The Ground Segment Problem

A satellite that cannot get its data down is an expensive way to store pictures. Every founder in this industry knows that. Far fewer run their companies as if they do.

The architecture conversation in most young space companies is about the spacecraft. The payload, the bus, the orbit, the launch slot. Ask about the ground segment and the answer is short. We use a provider. Antenna time is a line item. It scales with usage. The serious engineering, by implication, happens in orbit, and the ground is plumbing to be rented when the plumbing is needed.

That view has a respectable history. For decades, ground infrastructure was a capital burden that only governments and the largest operators could carry. It meant antennas at remote high-latitude sites, licences in several countries, fibre back to a processing centre, and people to run it all around the clock. When Amazon Web Services launched its ground station service, it framed the problem in exactly these terms. The cost and difficulty of ground infrastructure, it said, had kept satellite data out of reach for all but the best-funded organisations. Its answer was first announced in late 2018 and became generally available the following May. Antenna time was billed by the minute, with no long-term commitment, and AWS claimed customers could cut their ground station costs by up to 80 percent. Microsoft announced a competing service in 2020. For a seed-stage company, the ground segment became something you switch on with a contract and a console.

The incentives point the same way. A launch is an event, with a countdown, a livestream and a press release. First light from a new sensor makes a good image for a funding announcement. A ground network coming online has no launch day. Nobody tweets a contact schedule. Founders spend their attention where investors and the press spend theirs, and the ground segment rarely earns a headline until something goes wrong with it.

The rental model was genuine progress, and I am not arguing against it. Renting ground capacity lets a company with three satellites reach coverage that once required a national programme. It converts a heavy fixed cost into a variable one, which is exactly what an early balance sheet needs. The consensus is not foolish. But it rests on an assumption worth naming. The ground segment, in this view, is a commodity input, interchangeable and priced by the minute, and the decisions that shape the business are made somewhere else.

That assumption is where it breaks.

The ground segment is not where a satellite's product arrives. It is where the product is made. Nothing a customer pays for exists until the data is down, processed and delivered, and each of those steps runs through choices made on the ground. Treat those choices as procurement, and they get made by default, on someone else's terms, at a moment when the company is looking elsewhere.

Start with physics. A satellite in low Earth orbit is in view of any given ground station for minutes at a time. The European Space Agency describes the typical case plainly. An Earth-observing satellite gets roughly ten minutes in every hundred to send its data down. Everything gathered in between waits on board for the next window. Location changes the picture sharply. Copernicus describes the Svalbard station run by Kongsberg Satellite Services as the only commercial ground station that can support polar-orbiting satellites on every orbit, about fourteen times a day. A mid-latitude site sees far fewer passes. One British operator, explaining a move to Svalbard in 2015, set up to fourteen daily passes there against up to six from its UK stations.

So the questions that define the product are ground questions. How old is the data when the customer receives it? How much of what the satellite collects can actually come down? How quickly can an urgent tasking request reach the spacecraft? None of these is an antenna-rental question. Together they are the product specification, written in contact windows.

The clearest evidence comes from Europe's own flagship programme. In 2016 the European Data Relay System entered service. It uses laser links through geostationary relay satellites, so Sentinel imagery can reach the ground within minutes instead of waiting for the next pass. Faster delivery was the obvious gain. The less obvious one was volume. ESA later reported that the Sentinel-1 constellation had increased the amount of data it produced by around half since it began using the relay. Same satellites, same instruments. Part of the ceiling on output had been sitting on the ground side of the link.

Then there is dependence. A company that rents its ground segment also rents part of its data pipeline, its scheduling logic and, often, the cloud environment where processing begins. That integration is the selling point. Data lands straight in one provider's storage and compute. Contact planning runs through one provider's tools. Over two or three years, the company's operations software grows around those interfaces. Even the regulatory layer can sit with the vendor. When its first stations opened, the head of AWS Ground Station described the company completing licensing in each country where it was building antennas. New customers, he said, would be added to its own ground station licence. That is a genuine convenience. It also means that the permission to talk to your satellite from a given site may belong, in part, to someone else. Moving then means securing frequency authorisations at new sites, qualifying new links, rebuilding parts of the pipeline and re-testing operations, all while satellites are flying and customers are waiting.

That sounds abstract until a provider leaves. Microsoft began installing ground stations at its own data centres in 2021 and, in October 2024, quietly retired the service, giving existing customers until 18 December to move. The antennas were sold to a leasing company and placed with another ground service operator, which gave customers somewhere to go. The lesson survives the soft landing. Customers of one of the largest technology companies in the world discovered that their ground segment had a strategic owner whose priorities had shifted. No contract clause converts that into a risk the customer controls.

The ground is also where command authority lives. One of the most closely studied attacks on a commercial satellite network never touched a satellite. On 24 February 2022, by Viasat's own account, attackers entered the management network of its KA-SAT system through a misconfigured VPN appliance and pushed destructive commands to user modems. Tens of thousands went offline across Europe. The spillover cut remote monitoring and control for 5,800 wind turbines in Germany. The spacecraft worked throughout. The service did not. Who holds command access, how it is secured, and what happens when a partner network is compromised are product decisions with customer consequences. They are not IT housekeeping.

If the ground segment is part of the product, the decision belongs at the start of the design, not after it. Three things follow for the people building these companies.

The first is that ground cost is not the price of antenna minutes. It sits in at least six places. Contact capacity, meaning enough of the right windows at the right latitudes. Latency, the time from capture to customer. Licensing, which recurs with every new site and every new country. Integration, the pipeline from antenna to processing to delivery. Control, meaning who can command the spacecraft and how that access is defended. And exit, what it would cost in time and engineering to move. Only the first appears cleanly on an invoice. The rest surface later, as slower delivery, missed tasking, delayed expansion, or a migration project nobody budgeted for.

The second is that renting usually remains right, but it should be a decision with a shape. Owning everything is rarely sensible for a young company. The real question is which parts are core. A business selling rapid-response intelligence lives or dies on latency and should own, or tightly control, whatever determines it. A business whose customers buy archive imagery on weekly cycles can rent far more freely. The mistake is not renting. The mistake is renting the part of the system the product depends on, without noticing that it does.

The third is that exit should be designed on day one, while it is cheap. That means keeping operations software from hard-wiring itself to one provider's interfaces. It means qualifying a second network before it is needed. And it means treating frequency filings as part of the expansion plan, not paperwork that follows it. None of this is expensive early. All of it is expensive late.

There is a cruel timing to all of this. The ground problem gets worse with success. A company flying three satellites rarely struggles for contact time. A company flying thirty finds them competing for the same windows over the same stations. That happens at exactly the moment its revenue starts to depend on delivery. Every satellite added is another set of passes to schedule and another configuration tied to the current provider. The arrangement that was sensible at three becomes the binding constraint at thirty, and the cost of changing it has grown with every launch in between.

For capital allocators, this makes the ground segment one of the more revealing questions in a diligence conversation, precisely because founders rarely raise it themselves. It is the capability-business gap at a smaller scale. A working satellite is a capability. A delivered data product is a business. The ground is where one becomes the other. A company that can explain its contact strategy, its latency budget and its exit path understands what it sells. A company that answers "we use a provider" may have outsourced a part of its product it does not know it has. Neither answer appears in a launch announcement. Both appear, eventually, in margins and retention.

Relay and optical links. Moving data through relay satellites or laser downlinks shifts part of the ground problem into orbit. It removes some constraints and creates new dependencies, because a relay network is another provider with its own capacity, pricing and priorities. I am watching whether early relay customers treat it as product architecture or as one more line item.

Who owns the antennas. The Microsoft exit landed softly, helped by a buyer for the hardware. The next exit may not be so tidy. When I look at an operator's ground arrangements, I care less about who sells the minutes than about who owns the steel. And what happens to the operator if that owner's strategy changes?

India's ground rules. India's 2023 space policy explicitly allows private entities to establish and operate ground facilities, including telemetry, tracking and command stations and satellite control centres. IN-SPACe's May 2024 norms then set out which ground systems need its authorisation. In February 2025, TRAI recommended that those ground station categories be exempt from separate telecom network authorisation. The network authorisation rules the Department of Telecommunications notified in July 2026 created six categories, none of them for ground stations of that kind. Those rules also require authorised networks to keep their systems and associated data inside India. I am watching whether that separation holds in practice. I am also watching how Indian operators answer a question that will matter more each year. When their data comes down through someone else's antenna, where does it physically land, and under whose rules?

The vocabulary of pitches. Resolution and revisit still dominate how space data companies describe themselves. I expect the more serious ones to start leading with delivery: how fast, how reliably, and through what. When that shift happens, it will be a sign that the industry has started pricing the ground segment as what it is.

The satellite is the part everyone photographs. The ground is the part every customer actually touches. A company that designs its ground segment deliberately has decided what it is selling. A company that buys it by default has let someone else make that decision, and will learn the terms later, usually at the worst possible time.

Chandrim Chatterjee · Orbit Intelligence Labs

APOGEE · Issue 5 · Sunday 4 October 2026