Geospatial foundation models Market Outlook 2026–2035: Growth Drivers, Key Companies, Strategic Developments & Adoption Forecast — A Meticulous Next™ Foresight Brief
What This Brief Covers
This Meticulous Next™ brief examines geospatial foundation models, large models trained on satellite and aerial data to perform land, climate and infrastructure tasks with little task-specific training. It tests whether earth observation moves from specialist image analysis to general-purpose embeddings and models that any developer can call, and who captures the value in data, models and applications.
The first proof is orbital deployment. NASA reported in May 2026 that its open-source Prithvi model, developed with IBM, had been uploaded to the Kanyini satellite and to a Thales Alenia Space payload aboard the International Space Station, where a compressed version was tested for flood and cloud detection. Running a foundation model in orbit means analysis can happen before data is sent to the ground, and adding a task needs only a small decoder package.
The second proof is commercial packaging. Google announced Custom Satellite Embeddings in private preview on 29 July 2026, built on AlphaEarth Foundations and offering embeddings for custom regions and time periods at intervals as fine as every five days. A technology company converting a research model into an on-demand data product shows how the market will be sold: as embeddings and interface calls, not as image archives.
The third proof is the pace of model releases. Open and academic models including Prithvi-EO-2.0, TerraMind from IBM and the European Space Agency, and OlmoEarth from the Allen Institute for AI have appeared since late 2024, each with self-reported benchmark wins on different test suites. No neutral leaderboard exists, so buyers may need months of internal testing, which slows adoption and favours vendors that supply evaluation.
The fourth factor is cost and distribution. A global 10 metre embedding product can run to hundreds of terabytes, and data transfer fees can exceed the cost of running the model. Licences range from research-only to commercial. Distribution and licensing terms will shape who builds products as much as model quality.
It is a focused 101-page decision brief for satellite operators, cloud and model providers, insurers, agriculture firms and governments. It presents an indicative trajectory rather than a segmented market model. Its purpose is to identify which land and climate tasks adopt first, who captures value, and which indicators to track.
| Parameter | Details |
|---|---|
| Forward horizon | 2026–2035 (9 years) |
| Emerging force | Geospatial foundation models: general-purpose models and embeddings trained on satellite data for land and climate tasks |
| Technology readiness | Open models and embedding products available; orbital demonstration completed; no neutral benchmark; production use concentrated in agriculture, forestry and disaster mapping |
| Indicative market size & forecast | The geospatial foundation models market, covering foundation models trained on satellite and aerial data, embedding data products and platform and application programming interface services for land, climate and infrastructure tasks, is valued at USD 0.5–0.8 billion in 2026 and is projected to reach USD 5.7–9.1 billion by 2035, at a CAGR of 31% |
| Mainstream inflection | ~2031, when embedding and model services are a default input to land, climate and risk workflows at large users |
| Signal strength | Prithvi in orbit (May 2026); Google Custom Satellite Embeddings private preview (July 2026); steady open model releases since 2024 |
| Primary beneficiaries | Cloud and model providers; satellite data operators; insurers, agriculture firms and governments that buy faster monitoring |
| Brief length / format | 101 pages · PDF + executive summary deck · instant delivery |
Understanding the Technology
Foundation models are the first layer. Models pretrained on years of satellite imagery learn general representations that adapt to tasks such as crop mapping, flood detection and land change.
Embedding products are the second layer. Compact per-pixel vectors let users run classification and change detection without handling raw imagery.
Satellite data supply is the third layer. Optical, radar, hyperspectral and elevation data feed the models, and persistent cloud cover favours radar and fusion.
Edge and on-orbit inference is the fourth layer. Compressed models running on satellites cut the data that must be sent to the ground.
The limits are the lack of neutral benchmarks, distribution costs, incompatible licences and the need for local labels. The decade turns on standard evaluation and low-cost access.
Market Outlook
The geospatial foundation models market, covering foundation models trained on satellite and aerial data, embedding data products and platform and application programming interface services for land, climate and infrastructure tasks, is valued at USD 0.5–0.8 billion in 2026 and is projected to reach USD 5.7–9.1 billion by 2035, at a CAGR of 31%, and excluding raw satellite imagery sales and conventional geographic information software sold without foundation model features.
The 2026 base rests on model access and embedding data services, platform subscriptions and custom model work for governments and enterprises. From that base, growth follows the area monitored at high frequency, the number of applications built on embeddings and the share of satellite data consumed through models.
The forecast is sensitive to licensing terms, benchmark standardisation and satellite data prices. Regionally, North America leads on cloud providers and model developers, Europe on public programmes and open models and Asia-Pacific on agriculture, disaster monitoring and fast-growing satellite constellations.
Scenarios
The base case assumes steady model releases and gradual standardisation, giving the inflection around 2031. An accelerated case assumes a neutral benchmark and low-cost distribution emerge, bringing the inflection to 2029 and lifting the 2035 value to the top of the range. A delayed case assumes licence fragmentation and weak evidence of accuracy gains, holding growth at the bottom of the range.
Factors Behind Growth
Growth drivers
- Demand for frequent, low-cost land monitoring.
- Regulatory needs such as deforestation compliance.
- Open models from public agencies and research bodies.
- Cloud providers turning research models into products.
Enablers
- Large archives of open satellite data.
- GPU computing.
- Compressed models for on-orbit use.
- Cloud storage and interface services.
Restraints and barriers
- Lack of neutral benchmarks.
- Data transfer and storage cost.
- Incompatible licences.
- Need for local labelled data.
The Forces at Play
Five converging forces will determine how fast geospatial foundation models scale.
Adoption Outlook
How the shift is likely to unfold across three time horizons.
Embedding and model services move from preview to general availability, and benchmarks begin to converge.
Large insurers, agriculture firms and agencies use models for routine monitoring and risk scoring.
Near-continuous monitoring of land and infrastructure runs through models, with on-orbit inference common.
Latest Strategic Developments
| Date | Development | Type | Significance |
|---|---|---|---|
| Dec 2024 | NASA, IBM and partners release Prithvi-EO-2.0 with open weights | Product | Opens a major geospatial foundation model to developers |
| Apr 2025 | IBM and the European Space Agency present TerraMind, an any-to-any model across nine data modalities | Product | Extends models to multiple data types |
| Jul 2025 | Google DeepMind introduces AlphaEarth Foundations for annual global satellite embeddings | Product | Sets the pattern for embedding data products |
| Nov 2025 | The Allen Institute for AI releases OlmoEarth with an Earth Systems platform | Product | Adds an open model with a platform |
| May 2026 | NASA reports Prithvi running in orbit on the Kanyini satellite and a payload aboard the International Space Station | Demonstration | First reported foundation model of its kind in orbit |
| Jul 2026 | Google announces Custom Satellite Embeddings in private preview, built on AlphaEarth Foundations | Product | Brings on-demand embeddings at intervals as fine as five days |
Key Players & Competitive Landscape
The key players operating in geospatial foundation models include Alphabet Inc. (Google DeepMind and Earth Engine), National Aeronautics and Space Administration, International Business Machines Corporation, European Space Agency, Allen Institute for AI, Clay Foundation Initiative, Planet Labs PBC, Maxar Technologies, Airbus SE, Microsoft Corporation, Amazon Web Services, Inc., Environmental Systems Research Institute (Esri), Pixxel Space Technologies, Inc., Satellogic Inc., ICEYE Oy, EarthDaily Analytics Corp., Thales Alenia Space, NVIDIA Corporation, Hugging Face, Inc., Ubotica Technologies Limited, BlackSky Technology Inc., Capella Space Corp., Umbra Lab, Inc., Spire Global, Inc., Orbital Sidekick, Inc., Descartes Labs, Inc., Bayer AG, Deere & Company, Swiss Re Ltd., Munich Re AG, Woolpert, Inc., NGIS Pty Ltd, University of Cambridge, Jülich Supercomputing Centre, SmartSat Cooperative Research Centre, Adelaide University, ESA Φ-lab, U.S. Geological Survey, National Oceanic and Atmospheric Administration, Copernicus Programme and the European Commission.
The brief profiles representative players in each archetype and assesses which are positioned to lead as embeddings become a product.
The competitive landscape is forming around six archetypes. Cloud and model developers. Public agencies and research bodies. Satellite data operators. Geospatial software and analytics firms. End users in insurance, agriculture and government. Regulators and data policy bodies. Competitive intensity is moderate in 2026 and is expected to rise as cloud providers sell embeddings directly through 2031.
| Archetype | Representative players | Position in 2026 | Outlook to 2035 |
|---|---|---|---|
| Cloud and model developers | Alphabet, Microsoft, Amazon Web Services, NVIDIA, IBM, Hugging Face, Allen Institute for AI | Models, compute and distribution | Set the pace on model capability and access |
| Public agencies and research bodies | National Aeronautics and Space Administration, European Space Agency, Clay Foundation Initiative, University of Cambridge, Jülich Supercomputing Centre, SmartSat Cooperative Research Centre, Adelaide University, ESA Φ-lab, U.S. Geological Survey | Open models and archives | Release open models and data |
| Satellite data operators | Planet Labs, Maxar, Airbus, Pixxel, Satellogic, ICEYE, EarthDaily Analytics, BlackSky, Capella Space, Umbra, Spire Global, Thales Alenia Space, Ubotica | Imagery and on-orbit processing | Supply data and edge inference |
| Geospatial software and analytics firms | Esri, Woolpert, NGIS, Orbital Sidekick, Descartes Labs | Applications and services | Package models into workflows |
| End users | Bayer, Deere & Company, Swiss Re, Munich Re | Demand and requirements | Pay for monitoring and risk scoring |
| Regulators and data policy bodies | National Oceanic and Atmospheric Administration, Copernicus Programme, European Commission | Data access and rules | Define open access and compliance duties |
Where value migrates.
In 2026 value sits with cloud and model developers that own compute and distribution. By 2031 it moves toward data operators and software firms that supply proprietary imagery, local labels and workflows. By 2035 it settles in services that provide continuous monitoring and risk scoring to insurers, agriculture and governments. Margins fall for raw imagery and generic embeddings and increase for tailored, verified decision products.
Who Will Win — and Why
The archetypes best positioned to capture value as the shift matures.
Firms that turn research models into low-cost services.
Firms that supply differentiated inputs and on-orbit inference.
Firms that package models for specific sectors.
Regulatory Landscape
| Jurisdiction | Milestone | Indicative timing | Effect on adoption |
|---|---|---|---|
| EU Deforestation Regulation | EU rules requiring deforestation-free supply chains | Phased application | Drives demand for land monitoring |
| EU AI Act | EU regulation on artificial intelligence systems | Phased application | Frames AI systems used in public decisions |
| Copernicus data and information policy | EU policy on free and open satellite data | In force | Provides open training and monitoring data |
| USGS Landsat open data policy | US policy on free access to Landsat data | In force | Supports model training and monitoring |
| NOAA commercial remote sensing licensing | US licensing of private satellite operators | In force | Governs commercial imagery resolution and distribution |
Investment Signals
The signals converge on productisation. A public model reached orbit, a technology company moved from a research model to a data product and open models kept arriving. Each signal says geospatial foundation models are becoming a product category, while accuracy gains remain hard to compare.
Research activity reflects where the chain is weakest: neutral benchmarks, distribution cost, licences and local labels. The brief therefore tracks four indicators: area monitored at high frequency, applications built on embeddings, share of satellite data consumed through models and benchmark standardisation.
Regionally, North America leads on cloud providers and model developers, Europe on public programmes and open models and Asia-Pacific on agriculture, disaster monitoring and constellations. Agriculture and forestry, disaster mapping and insurance are the first demand pools in that order.
Questions This Brief Answers
Strategic Implications
- Cloud and model providers: publish benchmarks and simplify licences to reduce buyer testing time.
- Satellite operators: pair proprietary imagery with embeddings and test on-orbit inference.
- Software firms: build sector workflows that turn embeddings into decisions.
- Insurers, agriculture firms and governments: run side-by-side tests on local data before committing.
- Investors: treat area monitored, embedding applications, share of data consumed through models and benchmark standardisation as the leading indicators.
"The planet has been photographed for decades and read by few. The models change who gets to read it and how often."
Table of Contents
Access & Licensing
A focused foresight brief, priced to circulate. Every option is delivered instantly and backed by analyst support.
every brief