers Washington licenses: Gulf-in-US-frameworks, ASEAN allies, India, Japan, Korea, Australia
SovereigntyKorean NPU (hedge track, plausible 2028) + rad-tolerant FPGA/SoCN−2Lower-policy-concentration, not "independent" — residual US threads (EDA, HBM channels, equipment lineage) remain; control plane (TT&C, scheduling) ITAR/EAR-free by designWorkloads where jurisdictional cleanliness outranks FLOPs
RejectedCustom leading-edge ASIC (TSMC N3/N5)Allocation, $100M+ NRE and timeline all fail for a startup before 2030 (author inference, high confidence)

Radiation at 500–600 km / 3–5-year life with TMR + ECC + scrubbing is a demonstrated regime for commercial silicon (Starcloud-1, Google Trillium data) — an engineering budget line (~15–25% overhead, modeled), not a research problem. Not closed above 1,200 km; the constellation stays low.

No US part can ground the control plane — bounded, disclosed dependency08 / 17

09Unit Economics · Modeled

Orbital $/kW-year vs terrestrial — the premium is the product

$200k$150k $100k$50k$0 Terrestrial $14–17k 5–11× 2026rideshare 2.5–6× 2028competition 1.1–2.4× 2030-Rreusable works 1.8–4× 2030-Sreusable slips
Orbital delivered cost range (modeled, inputs published in report §6.3)Terrestrial benchmark band

All-in $/kW-year incl. ops, ground, insurance · conservative $17k denominator convention · zero residual value

Bulk compute from orbit is not investable this decade. A sovereignty-priced niche closes from ~2028: governments pay 1.5–3× for ring-fenced compute (modeled input), and more where terrestrial capacity is power-gated. Parity gate: launch <$1,000/kg AND hardware <$40k/kW AND life ≥5y AND utilisation >60% — no earlier than 2029–30.

Sensitivity: launch dominates 2026; hardware $/kW binds by 2030; life extension is the quiet lever · infrastructure cost comparison, not workload cost09 / 17

10Launch Procurement Input

Nobody in Asia beats SpaceX on price — buy them anyway

$0$5k $10k$15k$20k/kg Starship (claim) 0.1–1.5k · scenario Falcon 9 / rideshare 2.7–6k · benchmark LVM3 (India) 5–8k · modeled H3 / H3-30 (Japan) 6–12k · target PSLV (India) 9–18k SSLV (India) 10–14k Zhuque-3 (China) excluded · China rule
Venture's procurable non-US tierUS benchmark / scenario

Vikram-1 maiden window Jul 12–Aug 4 2026 · Nuri capability-not-market · a 2028 procurement faces ≥4 credible non-Chinese bidders + SpaceX

Indian and Japanese launch costs 2–5× Falcon 9 per kg. The venture buys them anyway — for dedicated-orbit control, schedule sovereignty, political acceptability — while using the SpaceX bid as permanent price discipline. The buyer's market OneWeb never had is arriving on exactly this venture's schedule.

Supply curve · force 1Commoditisation

Musk, May 2026: SpaceX launches competitors "with fair terms and pricing" — and, same post, "we reserve the right to reclaim the compute." Commodity price ≠ commodity market: ITAR keeps it politically segmented. Buy the cheap tier; never build on it.

Supply curve · force 2China's EV rerun

Founding wave 2015–19 → reuse inflection now (LM-10B netted an orbital-class stage Jul 10, 2026, first try; re-flight targeted year-end) → $/kg war ~2027–28 → shakeout → export flood ~2029–31 (EV pattern: BYD 56k→417k overseas in two years). Unusable under the China rule — but it bids every WP-L2 price down.

Supply curve · force 3The famished third tier

Astrobase 80t FFSC hot fires 2026 (2029 orbital target); Honda VTVL hop; ISC, Interstellar (Toyota-backed); Innospace. No domestic market feeds even one of them. Aggregated sovereign payloads — this venture's manifest — are the market that only a third pole creates.

WP-L1 dual award ~60/40 · $80–170M Phase-1 launch budget · full curve analysis §6.510 / 17

11Phased Roadmap

Sized by contracts, not coverage vanity

Phase 0 · 2026–27Fly payloads, not promises

2 + 2

Two hosted compute payloads (one per silicon stack) + two optical ground stations. Paid sovereign pilots at $2–8M. Gates Phase 1: flight heritage, monsoon-season optical availability data, two signed offtake term sheets. Capex $25–45M.

Phase 1 · 2028–29The sovereign edge shell

27–36 sats

Dawn-dusk SSO ~570 km, 3 planes, intra-plane OISL rings only. 0.7–1.4 MW nameplate → ~0.35–0.9 MW usable after bus/thermal/derating — ~3–7 rack-equivalents. Capex $400–700M.

Phase 2 · 2029–32 · gatedScale on triggers, not hope

150–230 sats

+30°-inclination shell, cross-plane relay added (true continuity), 5–8 MW generated → ~3–5 MW usable at improved second-gen retention. Hard gates: signed offtake ≥$150M/yr · launch <$1,500/kg · ≥95% fleet availability · two enclave audits passed. No gates, no Phase 2.

The rejection math, shown: continuous ±30° coverage at 570 km needs ~30 planes × 18 ≈ 550+ satellites, $4–7B — serving no workload the demand map requires before 2030.

Dawn-dusk SSO · ~570 km · in-plane optical ISL only coverage gaps — ground track shifts ~24° west each orbit in-plane spacing 30–40° footprint 10.4° half-angle fixed gateway · windows in minutes · recurrence in hours

Schematic, not to scale · conclusion: 27–36 satellites with in-plane-only links deliver asynchronous relay, not real-time cloud — by design, not by defect

The service-model correction: footprint (10.4° half-angle, ~1,160 km) is narrower than the ~30–40° in-plane satellite spacing, and Phase 1 has no cross-plane relay — so a fixed gateway sees any given plane only during few-minute windows recurring on an hour-scale cadence, not continuously. Honest product: bounded-latency asynchronous relay — store, carry, forward — matched to the workloads it serves (EO preprocessing, batch inference), not real-time cloud. Continuity is a costed Phase-2 upgrade, not a Phase-1 assumption.

Geometry: λ = arccos(R·cos ε/(R+h)) − ε at ε=20°, h=570 km11 / 17

12The Product · What a Customer Actually Signs

From sellable watts to a priced workload

Service (Phase-1 product card)Job windowMax result latencyAvailabilitySuited workload
Batch EO/ISR preprocessingcontinuous queue4–12 h~95% monthlyimagery pipelines, maritime domain awareness
Sovereign batch inference2–6 h uploads4–24 h bounded~95% monthlymodel runs on non-exportable data (own-jurisdiction gateway only)
Resilience reservepre-stagedactivation ≤12 h90% standbycontinuity of government/finance workloads

Sellable-capacity waterfall · modeled mid-caseWatts a contract can be written against

1.0 MW generated (27–36 sats)
→ 0.6 MW usable (payload fraction × derating)
→ 0.48 MW sellable (−15–20% reliability reserve)
→ 0.26 MW committed (40–60% utilisation)
≈ 4–15 customers at 20–50 kW enclaves (mid ~8)

One workload, priced end-to-end · scenario assumptionsSovereign EO preprocessing

3–8 TB/day raw → 85–90% cut in orbit
downlink + ground processing avoided: $4.5–30M/yr
orbital cost to serve (20–40 kW): −$2.5–5.5M/yr
net value created: $3–27M/yr (mid ~$10M)
contract at 30–50% capture: $3.5–8M/yr per customer

The floor product closes its own economics with no sovereignty premium at all — avoided downlink pays for the enclave at mid-case. The premium enclave business is upside on top, not the load-bearing floor.

SLA targets set from the §7.2 orbital geometry, not from marketing — Phase 0 exists to measure the monsoon variable12 / 17

13The Commercial Thesis · Modeled

One enclave P&L carries the whole model

Illustrative 200 kW sovereign enclave · 2029Per-anchor economics

Enclave capacity (200 kW × $120k/kW-yr)$24.0M
EO-preprocessing / relay tasking$3–6M
Capability-transfer programme$1–2M
Revenue per anchor$28–32M
Allocated fleet capex (÷4.5 yr)−$10.2M
Ops, ground, insurance (20–25%)−$2.0–2.6M
Contribution margin$15–20M · 54–62%

Fleet math · utilisation = sold fraction of USABLE capacityWhat Phase 1 supports

Base case (40–60% of 0.35–0.9 MW usable): sellable 0.14–0.54 MW → zero to three such anchors blended with the always-on tasking/relay line (the low case is zero anchors, not zero revenue), or 4–5 smaller enclaves → $30–95M/yr at maturity.
Expansion case: 4–6 anchors ($110–190M/yr) needs the top-end build near-saturated (>80%) or the first Phase-2 tranche.

Serviceable market (2030–32, modeled): government $96–450M + commercial maritime/energy $25–240M = $120–690M/yr; capture 30–50% at a ~45–55% WTP capture-rate assumption (stress-test this input first).

Trajectory: 2027 $5–15M → 2029 $20–50M → 2030 $30–95M → 2032 $120–300M (if Phase 2 fires). Operating break-even ~2031 mid-case.

Stated plainly: a bad pure-DCF venture bet, a reasonable strategic-capital bet — which is why OneWeb's soul is required, and DayOne's method is the only safe way to take it.

All figures modeled from §6.3/§7.2 inputs — re-price any assumption13 / 17

14Capital Plan · Modeled Design

$2–3.5B programme — a third of one year of Starlink capex

RoundTimingSizeInvestors (archetypes)Gate to raise
Seed2026$15–25MDeep-tech VC + Singapore platformsTeam · governance constitution locked · pilot LOIs
Series A2027$80–120MGlobal VC + first strategicHosted payloads flying · first paid pilot
Sovereign round2028$450–650MGulf + Japan + Korea + Singapore, per caps≥2 anchor offtakes, ≥$100M TCV — signed first
Phase-22030–31$1.2–2BPrior LPs + Asian export credit (JBIC/NEXI · K-SURE · India EXIM) tied to work packagesPhase-2 gates fire

Export credit is the design's financial punchline: Eutelsat just showed the template (€975M French export credit financing Airbus satellites). The same machinery in Tokyo, Seoul and Delhi finances procurement — which this venture distributes across those countries — rather than equity, which it caps. Supplier nations fund the venture heavily while staying off the cap table.

Phase-1 procurement $290–585M across 8 competed work packages14 / 17

15Risk Register · Honest Residuals

Every mitigation leaves a residual — say so

RiskL / IMitigationResidual
Sovereign-offtake cold startHigh / FatalOfftake-before-equity; pilots on hosted capacity; end-2027 kill-gateThe hardest problem — see verdict
US export-control contaminationHigh / HighDual-stack payloads; EAR-free control plane; enclave-level licensingPerformance stack permanently carries US policy risk
Governance capture (OneWeb relapse)Med / FatalCaps, no golden shares, 75% reserved-matter wall, annual re-competeEnforcement = founder discipline + Singapore law
Launch-cost scenario fails (2030-S)Med / HighPhase-2 gate; Phase 1 viable standaloneCompany plateaus as niche operator
China structural pressureMed / Med-HighStructurally clean of Chinese control, capital, critical supply from day one+15–30% on bus/launch vs China-inclusive procurement (estimate) — the price of existing

The governance-speed risk deserves adult language: this structure trades OneWeb's veto paralysis for key-person risk concentrated in a founder-CEO under Singapore law. Starlink is the existence proof that concentrated authority wins this industry — but it is a trade, not a free lunch.

Full register: 10 risks with mitigations in report §815 / 17

16The Verdict · 不和稀泥

Buildable — conditionally, and the conditions are specific

0%100%
~30–50%

Probability of Phase-1 revenue service · by 2029 · structured judgement, not a statistic

probability a competently executed version reaches Phase-1 revenue service (2029) — recalibrated down from an earlier 40–55%, because the technical gate was repriced: flown-class subsystems do not make first-of-kind system integration a 90% proposition. Gates (low/base/high): offtake by end-2027 ≈ 40/55/70% · sovereign round | offtake ≈ 65/75/85% · execution | funded ≈ 55/70/85% → naïve product 14/29/51%, and the gates are correlated, not independent. Phase-2 durably profitable by 2032–33: ~18–30%. Structured judgment calibrated against comparables, not a measured statistic. Against near-zero for a copycat bulk-compute play or a treaty-based multilateral programme.

Five bright lines · company vs vanity project

  • Offtake precedes sovereign equity — every round, no exceptions
  • No golden shares, no state board seats; caps behind a 75% wall
  • Procurement by published criteria, re-competed annually
  • Phase gates that can kill the programme, with dates
  • Constellation sized by contracts (27–36), not coverage (550 rejected)

The single hardest problemSovereign-offtake cold start

No government on Earth has yet paid real, repeatable money for orbital compute. This venture must sell a premium product to the slowest buyers before the cost curve arrives, while silicon hangs on US policy. The sequence: constitution locked before sovereign money → hosted payloads within 12 months → two paid government pilots in 2027 → sovereign round only against signed offtake. Sovereign capital arrives after the customers — the thing OneWeb never had.

"IRIS² with extra steps" steelmanned in §9 — governance holds at ~50–65%16 / 17

17Falsifiable Signals · Dated

What to watch

Independent analytical design study by Singapore Space Agency (private research platform; UEN 53448796C; not a government body). All data from public sources as cited in the full report; cost models, cap-table structures, constellation designs, market sizes and probabilities are the author's modeled estimates and inferences from stated assumptions. Not investment advice, not a solicitation, not a company announcement, and not an endorsement of any company or government programme. Read the full report and sources.

Singapore Space Agency · July 202617 / 17