A Cloud Native Geo Upgrade
for Spatial Data Infrastructure
Open Geodag 2026 · May 12
Chris Holmes
Planet
GeoServer
INSPIRE
The vision
A federated European spatial data infrastructure — discoverable, interoperable, openly accessible to all.
Upgrading
AI-First
Easy to Implement
Scalable
Low Cost
Sovereign
AI-First
Easy to Implement
Scalable
Low Cost
Sovereign
AI-First
AI
Current SDI was not built for AI or agents.
AI needs full data and rich metadata, fast.
Upgrade SDI so AI is a first-class citizen.
AI-First
Easy to Implement
Scalable
Low Cost
Sovereign
Easy to Implement
Database WFS Service WMS Service Schema Ops !
Today's SDI requires databases, services, and specialist staff.
The barrier is too high — most organizations can't participate.
AI-First
Easy to Implement
Scalable
Low Cost
Sovereign
Scalable
Traffic Cost Servers Files
AI agents crawl entire datasets, not single queries.
More popularity means more servers — until something breaks.
Infrastructure should scale without human intervention.
AI-First
Easy to Implement
Scalable
Low Cost
Sovereign
Low Cost
Servers always-on €€€
Staff ops engineers €€€€
Traffic spike emergency €€€€€
Sharing public data shouldn't be an organizational burden.
Server, staff, and licensing costs add up fast.
Popular datasets shouldn't cause budget shortfalls.
AI-First
Easy to Implement
Scalable
Low Cost
Sovereign
Sovereign
Control the full stack within your own jurisdiction.
SDI should not depend on foreign clouds or vendors.
Part II
Cloud-Native
Geospatial
A community-built set of formats that eliminate the need for a server — and scale through the power of cloud storage.
Cloud-Native Geospatial
COG
COG
Cloud-Optimized GeoTIFF
Zarr
Zarr
Cloud-Native Arrays
COPC
COPC
Cloud-Optimized Point Cloud
GeoParquet
GeoParquet
Columnar Vector Format
PMTiles
PMTiles
Single-File Tile Archive
STAC
STAC
SpatioTemporal Asset Catalog
Stream raster data for visualization and analysis with just HTTP range requests — no tile server needed.
Still a valid GeoTIFF, so every legacy tool reads it exactly as before.
COG demo
Cloud-native format for multi-dimensional data cubes — think HDF5 or NetCDF, built for the cloud.
Not backwards compatible like COG/COPC, but getting great traction in climate and earth science.
ESA is adopting Zarr for all Copernicus Sentinel missions.
Zarr demo
Stream point clouds for visualization and analysis — the COG equivalent for LiDAR data like AHN.
Still a valid .laz file, so it works with all existing LiDAR tools without conversion.
COPC demo
The cloud-native format for vector data analysis — query with SQL, no server needed.
Tools like DuckDB and Pandas read it natively, making AI integration natural.
GeoParquet demo
Packages map tiles into one file for visualization — no tile server needed.
Complements GeoParquet and COG: they handle analysis, PMTiles handles display.
PMTiles demo
Simple JSON format that adds metadata to any cloud-native format.
Search collections and items with STAC API or serverless with STAC-GeoParquet.
STAC demo
Part III
An SDI Without Servers
GeoParquet
COG
STAC
PMTiles
COPC
Zarr
Cloud-Native Spatial Data Infrastructure
Portolan
A new foundation for sharing geospatial data — faster, cheaper, and built for the age of AI.
Explore the demo live
portolan-sdi.github.io/portolan-nl-demo
QR code to demo
Portolan NL landing page
Explore the demo live
portolan-sdi.github.io/portolan-nl-demo
QR code to demo
Portolan NL browsing demo
QR code to demo
GeoParquet data browsing demo
Buildings PMTiles demo
Portolan styles demo
Mapbox style JSON example
STAC assets with style artifacts
COG aerial imagery demo
STAC Map interface demo
Portolan source view demo
Portolan viewers demo
Source Cooperative demo
Source Cooperative Portolan readmes and llms.txt
AI prompt to explore buildings data
Reading Portolan skill in Claude
AI exploring BAG Light dataset
Claude analyzing BAG buildings data
Claude's compliments on dataset formatting and follow-up suggestions
National Parks in Portolan NL
Claude performing spatial analysis on buildings and parks
Claude monuments analysis results
AI-generated tulip map application
Chat prompt asking about tulip fields
AI chat analyzing Dutch crop data
Part IV
How Portolan
Answers Each Point
Let's revisit the five requirements — now with Portolan as the answer.
AI-First ✓
Easy to Implement
Scalable
Low Cost
Sovereign
Built for AI
Portolan explains to AI in text files how to access the data, no API calls needed.
STAC and other metadata gives understanding of what's in the data, for more meaningful analysis and queries.
AI Agent STAC JSON GeoParquet INSPIRE schemas extracted as metadata Foundation Models World models · Agents
AI-First ✓
Easy to Implement
Scalable
Low Cost
Sovereign
AI Queries Your Data
AI translates natural language into GeoParquet queries — data answers questions directly.
Natural foundation for world models and agentic workflows.
"Where are the tallest buildings?"
-- AI-generated query
SELECT inspireId, name,
  heightAboveGround_value,
  ST_AsText(ST_Centroid(geometry))
FROM 's3://sdi/buildings.parquet'
WHERE heightAboveGround_value IS NOT NULL
ORDER BY heightAboveGround_value DESC
AI-First ✓
Easy to Implement ✓
Scalable
Low Cost
Sovereign
Ecosystem for Free
Standard formats mean a data browser and map come automatically.
Growing ecosystem will rapidly deliver more interactive tools.
GeoParquet · COG · STAC · COPC · PMTiles Portolan Data Viewer Map AI Chat QGIS Plugin All generated automatically from standard formats
AI-First ✓
Easy to Implement ✓
Scalable
Low Cost
Sovereign
Point, Convert, Push
"Claude, can you use the portolan skill to go through these data folders, extract relevant metadata, and publish a Portolan node to Source Cooperative?"
Portolan tools convert shapefiles, WFS, GeoPackage, BigQuery, or ArcGIS Feature Services into cloud-native formats.
AI tools excel at CLI + text files — non-experts can set up nodes.
AI-First ✓
Easy to Implement ✓
Scalable ✓
Low Cost
Sovereign
Scalability
No servers — scalability handled entirely by cloud storage.
Object storage is the most robust, proven layer any cloud offers.
0
servers to maintain
∞
concurrent requests
AI-First ✓
Easy to Implement ✓
Scalable ✓
Low Cost ✓
Sovereign
The Cost Picture

Only two costs: storage and egress. Click Edit to adjust estimates and compare providers.

Scenario Data Size Storage / mo. Transfer / mo. Egress / mo. Total / mo.
Municipality Vector (BAG, BGT, zoning)
Province Vector + orthophoto
National (PDOK-scale) All registrations + AHN + luchtfoto
Source Cooperative: free hosting for open government data € 0
AI-First ✓
Easy to Implement ✓
Scalable ✓
Low Cost ✓
Sovereign ✓
Sovereignty
EU flag
European digital
sovereignty
No foreign vendor in the stack — 100% open source, no servers.
Cloud-native just needs S3-compatible storage — many European providers qualify.
Leaseweb
Leaseweb
Netherlands 🇳🇱
Hetzner
Hetzner
Germany 🇩🇪
Scaleway
Scaleway
France 🇫🇷
OVHcloud
OVHcloud
France 🇫🇷
IONOS
IONOS
Germany 🇩🇪
MinIO
MinIO
Self-hosted
Any S3-compatible storage
Call to action
The Path Forward
Be the first.
A Dutch government Portolan implementation would be the world's first real proof point — and a model for SDIs across Europe, and the world. You have the open data culture, the technical capacity, and the geospatial ambition to lead.
Join the community
Portolan is open-source, community-governed. Contribute to the spec and tooling. Shape the direction.
Run a pilot
Pick a dataset, set up a Portolan node, share it. The barrier is low — we'd be glad to help.
portolan-sdi.org
Get the talk and explore the demos at
cholmes.github.io/open-geodag-presentation
QR code