An earth observation company can build a good satellite, launch it successfully, and still fail — because the part of the business that costs money every single day is on the ground.
The cost that never stops
A spacecraft is a capital cost. It is spent once, amortised, and largely forgotten. The ground segment is an operating cost that scales with the size of the constellation and never goes away: antenna time, downlink bandwidth, data storage, processing compute and the engineers who keep the pipeline running.
Companies that modelled the ground segment as a fixed overhead during fundraising discover in year two that it is the dominant recurring line item.
Antenna economics
Ground station as a service has genuinely changed the picture. An operator no longer needs to build a global network before its first satellite reaches orbit, and for a small constellation the commercial option is clearly cheaper.
The crossover comes sooner than most plans assume. Past a certain number of passes per day, contracted antenna time costs more than owned infrastructure would, and the operator is locked into a contract structure that was sized for a smaller business.
The data problem underneath
The harder issue is what happens after the data lands. Raw downlink is not a product. Turning it into one requires calibration, correction, cataloguing and delivery within a latency window the customer cares about.
Every one of those steps is software that has to be maintained against changing spacecraft behaviour, changing customer formats and growing volume. It is unglamorous, it is expensive, and it does not appear in the pitch deck.
Nobody has ever raised a round on the strength of their processing pipeline. Plenty of companies have failed for want of one.
What to look for
When assessing whether a space company has thought this through, the useful questions are on the ground. What is the cost per pass, and how does it change at ten times the current constellation size? Who maintains the processing chain, and what happens when that person leaves? What is the actual latency from downlink to a customer-usable product, measured rather than designed?
The spacecraft, by comparison, is the straightforward part.
