Why Every AI Project Eventually Hits Geospatial: A Primer on the Category That Anchors the Physical World

Why Every AI Project Eventually Hits Geospatial: A Primer on the Category That Anchors the Physical World

Primer · Geospatial
A field guide to the space under logistics, insurance, defense, and every AI project that touches the ground
$12.6B
GIS software market, 2026 (Precedence Research, 2026)
$123B
Geospatial analytics market, same year (The Business Research Company, 2026)
700K+
Organizations running Esri's ArcGIS platform (Esri, 2026)
9,000
Marriott properties monitored for risk on ArcGIS (Marriott, 2026)
Key Takeaway

Location is the enterprise data layer that stays tied to something physical. A customer record is an abstraction. A coordinate is a place a person can stand, which is why every AI project touching the real world eventually resolves against it.

start with what the category actually does, because the name hides it. A geographic information system stores what is where and reasons about why that matters. The distinction that trips up most buyers: consumer maps answer how to get somewhere. A geographic information system answers which of your forty thousand assets sit inside the flood zone, which crews are closest to each one, and what the failure history says about which to fix first. One is navigation. The other is a decision layer for anything that exists in physical space.

I am in San Diego from July 13 through 17 for the Esri User Conference, the largest gathering in the field, and I am treating it as a rare chance to learn this space at close range from the people who have run it for decades. I covered Esri the company last week ahead of the event. This is the wider view, written as a working primer for a technology leader who keeps meeting location data inside other systems and wants to understand the category it comes from before it becomes a dependency nobody chose.

What the category does

Underneath the map is a data model. Every road, parcel, pipe, vehicle, and service territory is stored as a feature with coordinates and attributes, and the software reasons about how those features relate. Proximity, overlap, containment, flow. Ask which substations lose power if one transmission line fails, and the system traces the network. Ask which addresses fall inside an evacuation radius, and it returns the list with the people and property attached.

That is the difference between a picture of a place and a model of it. A picture shows you the terrain. A model lets you ask it questions and route work off the answers. The commercial value of the category sits entirely in the second thing.

Key Takeaway

You are not buying maps. You are buying the ability to ask spatial questions and act on the answers, which means you are buying a data model and inheriting whoever built it.

Who uses it, and why most of them do not call it GIS

Governments run it at every level. All fifty US states, most national governments, and tens of thousands of cities and counties depend on it for zoning, emergency response, and public works. Utilities run it to manage the physical network. Telecom runs it to plan coverage. Insurance runs it to price catastrophe risk. Defense and intelligence run it for situational awareness. Logistics runs it for routing and yard management, and transportation alone accounted for close to a third of one recent read of category revenue (Mordor Intelligence, 2026).

The private-sector examples make the point better than the public ones. Marriott International runs ArcGIS as a global risk platform its security team calls the Risk Atlas, mapping more than 9,000 properties across 139 countries and monitoring each for natural and human-related hazards to support crisis response and climate adaptation (Marriott, 2026). A hotel company, using a geographic information system for security rather than mapping. On the retail side, chains use the same platform for site selection, scoring a location against demographics and drive-time before signing a lease, and Chick-fil-A and hospitality consultancy Horwath HTL are both public about doing exactly that (Esri, 2026).

Here is the part that matters for a technology buyer. Most of these organizations never bought a product called GIS. They met location data folded into a risk platform, a site-selection workflow, a fleet system, or a defense integration. The capability arrived embedded, which means the dependency did too.

The category is broad because location is a property of almost everything an operating business touches, not because the vendors sell aggressively.

The field divides into positions, not a single ranking

The useful way to see who runs the space is by layer, because a contract signed at one layer creates dependencies at the ones below it. A buyer who commits to a platform inherits its data model, its imagery sources, and its analytics assumptions whether or not anyone read them into the deal.

Platform
Esri, Hexagon AB, Bentley Systems. The full stack from data model to analytics to publishing. Esri's ArcGIS is the incumbent standard across government, utilities, and large enterprise. Hexagon and Bentley anchor the industrial and infrastructure side, where mapping meets engineering and the digital twin.
Imagery
Maxar, Planet Labs, Google. The pixels themselves. Maxar sells high-resolution satellite imagery to defense and commercial buyers. Planet Labs flies a large constellation for daily global coverage. Google Earth Engine pairs a multi-petabyte public archive with cloud compute, the default for planetary-scale environmental work.
Field & Survey
Trimble. Where location data gets born. Global positioning receivers, lidar, and survey hardware that feed everything upstream. Trimble and Esri have run a long partnership precisely because one captures and the other reasons.
Developer
Mapbox, CARTO, Google Maps Platform. Interfaces for teams building mapping into their own products rather than adopting a full workflow. Mapbox is developer-first and usage-based. CARTO is cloud-native and pitched at spatial data science inside the enterprise.
Location Data
HERE Technologies, TomTom. Roads, points of interest, real-time traffic, and routing tuned for navigation and automotive. The reference layer that fleet and mobility products build on.
Open
QGIS, the OSGeo community. Free, open-source desktop and server tools with a real base in research, education, and cost-sensitive agencies. The pricing floor every commercial vendor gets measured against.
Embedded
Autodesk, the ERP layer. Location capability folded into a system bought for another reason. Autodesk pulls authoritative geospatial reference data into design and planning. This is where most enterprises first meet spatial data without naming it.

Read down that list and the picture stops looking like a horse race. Trimble does not compete with Esri, it feeds Esri. Google competes at imagery and developer layers while partnering at the 3D-tiles layer. The market is moderately fragmented, with the top ten vendors holding roughly a quarter of revenue (The Business Research Company, 2026), which tells you no single vendor owns the category even though one owns the platform standard.

How the space changed, and the disruption that did not happen

Three shifts reshaped the category in the last few years. Cloud moved the software off the desktop into subscription platforms handling petabyte-scale data. The digital twin pulled GIS into engineering and simulation, so a city can now model flood risk or place electric-vehicle chargers on a live spatial base rather than a static map. And artificial intelligence made satellite imagery interpretable in minutes where it once took analysts weeks, while putting natural-language querying on top of the platform. Esri shipped generative AI into ArcGIS Pro this year and reported it cut analysis time roughly a third for non-specialist users.

The expectation a few years ago was that cheap models and satellite inference would flatten the field, that anyone could query an imagery archive and skip the platform. That displacement did not arrive. The natural-language query still runs against a data model somebody spent decades building, and lowering the skill needed to use the platform raised the value of owning the data underneath it.

AI was supposed to make the platform optional. It made the data model more valuable instead.

New entrants show where the energy went. NV5 launched an agentic geospatial platform this year that discovers and analyzes imagery, lidar, and drone data through natural language. EarthDaily Analytics absorbed Descartes Labs to fuse open-source and geospatial intelligence for government buyers. The startups are not rebuilding the platform. They are building agents and models that sit on top of it.

Where it goes next

Real-time and predictive is the direction. Streaming sensor feeds, on-orbit inference that classifies imagery before it lands, and spatial reasoning wired into agentic AI so a system can act on where something is, not just report it. The digital twin keeps expanding the addressable market past mapping into simulation and asset management.

The constraint that decides winners is not model quality. It is data sovereignty. Rules on geospatial personal data, and national mandates in India, Australia, and the United States pushing open standards, are forcing fresh procurement cycles and putting new limits on where spatial data can be processed and stored. The vendor that holds the authoritative data other systems resolve against is positioned to win regardless of who ships the best model on top.

Why this layer sits closest to physical reality

Most enterprise software categories model a business abstraction. A customer relationship record, a ledger entry, an inventory count. Useful, but a construct. Location is different because a coordinate points at a place that exists whether or not the software does.

That is why every AI initiative touching the physical world eventually resolves against it. A crew, a building, a shipment, a storm surge. The model has to meet the actual ground somewhere, and the coordinate is where it happens.

That physical anchor is the real reason to understand this category. The location dependency is probably already in your stack, arriving through a system you bought for another purpose. Knowing which layer it came from is the difference between a vendor relationship you chose and one you inherited.

CIO/CTO Viability Question

Trace one AI or analytics initiative in your organization that depends on location, and name the layer it resolves against. If the answer is a data model owned by a vendor you did not select, the question is not whether that vendor is good. It is what leaving would cost, and whether anyone has priced it.

Sources
Precedence Research. "Geographic Information System Market." Precedence Research, 10 Apr. 2026, https://www.precedenceresearch.com.
The Business Research Company. "Geospatial Analytics AI Market Report 2026." The Business Research Company, 28 Apr. 2026, https://www.thebusinessresearchcompany.co
Disclaimer: This blog reflects my personal views only. Content does not represent the views of my employer, Info-Tech Research Group. AI tools may have been used for brevity, structure, or research support. Please independently verify any information before relying on it.