Spatial data is treated as a specialized category of enterprise data. It is not. All data is spatiotemporal because everything exists somewhere and everything happens at some point in time. A maintenance record belongs to an asset at a location. A shipment moves through a sequence of places. A customer interacts with a store, service territory, facility, or network. A financial transaction is connected to people, companies, offices, assets, and markets that all exist within the real world. Even data that appears abstract becomes spatial and temporal once it is connected to the entity or event it describes.
This does not mean every dataset needs a latitude and longitude field. It means every enterprise architecture needs to understand how data relates to place, time, entities, and events. Traditional data systems tend to flatten those relationships into tables and transactions. Traditional GIS systems tend to express them as geometries and layers. AI needs the complete picture.
With the emergence of AI and, more recently, world models and digital twins, space and time will be native parts of the enterprise data foundation, not optional metadata or specialty functions. Spatial formats must move naturally through the same pipelines, notebooks, and AI workflows as conventional enterprise data. At the same time, traditional enterprise data must now leverage spatial and temporal context without the cost of being copied into a separate GIS environment.
AI does not recognize the departmental boundaries enterprises have created around data. An agent trying to answer an operational question does not care whether a relevant source belongs to the GIS team, the finance system, the maintenance platform, or the central data warehouse. It needs access to all of them, and it needs to understand how they fit together.
Consider a question about a pipeline anomaly. The answer may depend on sensor readings, maintenance history, inspection records, weather, soil conditions, imagery, land ownership, regulatory filings, and prior incidents. In most organizations, those sources sit across separate systems, use different identifiers, and are governed by different teams. The model may be capable of reasoning over that information, but the enterprise has not created the architecture that allows it to find and contextualize it.
Most enterprises will no longer need a completely separate geospatial warehouse, duplicated pipelines, and a segregated GIS data function operating beside the core data organization. In fact, most enterprises will not be able to afford such an architecture, because it will fragment their decision-making capabilities.
Three years from now, most enterprises will not talk about corporate data and geospatial data as separate disciplines. They will talk about enterprise knowledge, made available through one architecture to whatever employees, applications, models, and agents need to use it.
The GIS layer cake helped people see the world. The next architecture must help AI understand it.