An own minimal Starlink for Ukraine: 50–180 satellites, $1.4bn–3.6bn and about four years to service

01 · DependenceFour years on someone else’s off switch: from 5,000 terminals in 2022 to a speed cap in 2026

Starlink arrived in Ukraine on 26 February 2022, two days after the invasion, when the then deputy prime minister Mykhailo Fedorov asked Elon Musk for it on social media 27 M. By April SpaceX and the American agency USAID had sent 5,000 terminals; by August more than 20,000 were in use, and about 85% of them were paid for in full or in part by governments and other organisations 28 M. In September 2025 Fedorov cited more than 50,000 terminals, of which more than 29,000 were supplied by Poland 29 M.

Nobody publicly counts how many terminals actually work. Estimates for 2025–2026 differ fivefold: from the official 50,000 to 160,000–250,000, of which about 100,000 are in the army 30,31 L. The reason is that the official figures cover terminals supplied by the state and by partners, while individuals and units also bought their own. Since February 2026 all terminals have to be registered, so the Ministry of Defence and SpaceX now have the exact figure, but public sources do not.

Over these years the dependence has shown itself in at least five different ways. Money: in September 2022 SpaceX wrote to the Pentagon that it could not go on funding the service, and put a year of operation at almost $400m 32 M. Geography: that same month Musk declined to switch on coverage near Crimea for an attack by naval drones 33 M. Purpose: in February 2023 SpaceX restricted the use of Starlink for controlling strike drones 34 M. Politics: in 2025 Reuters reported that American negotiators had raised the possibility of a shutdown in the bargaining over minerals 35 M. And technology: in October 2026 a cap was placed, without warning, on the speed at which terminals can move 1,36 M.

The same mechanism cuts both ways. In February 2026 Ukraine and SpaceX introduced the whitelist: the military register terminals through the DELTA system, businesses through Diia, citizens at administrative service centres (TsNAPs), and unregistered terminals are disconnected 37,38 H. According to Ukrainian officials, this cut off terminals used by Russian forces; some Ukrainian units whose terminals had not been registered in time also briefly lost connection 39 L. The switch that helped Ukraine in February worked against it in October. It remains in the hands of the supplier and of the government of the supplier’s country.

Even so, Starlink is the best service Ukraine can get. The network has more than 11,000 satellites in orbit 40 H, 13.6m subscribers and $5.2bn of revenue in its connectivity segment in the third quarter of 2026 41,42 H. Its nearest competitor, OneWeb, has ~650 satellites and bulkier terminals; in April 2025 fewer than 1,000 of its terminals were working in Ukraine 43 M. So the question is not what to replace Starlink with, but what to have ready for the day it is switched off or restricted.

02 · PhysicsHow many satellites are needed: from 52 at 1,200 km to 160 at 550 km, and why ten will not be enough

A geostationary satellite at an altitude of 35,786 km turns with the Earth and “hangs” over a single point. A satellite in low Earth orbit cannot do this: at 550 km it circles the planet in 95.5 minutes, at 1,200 km in 109 44 H. From any one place on the ground it is visible for only a few minutes: on a pass directly overhead, up to 4.5 minutes at 550 km and up to 9.3 minutes at 1,200 km, if the antenna sees the sky from 25° above the horizon 45 L. Then the terminal has to switch to the next one.

This gives the main constraint. While a satellite is flying over Ukraine, the Earth rotates beneath it, and within an hour it will be over Australia. So a “regional” constellation in low Earth orbit in fact covers the whole band of latitudes in which the region lies, around the entire planet 46 H. For Ukraine this is the 44–53°N band: Europe, Kazakhstan, China, Canada, the north of the United States. At any moment only 0.1–0.3% of the satellites in such a network are over Ukraine itself 47 L.

To calculate the minimum we wrote our own simulation L. It runs through thousands of Walker-type configurations (equally spaced orbital planes with the same number of satellites and a phase offset), checks whether every point in the 44–52.5°N band sees at least one satellite at every moment, and looks for the smallest constellation with no gaps. A check on a known example: the Iridium scheme (66 satellites, 781 km) gives 100% global coverage in the model, as in reality. Another researcher, using their own code and the same method, independently obtained 48, 84, 60 and 144 satellites for the same conditions at altitudes of 1,200, 780, 1,000 and 550 km — within 6–10% of ours 8 L.

Two things affect the number most. The first is altitude: the higher the orbit, the wider the footprint. At 550 km one satellite is seen within a circle of radius ~940 km, at 1,200 km ~1,700 km 48 L. The second is the elevation angle below which the antenna does not work. A flat antenna on a vehicle among trees and buildings, or on a drone with low gain, sees a satellite near the horizon poorly. So for military use we calculated two variants: from 25° and from 35°. Raising the angle from 25° to 35° increases the number of satellites by 1.6–1.9 times.

Inclination matters too. Satellites in orbits inclined at 50–56° spend more time precisely over Ukraine’s latitudes, so a regional constellation is noticeably smaller than a global polar one. OneWeb, which covers the whole planet including the poles, has 648 satellites at the same altitude of 1,200 km (600 operational and 48 spares) and requires a higher elevation angle 49 M. Iridium, by contrast, covers the whole world with 66 satellites in polar orbits, because its phones work with a satellite low above the horizon L. In the best configurations for Ukraine the inclination was 50–56°, as in the other researcher’s calculation 50 L.

But “minimum” here means zero margin. No spare satellite, no failure, no second satellite to share the load. Iridium keeps 66 operational satellites and 9–14 spares in orbit, OneWeb 600 operational and 48 spares 51,49 M. For two satellites to be in view of every point at all times, our model requires 90 satellites at 1,200 km instead of 51–52, and 250 instead of 152–160 at 550 km L. And one more limitation: we calculated the 44–52.5°N band, that is, the land of Ukraine. To cover the south of the Black Sea as well, down to 41°N, another researcher’s calculation says 12–17% more is needed: 54 satellites at 1,200 km instead of 48, 98 at 780 km instead of 84, and 162 at 550 km instead of 144 52 L. We did not check coverage of deep-strike targets in Russia, at latitudes of 55–60°.

03 · CapacityWhat the minimum delivers: 1–3 satellites overhead instead of ~58, and 2–20 Gbit/s for the whole country

Coverage answers the question “is there a connection”. Capacity answers “for how many”. Here the minimum constellation loses to Starlink most. In our model, over Ukraine (on average across the 44–52.5°N band) 1.4 satellites are in view with the 1,200 km scheme, 1.9 at 550 km and 2.7 with double coverage L. A geometric model for Kyiv gives Starlink a median of ~58 satellites in the field of view and OneWeb ~11 9 L. More than three-quarters of the operational second-generation Starlink satellites fly in orbits inclined at 43° and 53°, which pass right over Ukraine’s latitudes 53 M.

The capacity of a single satellite grew roughly tenfold from the first generation of Starlink to V2 mini, and under the V3 design by ~150 times. The first generation of Starlink, according to a model from the Massachusetts Institute of Technology (MIT) that calculated from the companies’ filings with the regulator, delivered 6.5 Gbit/s on average and 19.7 at maximum; the first generation of OneWeb delivered 2.0 on average and ~10 at maximum 54,55 M. For Starlink V2 mini SpaceX does not publish numbers: the model of the bank Stifel assumes 60 Gbit/s 56 L, ISPreview’s estimate is ~96 57 L. The new V3, according to SpaceX, will deliver more than 1 Tbit/s, but weighs ~2 tonnes and is launched only by Starship 58 L.

How much of this reaches Ukraine depends on whether the satellite can steer its beams at the country specifically. Ukraine covers ~0.6m km², while the footprint of one satellite at 550 km is ~2.8m km² 48 L. Ukraine is ~22% of the footprint at 550 km and only ~7% at 1,200 km. A satellite with steerable beams can direct more at it; we assumed 30% of capacity L. A commercial satellite of the Telesat class (~25 Gbit/s on average in the MIT model) then gives the minimum constellation 10–20 Gbit/s for the whole country. If 20,000 terminals are working and a fifth of them are active at any moment, that is 2.6–3.5 Mbit/s per active terminal; with 50,000 terminals and double coverage, ~2 Mbit/s L. If the satellite gives Ukraine only a share proportional to area, the result is 2–10 Gbit/s and 0.5–2.6 Mbit/s per terminal L. This is the least certain number in the whole briefing.

Whether this is enough depends on the task. We found no published bandwidth requirements for Ukrainian drones. What is known is this: because of overloading, Starlink at the front gives about 10 Mbit/s per terminal at a peak speed of 200 Mbit/s 10 M, and operators of ground robots say that safe control requires at least 30 frames per second 59 M. Our conclusion: the minimum constellation will provide control and telemetry, text and voice channels for headquarters, and from a few hundred to a few thousand simultaneous video streams at ~5 Mbit/s each (2–20 Gbit/s ÷ 5 Mbit/s ≈ 400–4,000) L. It will not serve hundreds of thousands of terminals with video, as Starlink does now.

Latency explains why geostationary satellites do not solve the problem. Even at the speed of light the signal takes at least 477 ms to reach geostationary orbit and back, and in practice 600 ms or more 60 M. For reconnaissance, logistics and headquarters messaging that is enough; for piloting a drone in real time, by our estimate, it is not L. In low Earth orbit the physical floor is 15 ms at 550 km and 29 ms at 1,200 km 61 L; Starlink’s median in the US, according to SpaceX, is 25.7 ms, and by an independent Ookla measurement about 50 ms 62,63 M. The difference between 550 and 1,200 km is small for drones, while three times fewer satellites are needed at 1,200 km. The price is larger antennas: at twice the distance the signal weakens about fourfold, so OneWeb-class terminals weigh more than 10 kg and cost $7,000–10,000 64 M.

Two more technical decisions determine how far the network is really “ours”. The first is inter-satellite laser links. Without them the terminal and the gateway have to see the same satellite; at 550 km they can be up to ~2,600 km apart, at 1,200 km up to ~4,400 km, if the terminal sees the satellite from 25° above the horizon and the gateway from 10° 65 L. So gateways in Poland or Romania cover all of Ukraine, but each gateway becomes a point that can be attacked. The best-known attack on satellite communications in this war hit precisely the ground segment: on 24 February 2022 a cyberattack on the KA-SAT network knocked out tens of thousands of modems, and ~30,000 routers had to be replaced 66 M. The second is protection against jamming. According to Ukrainian electronic warfare specialists, the Russian Volna Kupol Garant station jams Starlink over an area of about 20 km² 67 M. In 2026 the Secure World Foundation recorded that in wars so far only non-destructive means have been used — jamming and cyberattacks 68 H.

04 · LaunchesHow many rockets: 4–10 Falcon 9s at $74m each, and without SpaceX, a queue

As recently as 20 years ago, putting a satellite into orbit could cost more than the satellite itself. By NASA’s calculation, the Space Shuttle cost ~$54,500 per kg to low Earth orbit in nominal dollars (~$61,700 in 2018 dollars); the industry average for 1970–2000 was $18,500 in 2018 dollars 69,70 H. At its 2016–18 list price ($62m for 22.8 tonnes), Falcon 9 cut the price to ~$2,700 per kg, a 20-fold drop 69 H. In the CSIS database, which gives prices in 2021 dollars, Falcon 9 is 3–4.5 times cheaper than Atlas V, Ariane 5 and Delta IV Heavy 71 M.

In 2026 a standard Falcon 9 launch costs $74m; in 2025 it cost $70m 11 M. The rocket lifts 17.5 tonnes to low-inclination low Earth orbit with the first stage recovered 72 M, so the list price works out at $4,229 per kg (or $3,246 if calculated on the maximum 22.8 tonnes without recovery) 73 L. SpaceX does not publish how much Falcon 9 carries to 1,200 km; our model takes 13 tonnes, the average payload of its launches according to the prospectus. In practice a kilogram costs more: the typical customer of a dedicated launch carries 3.4 tonnes on average and effectively pays ~$20,800 per kg, while Starlink missions fill the rocket to 96% 74 M.

This brings a first surprise for a sovereign constellation. The number of launches is set not only by the mass of the satellites but also by the number of orbital planes. One launch puts satellites into one plane; changing plane in orbit is expensive, and spreading satellites out by natural drift is slow (months). A scheme of 52 satellites in four planes at 1,200 km (60 with spares, 700 kg each) needs four launches; 160 satellites in ten planes at 550 km (176 with spares) needs ten. Here planes and mass give the same number L. But for lighter satellites the planes become the binding constraint: the rocket flies half-empty, yet you pay for a full launch. At list price this is $0.3–0.74bn, or $4–6m per satellite. SpaceX, by outside estimates, pays $15–28m per launch for its own Starlink, or $0.5–1m per satellite 75 L. That is a hidden advantage the buyer will not enjoy.

The second surprise is where to find rockets. In 2025 SpaceX made 165 of the world’s 329 launches 76,77 H. By its prospectus (not independently verified), it put up more than 80% of all the mass that reached orbit. All other rockets combined lifted about 390 tonnes (our calculation: 2,600 tonnes minus 2,213 tonnes) 12 M. Of SpaceX’s 165 launches, 122 carried its own satellites 76 H. Outside customers do get slots: after Russia cancelled its Soyuz launch in March 2022, OneWeb flew on Falcon 9 8.5 months after signing the contract, and on India’s LVM3 after 6 78 M. The satellites were already built. But the trend runs the other way: in the first quarter of 2026 SpaceX made only 7 launches for outside customers, against 12 a year earlier, and with the arrival of Starship it expects fewer Falcon 9 launches (per the prospectus, which we could not verify independently) 79 L.

European options are so far narrow. Ariane 6 flew four times in 2025 against a plan of six, and for 2026 the plan was cut from eight to six or fewer because of a turbopump fault 13,80 M. An Ariane 64 costs about €115m (an estimate; Arianespace publishes no prices), or ~$6,200 per kg 81 L. Small European rockets such as Isar’s Spectrum (which reached orbit on its second attempt on 5 September 2026) lift ~1 tonne 82 M; for a constellation of dozens of satellites of 500–700 kg, that means dozens of launches. Amazon has more than a thousand satellites built but has launched only 396, for lack of rockets 83 M.

Starship promises prices per kilogram several times lower. On 28 September 2026 it reached orbit for the first time and deployed 26 Starlink V3 satellites, although one of the ship’s engines shut down early 84 M. There is no public price; Voyager Technologies’ annual report implies ~$90m per launch around 2029, or ~$900 per kg at 100 tonnes 85 L. A model by Cambridge economists forecasts an average industry price of ~$300 per kg in 2040 if Starship reaches full scale 86 M. For a 2027 decision this is not an argument: commercial Starship launches for outside customers are the end of the decade at the earliest, and they, too, are SpaceX’s.

05 · PricesSatellites, gateways and terminals: from ~$0.8m to ~$29m per satellite and up to $10,000 per terminal

A constellation is more than satellites. Here is what it includes and what costs the most.

Satellites are the biggest cost item. Contract prices in 2022–2026 fall into two groups. A commercial broadband satellite of 500–750 kg costs about $8m: Telesat’s contract with MDA is ~$1.55bn for 198 satellites 16 M, and Rivada agreed a $2.4bn deal with Terran Orbital for 300 87 L. Eutelsat ordered 264 platforms for IRIS² from Aerospacelab for €2.4bn, or, by our calculation, ~$10.6m per platform; the contract also covers integration of the payload, which Thales builds separately 88 H. A military satellite for the Pentagon’s network, with laser links and secure communications, costs $9–29m. York won a $382m contract for 42 satellites and Lockheed Martin $700m for 42 17 M; in the second tranche prices reached $20–23m at Lockheed and Northrop and $28.6m at Rocket Lab, but the contracts included the ground segment and support 18,19 M.

Against this background, UASAT’s claim of 360 satellites for $1.1bn including launch, or ~$3m per satellite by our calculation 22 L, is a third of commercial contract prices. It is possible for the small satellites of Denmark’s GomSpace, but then the capacity of each is smaller. The opposite pole is IRIS²: €15.6bn for 348 satellites, or, by our calculation, ~$52m per satellite 26 L, because the money covers 18 large satellites in medium Earth orbit, the ground segment and 12 years of operations. Germany’s SATCOMBw 4 programme is €8–10bn for 100–200 military satellites 24 M.

There are hundreds of times more terminals than satellites, but in our cost estimate they take only 3–10% of up-front costs; they become a large item if you buy tens of thousands of military terminals at $5,000–10,000. Within a few years SpaceX cut the internal cost of its terminal from ~$3,000 to under $600 89 M, and in 2026 it sells the Starlink Mini in the US for $149 90 M. But a price that low is reached only in runs of millions. Britain reportedly paid £4,000–7,000 (~$5,300–9,300) a unit for military Starshield 91 L, and a OneWeb-class terminal weighs more than 10 kg and costs $7,000–10,000 64 M. Terminals from a new manufacturer for a Ukrainian network will cost closer to the second group: $5,000–10,000 each. 20,000 terminals at $5,000–7,000 come to $100–140m; 50,000 come to $250–350m L.

The remaining cost items are smaller, but the network cannot exist without them. A gateway costs $5–15m 92,93 L. An optical terminal for an inter-satellite laser link cost ~$1m in 2023, and analysts expect the price to fall to $100,000 by 2030 94 M. Spectrum is the cheapest and slowest item: fees of the International Telecommunication Union (ITU) amount to thousands of Swiss francs 95 L, but a new filing must be coordinated with other operators. Operators have 7 years to bring the system into use (one satellite in orbit is enough); after that the rules require 10% of the filed satellites to be in orbit within 2 years, 50% within 5 and 100% within 7, that is 9, 12 and 14 years after filing 96,97 H. Operations are not free either: Iridium, with 66 satellites, spent $197.6m on cost of services in 2025 and another $100m on capital expenditure 98 H.

06 · Cost estimateThree scenarios: $1.4bn for Minimum at 1,200 km, $3.2bn for Like Starlink at 550 km, $3.6bn for Military with margin

We drew up a cost estimate for three scenarios using a single model. The same model runs in the calculator below, so every number can be checked and changed L. The shared assumptions are these. Spares: 10–15%. A Falcon 9 launch costs $74m at list price and carries ~13 tonnes (the average across Falcon 9 missions according to SpaceX's prospectus, which we could not fully verify; the maximum to low Earth orbit is 17.5 tonnes) 99 L; each orbital plane needs at least one launch. Gateways: 6–8 at $10m each, plus $100m for a control centre and ground network. A further 30% goes on development, integration, insurance and programme management. Operations cost $100–150m a year. Terminals are included in the up-front cost.

Three conclusions follow from these numbers. First: launch is not the main cost item. In scenario A it accounts for 21% of the up-front cost, in C 12%; the satellites themselves cost more (37% and 52%) L. Twenty years ago it was the other way round: at the historical price of $18,500 per kg, putting a 700 kg satellite into orbit would have cost ~$13m — more than the satellite itself 69 L.

Second: orbit altitude affects cost more than the price of a rocket. A constellation at 550 km is three times as large, and its satellites live half as long, because the atmosphere drags them down and they decay from orbit within a few years. For example, in 2026 SpaceX is lowering ~4,400 of its satellites from 550 to 480 km so that failed ones burn up sooner 100 M. Over ten years scenario B costs $6.5bn, of which $2.2bn is the replacement of all the satellites L. At 1,200 km a satellite lasts ~10 years, as Telesat assumes 101 L, but it will not come down from such a height by itself: even at 800 km natural decay takes 100–150 years, according to ESA 102 M. A thruster is therefore needed to deorbit the satellite at the end of its life. If satellites at 1,200 km last not 10 but 5–7 years, the ten-year cost rises to $2.7–7.4bn L.

Third: all the scenarios fit within a single year's budget. $1.4–3.6bn of up-front cost is 2–6% of the defence spending in the draft 2026 budget (UAH 2.8trn at UAH 45.7 to the dollar, or ~$61bn) 14 M. For 2026 alone the EU has offered Ukraine €28.3bn of loans for the defence industry, within €90bn for 2026–2027 103 H. Finding money for a first stage is realistic; the question is who will decide what exactly to spend it on.

07 · TimelinesHow long it takes: 2.6–6.6 years to a first launch in real programmes; in our base schedule, about three years to launch and four to service

Money is easier to find than time is to save. In the nine programmes we compared, the time from a formal decision (a contract, a regulator's permission) to the first launch ranged from 2.6 years (SDA's demonstration series) to 6.6 (Iridium NEXT), and to first service 5.3–7.8 years in completed programmes (4.6–6.4 in plans not yet fulfilled; for OneWeb the 7.8 years is the time to global coverage) 20,21 L. Most took 3–4 years to launch. Even SpaceX, which has its own rockets and factory, took 5.8 years from the announcement of Starlink in January 2015 to the public beta test in October 2020 104 M.

Those who bought something standard moved fastest. In 2020 SDA ordered 20 satellites on ready-made platforms from two suppliers at a fixed price and launched the first ones 2.6 years later 105 M. But even SDA ran late: its first operational batch was launched a year behind schedule, and launches were then halted for six months over problems with cooling, electric thrusters and laser terminals 106 M. Poland received from ICEYE a system of four radar satellites and a ground segment less than 12 months after a contract worth ~€200m 107 H — fast, but for a finished observation product, not a communications network.

We drew up three deployment schedules, with the decision taken in early 2027 L. Fast: buy ready-made satellites from an existing production line (for example, on a platform already being built for another customer), use a partner's spectrum and secure slots on Falcon 9 — service around mid-2029. Base: a new satellite on a ready-made platform, our own filing with the International Telecommunication Union (ITU) — late 2030. Slow: our own development in cooperation with European partners, as with IRIS² — 2033.

Spectrum is a separate cause of delay. A new system must file with the ITU and coordinate with those already operating in the same bands — above all SpaceX, OneWeb and Amazon. From filing to notification of the system takes at least several months, and coordination usually takes years 108 M. This can be sidestepped if the satellites operate under the filing of a partner state: such deals are possible, but we found no public examples for a military system L. A second cause is production: Amazon has built more than a thousand satellites but has still not launched a commercial service 83 M, and Telesat has postponed the start of its global service because of a delay in chips for antennas 109 M.

The industry's history is a reminder that the first constellations rarely pay off. Iridium began operating in November 1998 and went bankrupt nine months later; assets worth ~$5bn were sold in 2000 for about $25m, and it was the US government that paid $72m to keep the network from going dark 110,111 M. OneWeb went bankrupt in March 2020, in the middle of deployment 112 M. For a military network this matters less: its aim is not profit but guaranteed communications. But the programme must be funded in advance for its whole term, or it will stall halfway.

08 · Who is buildingWho is already building their own networks: Ukraine, Germany, the EU, Turkey — and on what timelines

Ukraine is not the first to decide against relying on a single supplier. According to a count by the International Institute for Strategic Studies (IISS), European countries have announced at least $109bn of military space spending to 2030, including €35bn by Germany 113 M. These are the main projects concerning communications in low Earth orbit.

Another route is to buy capacity from a European operator. Eutelsat, which owns OneWeb, was recapitalised with €1.5bn in 2025 and is now largely owned by the French state (29.65%) and the British government (10.89%) 122,123 M. It ordered 440 new satellites from Airbus to replace its first generation, and in September 2026 gave the go-ahead for 229 more (669 in all) 124 H; France, for its part, has bought sovereign capacity from it since 2026 under the CENTAURE programme — up to €350m over eight years 125 H. For Ukraine this route has two limits: OneWeb terminals are large and expensive, and Germany, which paid for the service in Ukraine, funded fewer than 1,000 terminals in April 2025 43 M.

A comparison with other middle-sized states shows that the normal scale for them is dozens or the first hundreds of satellites and 5–10 years. Taiwan did not get Starlink because SpaceX demanded a controlling stake in a joint venture 126 M, and in 2025, according to the Taipei Times, it cut its own plan to four satellites, with a first launch in 2029 127 M. South Korea formally began a programme only in July 2026: it wants a sovereign network by 2035 128 L.

09 · RentRent or build: known bills for Starlink come to $23–50m a year, an own network to $0.24–0.65bn a year

The comparison with renting is awkward for advocates of an own network. Poland pays about $50m a year for Starlink subscriptions for Ukraine 15 M, and spent ~$84m over 2022–2024, terminals included 129 M. The Pentagon’s one-year Starlink contract for Ukraine (June 2023 – May 2024) was worth $23m 130 M. France buys sovereign OneWeb capacity for ~€35m a year 125 H. By our model an own network costs $0.24–0.65bn a year, averaged over ten years L — 5–28 times more than the known bills of individual payers. Nobody publishes the total that all payers hand over for Starlink for Ukraine.

But renting is cheap only until the seller changes the terms. In spring 2026 SpaceX raised the price of a terminal on the Pentagon’s American LUCAS drones from ~$5,000 to ~$25,000 a month — the aviation tariff 131 M. At that price 1,000 terminals on drones would cost $300m a year — a quarter more than the average cost of an own network under scenario A (~$240m) L. In January 2025 Italy discussed a €1.5bn, five-year deal with SpaceX for government communications alone, and did not sign it 132 M.

So an own network is not a saving but insurance. Its price should be compared not with a subscription but with the losses caused by a single night without communications for strike drones and interceptor drones, as on 14 October 2026 2 M. The cheapest insurance is probably a combined one (we did not cost it separately): keep Starlink as the main channel for the mass user, and hold our own minimal constellation and the capacity of European operators for critical tasks — control of deep-strike drones, command networks, naval drones. For these, 2–20 Gbit/s per country is enough for control and telemetry, while how much is left for video depends on what share of its capacity the satellite directs to Ukraine L.

10 · CalculatorBuild your own constellation: altitude, satellite class, rocket, terminals — and what comes out

The calculator uses the same model as the scenarios above L. The number of satellites and orbital planes for each altitude comes from our simulation. The rest can be changed: satellite price and mass, the rocket, the number and price of terminals and gateways, design life, and the share of capacity the satellite directs to Ukraine. The model is deliberately simple: it ignores financing, inflation, insurance payouts and delays.

A few observations that are easy to check in the calculator. Moving from a commercial to a military class raises the up-front cost roughly 1.5 times — more than any change of rocket. Starship at ~$90m per 100 tonnes does not cut the number of launches for 550 km: a separate launch is still needed for each of the 10 orbital planes, so launches cost more ($0.9bn against $0.74bn) L. And double coverage at 1,200 km (90 satellites) gives more satellites overhead than single coverage at 550 km (160 satellites): 2.7 against 1.9, and with a longer design life.

11 · LimitsHonesty and limits

12 · QuestionsFrequently asked questions

13 · MoreFurther reading

14 · SourcesSources and methodology

A number in the text leads to its source. Labels: H — primary source (company reporting, a government or regulator document, a peer-reviewed paper), M — secondary source or checked against several, L — estimate, a statement by a company or official, a single unconfirmed source or our calculation. The data was gathered and cross-checked by 16 AI agents: for each of eight research areas, one gathered facts and a second tried to refute them. No claim was refuted outright; of 296 checked, 41 were corrected and 21 could not be confirmed (such claims are marked L or carry caveats in the text). Two further reviewer agents checked the text against the research. We wrote the coverage simulation ourselves (Python, validated against the Iridium design), and the calculator uses the same cost model as the scenario charts. All sums are in US dollars unless stated otherwise; euros are converted at a rate of ~1.165.

1. Ukrainska Pravda — terminal speed cap, 14 October 2026. pravda.com.ua M
2. Reuters — the cap hits interceptors and medium-range strikes. devdiscourse.com M
3. WSJ via Kyiv Post — who was hit by the cap, 21 October 2026. kyivpost.com M
4. Reuters citing WSJ — Musk ordered a slowdown. staradvertiser.com M
5. Musk: Starshield belongs to the US government. cbsnews.com M
6. Letter of intent between the MoD, Nokia and ICEYE, 22 October 2026. aljazeera.com M
7. Zelensky: Ukraine is building a Starlink-like system, 4 October 2026. euromaidanpress.com M
8. Independent minimum-constellation computation for Ukraine (researcher). arxiv.org L
9. KeepTrack — geometric model of coverage over Kyiv. keeptrack.space L
10. Supercluster — real Starlink throughput at the front. supercluster.com M
11. Spaceflight Now — SpaceX price sheet: $74m in 2026, $70m in 2025. spaceflightnow.com M
12. SpaceX — prospectus, 2026: mass to orbit. sec.gov M
13. Ariane 6 launches in 2025–2026. en.wikipedia.org M
14. Ministry of Finance, draft 2026 budget (via DOU and Dev.ua). dev.ua M
15. Sikorski and Musk, 9 March 2025. aljazeera.com M
16. Telesat and MDA — contract for 198 Lightspeed satellites, 2023. kratosspace.com M
17. Air & Space Forces — SDA: $1.8bn for 126 satellites (York $382m, Lockheed $700m, Northrop $692m). airandspaceforces.com M
18. SDA — Tranche 2 Transport Beta awards. bcsatellite.net M
19. Rocket Lab — $515m for 18 SDA satellites. meritalk.com M
20. Years from decision to first launch (researcher’s calculation). en.wikipedia.org L
21. Years from decision to first service (researcher’s calculation). en.wikipedia.org L
22. Kyiv Post and DCD — Stetman’s 360-satellite plan. kyivpost.com M
23. Militarnyi — UASAT funding. militarnyi.com M
24. heise — SATCOMBw 4: the Bundeswehr’s own constellation. heise.de M
25. European Commission — IRIS² implementation agreement, 7 August 2026. euspa.europa.eu H
26. Atlas Institute — IRIS² and the satellite race: the €15.6bn cost estimate. atlasinstitute.org M
27. Fedorov asks Musk for Starlink, 26 February 2022. en.wikipedia.org M
28. CNN and USAID — Starlink terminals in Ukraine, 2022. en.wikipedia.org M
29. Interfax-Ukraine — Fedorov: over 50,000 terminals, September 2025. interfax.com.ua M
30. SkyCommons — satellite systems tracker: Ukraine, March 2026. skycommons.okfn.org L
31. Supercluster — Volodymyr Stepanets on terminals in Ukraine, March 2026. supercluster.com L
32. CNN — SpaceX letter to the Pentagon of 8 September 2022. kvia.com M
33. CBS News — Musk denied coverage near Crimea. en.wikipedia.org M
34. Gwynne Shotwell on limiting Starlink for drones, February 2023. en.wikipedia.org M
35. Reuters — Starlink in the minerals talks, February 2025. aljazeera.com M
36. CBS News — Zelensky on the cap, 17 October 2026. cbsnews.com M
37. Ukraine’s Ministry of Defence — Starlink terminal verification, 2 February 2026. mod.gov.ua H
38. Kyiv Independent — how Starlink terminals are registered. kyivindependent.com M
39. Reports on Russian terminals cut off, 5 February 2026. yahoo.com L
40. Jonathan McDowell — Starlink statistics, 26 October 2026. planet4589.org H
41. SpaceX — third-quarter 2026 results. sec.gov H
42. SpaceX — Connectivity segment, Q3 2026. sec.gov H
43. Reuters — Germany pays for OneWeb in Ukraine, April 2025. yahoo.com M
44. Circular-orbit period (calculation; checked against Turner, SSC16). s3vi.ndc.nasa.gov H
45. Satellite visibility duration (calculation). arxiv.org L
46. Turner — regional LEO constellations, SmallSat 2016. s3vi.ndc.nasa.gov H
47. Share of a constellation over Ukraine (researcher’s calculation). arxiv.org L
48. Satellite footprint by elevation angle (calculation). arxiv.org L
49. First-generation OneWeb: 648 satellites at 1,200 km. en.wikipedia.org M
50. Best inclination for 44–53°N (calculation). arxiv.org L
51. Iridium NEXT — operational and spare satellites (9 in orbit in 2019, 5 more in 2023). en.wikipedia.org M
52. Minimum constellation for Ukraine incl. the Black Sea, 41–53°N (researcher’s calculation). arxiv.org L
53. Jonathan McDowell — Starlink satellites by orbital shell. planet4589.org M
54. Pachler et al., MIT — comparison of constellations by filings. nitrd.gov M
55. Pachler et al., MIT — maximum data rate per satellite. nitrd.gov M
56. Stifel — Starlink economics model, June 2024. stifelinstitutional.com L
57. ISPreview — Starlink V2 mini capacity. ispreview.co.uk M
58. SpaceX — 2024 progress report: Starlink V3. ispreview.co.uk M
59. Mezha — Starlink on ground robots. oboronka.mezha.ua M
60. Minimum latency via GEO. en.wikipedia.org M
61. Physical latency floor in LEO (calculation). arxiv.org L
62. SpaceX — Starlink network update, July 2025. theregister.com M
63. Ookla — Starlink latency in the US, H2 2025. benton.org M
64. Supercluster, KeepTrack, Euronews — OneWeb terminal prices. euronews.com M
65. First-generation OneWeb: no inter-satellite links. en.wikipedia.org M
66. The Record — effects of the KA-SAT attack. therecord.media M
67. Janes — Ukrainian EW officials on Volna Kupol Garant, 1 October 2026. janes.com M
68. Secure World Foundation — Global Counterspace Capabilities 2026. swfound.org H
69. Harry Jones (NASA) — The Recent Large Reduction in Space Launch Cost, 2018. ntrs.nasa.gov H
70. Harry Jones (NASA) — launch prices in 2018 dollars. ntrs.nasa.gov H
71. CSIS Aerospace Security — cost of launch to LEO. ourworldindata.org M
72. Falcon 9 — payload to LEO. en.wikipedia.org M
73. Falcon 9 price per kg (calculation from list price). payloadspace.com M
74. Payload Space — effective Falcon 9 price per kg. payloadspace.com M
75. Outside estimates of Falcon 9 internal launch cost. en.wikipedia.org L
76. SpaceX — prospectus (S-1/A), 2026: launch counts. sec.gov H
77. Jonathan McDowell — orbital launches by year. planet4589.org H
78. OneWeb after losing Soyuz: LVM3 and Falcon 9 lead times. en.wikipedia.org M
79. SpaceX — prospectus: launches in Q1 2026. sec.gov M
80. SatNews — turbopump anomaly caps Ariane 6 plan. satnews.com M
81. Wikipedia — Ariane 6 price estimates. en.wikipedia.org L
82. Isar Aerospace — Spectrum reaches orbit on its second flight. en.wikipedia.org M
83. Light Reading — Amazon Leo: production versus launches. lightreading.com M
84. Wikipedia and TelecomTV — Starship Flight 14. telecomtv.com M
85. Motley Fool on Voyager Technologies’ 10-K: $90m per launch. fool.com L
86. Terzi, Nicoli — From Sputnik to Starship, PNAS Nexus, 2026 (model forecast). bennettschool.cam.ac.uk M
87. Terran Orbital — Rivada contract for 300 satellites. advanced-television.com L
88. Eutelsat and Aerospacelab — contract for 264 IRIS² platforms. aerospacelab.com H
89. SpaceX — terminal production cost, September 2023. driveteslacanada.ca M
90. Notebookcheck — Starlink Mini US price, October 2026. notebookcheck.net M
91. Reuters via Wikipedia — Starshield terminals for the UK. en.wikipedia.org L
92. OneWeb ground station in Angola: costs. mobileeurope.co.uk L
93. Hughes — OneWeb ground network contract. satellitetoday.com M
94. Euroconsult — optical inter-satellite terminal prices. nova.space M
95. ITU — cost-recovery fees for satellite filings. itu.int L
96. ITU — bringing into use of non-geostationary systems. itu.int H
97. ITU — Resolution 35 (Rev. WRC-23): deployment milestones. itu.int H
98. Iridium — 2025 annual results. sec.gov H
99. SpaceX — prospectus: average mass per launch. sec.gov M
100. SpaceX lowers ~4,400 satellites to 480 km, January 2026. en.wikipedia.org M
101. Telesat Lightspeed — satellite design life (2016 plan). spaceq.ca L
102. ESA — natural orbital decay times. universetoday.com M
103. European Commission — €90bn Ukraine Support Loan. eeas.europa.eu H
104. Starlink timeline from announcement to beta. en.wikipedia.org M
105. SDA — Tranche 0 contracts, 2020. en.wikipedia.org M
106. Breaking Defense — SDA resumes launches, July 2026. breakingdefense.com M
107. ICEYE — MikroSAR delivered to Poland in under 12 months. iceye.com H
108. ITU — filing procedure timelines. itu.int H
109. Telesat — Lightspeed global service slip. spaceq.ca M
110. Spaceflight Now — Iridium sold for about $25m, December 2000. spaceflightnow.com M
111. US government pays $72m to keep Iridium alive. en.wikipedia.org M
112. OneWeb — funding rounds and 2020 bankruptcy. en.wikipedia.org M
113. IISS (via Defence24) — the price of European military space autonomy. defence24.com M
114. GomSpace and Stetman — UASAT joint venture. satnews.com M
115. Space Intel Report — EU guarantees for UASAT. spaceintelreport.com M
116. Nokia and ICEYE — sovereign connectivity partnership, 1 October 2026. nokia.com L
117. esut.de — SATCOMBw 4: 40 satellites by 2029. esut.de M
118. EUSPA — IRIS²: first launches by 2029. euspa.europa.eu H
119. GOVSATCOM became operational on 27 January 2026. en.wikipedia.org M
120. Quwa — ASELSAN presents ASTRALINK. quwa.org M
121. Supercluster — Russia’s Rassvet constellation. supercluster.com L
122. Eutelsat — €1.5bn capital increase, 2025. eutelsat.com H
123. Eutelsat — shareholders after the capital increase. whitecase.com M
124. Eutelsat — order of 440 OneWeb satellites from Airbus. eutelsat.com H
125. Eutelsat — CENTAURE contract with DGA. eutelsat.com H
126. Taiwan and SpaceX’s majority-ownership demand. dtnext.in M
127. Taipei Times — TASA to launch four low-orbit satellites, 31 March 2025. taipeitimes.com M
128. DIGITIMES — South Korea launches a LEO communications programme. apps.digitimes.com L
129. Poland’s Digital Ministry — spending on Starlink for Ukraine. en.interfax.com.ua M
130. Bloomberg — Pentagon contract for Starlink in Ukraine. unn.ua M
131. Reuters (via LIGA.net) — SpaceX raised the Pentagon price fivefold. biz.liga.net M
132. Reuters — Italy discusses a SpaceX deal, January 2025. yahoo.com M
133. Ukrainska Pravda — drone schedules under the EU loan. pravda.com.ua M

Read this briefing in Ukrainian →