The original Earth observation stack

The Earth observation industry didn’t begin as a software industry. It began as an infrastructure problem.

For decades, the hardest part of observing Earth was getting access to imagery at all. Satellites were expensive, launches were rare, downlink bandwidth was constrained, and processing pipelines were slow. Most organizations couldn’t operate space systems themselves, so the industry organized around one principle: control access to pixels.

every business model, workflow, and tool downstream inherited this one constraint.

That principle shaped everything that followed. The original Earth observation stack was designed around acquisition. Every company, workflow, and business model ultimately revolved around the movement of imagery from orbit to human operators on the ground. The result was an ecosystem optimized for collecting data, cataloging data, and selling data — not understanding it.

In the early years of remote sensing, satellites were essentially scientific instruments operated by governments and defense organizations. Programs like NASA Landsat transformed humanity’s ability to observe the planet. For the first time, it became possible to repeatedly image Earth at planetary scale. But access was limited. Imagery was expensive to capture and difficult to distribute. Analysts often waited days or weeks to receive datasets. Data arrived in specialized formats that required domain expertise to interpret. Most workflows depended on dedicated geospatial software running on isolated workstations.

Earth observation was a specialist ecosystem, not a developer one. Remote sensing experts became the operating system of the industry. They understood spectral bands, atmospheric corrections, coordinate systems, radiometric calibration, and sensor artifacts. The average engineer couldn’t simply build on top of EO infrastructure the way developers build on cloud infrastructure today.

The industry evolved accordingly. Instead of open platforms, EO became a chain of vertically integrated systems: satellite operators captured imagery, ground stations received it, processing systems corrected it, catalogs indexed it, analysts interpreted it, and governments and enterprises purchased reports derived from it. Every layer existed to support the movement of imagery through the pipeline.

As commercial Earth observation matured, competition centered on three metrics: resolution, revisit frequency, and coverage.

Higher resolution meant smaller objects could be detected. Companies competed to image roads, buildings, vehicles, ships, and eventually individual pieces of infrastructure. Revisit frequency determined how often a location could be observed. A single image of a port had value; daily monitoring had far more; near real-time monitoring became the long-term ambition. Coverage defined how much of the planet could be captured at scale. Governments wanted continental visibility, agriculture required seasonal imaging across massive regions, and climate monitoring demanded persistent global archives.

The entire industry became a race to optimize these dimensions — better sensors, more satellites, larger constellations, faster downlinks, cheaper launches. The assumption was simple: if more imagery could be captured, more value would emerge. For a long time, that assumption held true.

the unspoken premise of an entire generation of EO companies. it broke quietly, then all at once.

As satellite imagery became commercialized, a new layer emerged above operators: imagery marketplaces. These platforms aggregated datasets from multiple providers and exposed searchable catalogs for customers. Instead of negotiating directly with individual satellite companies, users could discover and purchase imagery through centralized interfaces. The business model resembled stock photography marketplaces, except the assets were observations of the planet.

Customers searched by location, time range, cloud cover, resolution, and sensor type. If existing imagery was unavailable, users could submit tasking requests to satellites for future collection opportunities. This became one of the defining workflows of commercial EO — find imagery, purchase imagery, task imagery, download imagery, analyze imagery. The industry standardized around scenes, tiles, and archives.

Even when APIs emerged, they existed to reach imagery catalogs programmatically. Most APIs still exposed pixels. Earth observation infrastructure became very good at moving images through systems, and almost nothing existed for understanding what those images meant.

Traditional remote sensing workflows were heavily manual. An analyst might spend hours or days searching imagery archives, evaluating cloud contamination, running preprocessing pipelines, performing atmospheric correction, calculating spectral indices, training classification models, and exporting shapefiles and reports. Every workflow required deep domain expertise.

Even simple questions became operationally complex.

The imagery rarely contained the answer, so humans had to derive it. That was the ceiling on the original EO stack. The system delivered observations, not intelligence.

It is easy to criticize the early Earth observation industry for fixating on acquisition. But that focus was rational. Acquisition was the bottleneck. Before reusable rockets, cheap launches, cloud computing, and modern AI, collecting planetary-scale imagery was an extraordinary engineering challenge. The industry optimized around the scarcest resource: pixels.

storage, marketplaces, APIs, business models. every layer inherited the same scarcity assumption.

Storage systems were built to archive pixels. Marketplaces were built to distribute pixels. APIs were built to retrieve pixels. Business models were built to monetize pixels. Very few systems were designed around semantic understanding.

EO evolved much like the early internet. The first phase connected computers. The second organized information. Only later did platforms emerge that hid the complexity and opened the door to new applications. Earth observation is entering that third phase now.

The original stack solved access to planetary imagery. The next one has to solve access to planetary understanding.