One figure sums up the economic rupture that transformed the sector. An analysis drawing on the American space agency’s cost documentation notes that the dominant reusable launcher reduced the cost of placing payload in low Earth orbit from approximately 54,500 dollars per kilogram in the Shuttle era to under 2,700 dollars per kilogram as of 2024.
A factor of twenty on the line that was the principal barrier to entry in Earth observation. That fall changed the very nature of missions, making possible what had been economically inconceivable.
This article describes that transformation, its mechanisms and its consequences for available data. It extends the article on the Earth observation market.
The three mechanisms behind the Earth observation cost fall
It results not from a single innovation but from three developments that reinforced one another.
Launcher reusability is the first of those mechanisms. Amortising the manufacturing cost of a stage across many flights rather than one changes launch economics at its root.
Satellite standardisation is the second mechanism. A source in the field recalls that the CubeSat standard, developed in 1999 by two American universities, standardised satellite design by introducing a universal ten centimetre building block, the associated deployer acting as a universal interface between launch vehicles and satellites and eliminating custom integration work.
Rideshare launch is the third of them. Rather than chartering a whole launcher, an operator buys a slot on a shared flight, which divides the cost of access for small payloads.
Those three mechanisms combine and reinforce one another. Standardisation makes sharing possible, sharing makes access affordable, and reusability lowers the reference price of the whole.
What orbital access costs today
Public orders of magnitude allow the change to be measured, and they deserve stating with their caveats.
A sector analysis notes that the reference rideshare programme has set a market benchmark, with published prices as low as 325,000 dollars for a 50 kilogram payload.
That order of magnitude deserves two caveats. It covers launch alone, not design, manufacture, operations or the ground segment, which represent most of a mission’s cost. And it assumes accepting the orbit and schedule of a shared flight, a constraint a commercial constellation cannot always bear.
That second caveat explains the coexistence of two regimes. A university or demonstration mission tolerates rideshare constraints in exchange for a cost an order of magnitude lower; a constellation needing a precise orbital plane to generate revenue pays the premium of a dedicated launch.
The change of scale in Earth observation
The most visible consequence concerns the place small satellites now occupy in orbital activity.
A statistical compilation notes that small satellites represented under 2 per cent of orbital launch mass in 2013 and have become the leading spacecraft class globally.
Another source notes the intensity of launch activity, the American aviation authority having recorded over 250 commercial orbital launch attempts globally in 2023 alone, the highest annual total in history.
That change of scale has a direct consequence for available data. More satellites means more acquisitions, therefore a volume growing faster than human analytical capacity, the mechanism the opening article set out.
From satellite to constellation in Earth observation
A conceptual transformation accompanies the economic one.
The historical logic of Earth observation rested on a single, expensive satellite, carrying the best possible sensor and designed to last a decade. Its performance rested on instrument fineness.
Constellation logic rests on many satellites, individually less capable, frequently replaced. Its performance rests on revisit frequency rather than on fineness.
Three consequences follow. Temporal resolution becomes a differentiating factor at least as important as spatial resolution, whereas the opening article showed a physical trade-off binds those parameters on a single sensor. Frequent renewal allows technological progress to be incorporated by generation rather than by mission. And the failure of one satellite becomes an incident rather than a programmatic catastrophe.
That third point profoundly alters the sector’s risk appetite, and it explains the speed at which young companies were able to launch.
The new Earth observation capabilities
Four of them did not exist under the previous regime, and all rest on frequency.
Daily or sub-daily revisit over a given area, making it possible to follow fast phenomena rather than observe states.
Repeated global coverage over time, allowing homogeneous archives to be built across all land surfaces.
Diversification of sensor types, specialised operators having been able to launch on technologies the major programmes did not prioritise, notably the radar and hyperspectral described in the opening article.
And commercial responsiveness, with delays between order and delivery measured in hours for some suppliers.
Those four capabilities opened up uses that the articles on climate, security and insurance examine, and all rest on frequency rather than fineness.
What democratisation does not solve
The sector’s discourse dwells little on these limits.
The cloud constraint remains entire. The opening article noted that an announced revisit is an orbital capability and not a guarantee of observation, and multiplying optical satellites does not make clouds disappear.
Data quality does not follow satellite count mechanically. A miniaturised sensor offers radiometric and geometric performance below that of an institutional instrument, which limits some quantitative uses.
Homogeneity poses a new difficulty. A constellation contains satellites of different generations, with slightly distinct calibrations, introducing a heterogeneity exploitation must handle.
And durability is not assured. The market article noted the large number of entrants, not all of which will survive, which is a risk for a project depending on a single source.
The question of orbital congestion
It conditions the sustainability of the model and it is rarely raised commercially.
A statistical compilation notes that over 3,665 new debris objects were added in 2024, raising collision risks and prompting sustainability measures such as satellite deorbiting, debris tracking and mission planning.
That finding has three consequences for a data project. Deorbiting constraints shorten satellites’ useful lives, accelerating renewal and therefore the heterogeneity noted above. Regulatory developments could limit future deployment, making capacity projections uncertain. And sustainability becomes a selection criterion for some institutional clients.
That dimension is rarely addressed in commercial material and it is nonetheless a real variable of the sector’s future.
Shorter development cycles
This less visible consequence has changed the rhythm of the whole sector.
A source in the field notes that the small satellite model permits 18-month development timelines versus 7-year traditional programmes.
That shortening has three effects. Onboard technology corresponds to the state of the art at launch rather than to that of a decade earlier, which was not the case for major programmes. Design errors are corrected in the next generation rather than the next programme. And return on investment is assessed over a horizon compatible with private financing.
That third point largely explains the influx of private capital into a historically institutional sector.
One practical consequence follows for a project. A constellation’s characteristics evolve from one generation to the next, which requires documenting which generation a corpus was built from and forbids assuming a continuity the operator itself does not guarantee.
The Earth observation sensors benefiting most from the model
Not all technologies have benefited equally from democratisation.
Optical has benefited most in Earth observation, miniaturising a passive sensor being relatively accessible and the model’s economic validation having been established there first.
Radar benefited later, miniaturising an active instrument posing antenna and electrical power difficulties. Specialised operators have nonetheless established radar constellations, making accessible the all-weather acquisition the opening article described as reserved to major programmes.
Hyperspectral is benefiting now, several operators deploying dedicated constellations, which opens material characterisation uses hitherto confidential.
Thermal remains less served, the sensitivity required sitting poorly with miniaturisation.
That gradation indicates where data will become abundant in the short term, useful information for anticipating forthcoming annotation needs.
What democratisation changes for the data
It has direct consequences for the work of exploiting Earth observation imagery.
Available volume grows faster than analytical capacity, increasing dependence on automation and therefore on models, and therefore on training corpora.
Heterogeneity between sources becomes the norm rather than the exception. A project exploiting several constellations must handle differences in calibration, geometry and spectral bands, work the validation article details.
Frequency permits new temporal approaches, change detection and time series analysis replacing single-image analysis, which changes the nature of the annotation required.
That last point deserves developing because it is poorly understood.
Earth observation annotation in the era of time series
This development is specific to the constellation regime.
Moving from a single image to a series profoundly alters annotation in Earth observation, since annotating an image means describing a state while annotating a series means describing an evolution, which requires different conventions.
Four specific difficulties appear. Defining the moment of change, a gradual transition having no obvious date. Distinguishing real change from a variation in appearance, one parcel changing aspect with season without anything having happened. Handling missing acquisitions, a cloud potentially masking precisely the period of interest. And temporal coherence, an annotation having to remain stable over periods without change.
Those four difficulties explain why temporal annotation costs appreciably more than state annotation, and why it requires understanding the phenomenon observed rather than simple visual recognition.
They also constitute a real differentiation opportunity for a provider, that competence being markedly less widespread than still-image annotation.
The persistent role of major programmes
Democratisation has not made institutional missions obsolete, and the perspective is worth setting out.
Three functions remain specific to them and none of them delegates to the market.
Long-term continuity. A climate series requires homogeneous measurement across decades, a commitment a commercial operator cannot guarantee and which the climate article details.
Metrological precision. Some measurements require a calibration and stability only an expensive institutional instrument reaches, and they serve as the reference for calibrating others.
And free access, which the market article showed structures the whole economics of the sector.
That complementarity deserves understanding correctly. Commercial constellations do not replace major programmes; they densify observation between their passes and on segments they do not cover. A serious project generally uses both, which returns to the heterogeneity question raised above.
The ground segment, an invisible Earth observation bottleneck
This dimension escapes attention and is nonetheless a real constraint.
A satellite acquires data it must transmit to the ground, requiring receiving stations, visibility windows and link capacity. Multiplying satellites multiplies that need.
Three consequences follow. Downlink capacity sometimes limits acquisition, a satellite recording only what it will be able to transmit. The delay between acquisition and availability depends on the ground segment as much as on the orbit, which the near real time article details. And onboard processing is developing, some operators preferring to transmit a detection rather than a full image.
That last point deserves flagging for a data project. Onboard processing produces an output whose source data is unavailable, which makes it impossible to verify or re-annotate. A project requiring full traceability must therefore ensure it receives imagery rather than an already completed detection.
That requirement matches a principle earlier clusters established: the original data carries what its derivatives lost, and it is better retained than requested again.
What democratisation changes for a buyer of Earth observation data
Choosing a supplier becomes a decision in its own right.
The number of suppliers having risen sharply, that choice rests on four criteria rather than on price alone.
Sensor characteristics, determining what the data permits, the opening article having shown optical and radar sensors do not answer the same questions.
Effective revisit over the target area, distinct from announced revisit, and verified by requesting the actual acquisition history over a perimeter rather than a theoretical capability.
Archive depth, decisive for any time series analysis and favouring established operators over new entrants.
And licence conditions, which the market article noted may forbid delivering a derived product.
Those four criteria are checked before purchase and they remove most unpleasant surprises. The first two determine feasibility, the other two determine usability, and a project needs all four.
What this means for an Earth observation data provider
Four commercial consequences follow from everything above.
The first is that the number of potential clients grows. Each new operator and each new service provider is a potential need, and the market article noted the large number of entrants.
The second is that the need shifts towards the temporal annotation described above, a less widespread and therefore more valuable competence.
The third is that growing source heterogeneity increases the need for multi-source evaluation corpora, the only way to measure the robustness of a model exploiting several constellations.
A fourth consequence, less obvious, deserves flagging. Young companies in this sector have short decision cycles and constrained means, which favours targeted engagements and scoping over volume orders.
An Earth observation transformation that is not finished
A forward-looking reading is warranted, distinguishing the observable from the speculative.
Three Earth observation developments are under way and their continuation appears probable given the above.
Continued falls in access costs, several next-generation launchers being in development with further reduction targets. Caution applies to announced timetables, the field being accustomed to delays.
Extension of the model to active sensors, radar having already made the transition and other technologies following the gradation described above.
And the rise of onboard processing, with the traceability caveat noted above.
A fourth development is less certain and worth tracking: sector consolidation, the market article having noted the large number of entrants and the uncertain nature of their survival. A reduction in the number of operators would alter the structure described here without cancelling the cost fall that permitted it.
One observation nonetheless holds under any of these scenarios. The volume of available data continues to grow faster than analytical capacity, which keeps the bottleneck where the opening article located it.
What to verify before committing to a constellation
Five checks qualify a commercial Earth observation source before a project depends on it, and all five are answerable by the supplier.
Request the actual acquisition history over your area for the past two years, not the theoretical revisit. The gap between the two, described above, is where most disappointments originate.
Ask which satellite generations are in the constellation and whether their calibration is documented as consistent. The answer determines whether a time series is comparable across its span.
Establish the archive start date over your area. A supplier launched three years ago cannot support a ten-year trend analysis however good its current imagery.
Clarify what happens to already purchased data if the supplier ceases operations, a question the market article raised and which is rarely in standard terms.
And verify whether the delivered product is imagery or a processed detection, since the ground segment section above showed the difference determines whether the result can be verified at all.
Those five take one exchange with a sales contact. They are worth more than any comparison of published specifications, because they concern what the supplier actually delivers rather than what its constellation is capable of.
Common errors of reading
These misreadings recur often enough that naming them is usually enough to avoid them.
- Confusing launch cost with mission cost.
- Assuming that multiplying optical satellites resolves the cloud constraint.
- Expecting institutional instrument performance from a miniaturised sensor.
- Ignoring heterogeneity between generations within one constellation.
- Depending on a single source in a sector with many entrants.
- Handling a time series with annotation conventions designed for one image.
- Confusing real change with a seasonal variation in appearance.
- Neglecting missing acquisitions in a time series analysis.
- Assuming all sensor technologies democratise at the same pace.
- Ignoring orbital sustainability constraints in a capacity projection.
- Buying data without verifying effective revisit over the target area.
- Exploiting an onboard detection without holding the source imagery.
Why abundance made annotation scarcer
A closing observation captures the paradox running through this whole account of Earth observation.
Every mechanism described here increased the supply of imagery. Cheaper launch, standardised satellites, shared flights, shorter cycles, more operators: all of them push in the same direction, and the result is more data than anyone can look at.
None of them produced a single additional piece of ground truth. Knowing what is on the ground still requires a survey, an administrative record or an expert interpreting an image, and none of those became twenty times cheaper.
The consequence is a widening gap. Imagery has become abundant and annotation has not, which means the ratio between what can be observed and what can be learned from it is deteriorating rather than improving.
That is an uncomfortable conclusion for a sector that measures its progress in satellites launched, and it is the reason the bottleneck this series describes has not moved. It also explains why building annotated Earth observation corpora is a durable position rather than a transitional service: the forces making imagery cheap do not touch what makes it interpretable.
What to take away
The democratisation of Earth observation rests on a twentyfold fall in the cost of reaching orbit, achieved through launcher reusability, satellite standardisation and rideshare launch.
Three readings emerge. Constellation logic has replaced that of the single satellite, making temporal resolution a differentiating factor at least as important as spatial resolution and turning a satellite failure into an incident rather than a programmatic catastrophe. That transformation resolves neither the cloud constraint, nor the lower performance of miniaturised sensors, nor heterogeneity between generations within one constellation, three limits the sector’s discourse passes over. And the move from single image to time series profoundly alters annotation, which now describes an evolution rather than a state, with distinct conventions and an appreciably higher cost.
For a use this frequency made possible, the article on Earth observation and climate examines environmental measurement. For the economic structure in which this development sits, the article on the Earth observation market describes the players.
To explore delivery arrangements, supported formats and applicable control mechanisms, see our dedicated page on geospatial data processing. And if you work with time series requiring change annotation, let us discuss your project.
