The Next SDI — CARTO @ OGC Connect Helsinki

Hi, I'm Javier de la Torre, founder of CARTO. Excited to be here at OGC Connect. I don't need to tell this audience how big the AI transformation is — every software category is being redefined by the ability to ask in natural language and get answers. And I don't need to tell an audience in Finland about the geopolitical moment: data control, resilience, sovereignty are strategic requirements now. Today I want to connect those two forces — AI and sovereignty — to SDIs. Main message: SDIs are becoming more relevant than ever, but their requirements are changing. Historically SDIs served geospatial experts; in the AI era they must also serve agents — and those agents answer questions directly for citizens, planners, ministries, companies, governments. That de-intermediation makes SDIs more important — but it means a new architecture.

One quick word about CARTO. Born in Spain; today we work with leading enterprises, cities and public-sector organizations worldwide. Our focus: make geospatial more accessible, scalable and connected to the broader analytics ecosystem — moving organizations from traditional GIS into cloud-native spatial analytics. That's why we care about the future of SDIs, interoperability, sovereignty and AI. Geospatial is about to go through another interface shift. Think about maps: agencies had valuable authoritative data, but maps became part of everyday life when the interface changed — when anyone could use them instantly. The same is about to happen with SDIs: not only a portal for experts, but trusted infrastructure for agents that serve everyone.

For twenty years the SDI promise was discoverable, accessible, interoperable spatial data. Right vision — but the context changed. ① AI-ready: users aren't satisfied with portals, layers and endpoints; they expect to ask and get answers. ② Cheap and fast: agents query far more data than humans ever did; a heavy WFS or custom GIS service per request breaks the cost/performance model. ③ Interoperable with the rest of the analytics world: geospatial data is data — SQL, lakes, AI, cloud-native formats. ④ Sovereign: data, compute, AI and the GIS platform itself must be controlled, auditable, deployable under the org's rules. So the next SDI has four requirements: AI-ready, cheap and fast, interoperable, sovereign. Let me make it real.

AI isn't only changing software — it's creating massive physical infrastructure needs. Data centers need land, power, cooling, connectivity; they affect communities, raise environmental questions, become national digital infrastructure. Imagine a citizen or public-sector analyst in the Helsinki region reading about new data-center projects. They want to understand one contested site — without five geoportals, without knowing the dataset names or which catalog belongs to which publisher. They just want to ask: what's actually there? Close to power? Forest? Close to water? Protected? Environmental concerns? What evidence supports the answer? For the demo we federate three trusted sources: National Land Survey of Finland (infrastructure & physical context), SYKE (environmental datasets), Copernicus (Sentinel-2). The point isn't a giant new portal — it's that these sources are exposed as machine-readable catalogs an agent can use. Now let's ask the SDI.

▶ Open the live demo

SWITCH TO THE LIVE WEB APP (webapp/index.html → ask a question). The agent is connected to a federation of machine-readable spatial catalogs describing datasets, schemas, coverage, access methods, and where the cloud-native data lives. Ask: "I've been reading about proposed data centers in the Helsinki region. Investigate one contested site and assess the infrastructure and environmental context — use trusted Finnish and European open-data catalogues. Show me the datasets, queries, map, sources, and uncertainties." Flow: plan → search public reporting → three contested sites (Espoo/Hepokorpi, Kirkkonummi/Kolabacken, Vihti/Nummela) → choose Espoo → connect to the three publishers → discover datasets (NLS: sahkolinja, korkeusmalli_2m, jarvi, luonnonsuojelualue · Copernicus: Sentinel-2 NDVI · SYKE: Natura 2000, flood maps, CORINE — catalogued, convert-on-demand) → query. Honest results: transmission line ~64 m (favorable proximity — NOT capacity; depends on capacity & permitting). Elevation ~35.2 m, gentle slope, inland (no obvious topographic flood concern from the DEM; SYKE flood maps would sharpen it). Nearest lake ~931 m (cooling context, not a permission). NDVI ~0.66 — vegetated/forest: the catch. Nearest protected area ~830 m, no direct overlap (scopes review). Assessment: favorable infrastructure & terrain, but clears forest/green land at a residential edge — the core local objection, needs environmental review. Initial spatial screening, NOT a permitting decision. Every figure links to a publisher, dataset and query. THE POINT: no geoportal, no dataset names, no downloads, no manual joins. The catalog became a machine-readable contract for an agent. In the AI era, provenance is part of the answer.

▶ Open the live answer

First big requirement: AI-readiness. People expect to just ask. That does NOT mean GIS expertise disappears — the opposite: it becomes even more important, but encoded into metadata, catalogs, standards, tools, workflows and governance. For years SDIs served experts who knew which portal, which dataset, how to read the metadata, how to join layers. AI changes the interface: the next SDI serves agents, and agents answer questions directly for final users. Same shift as maps going from specialist-managed to something everyone uses daily — authoritative data stayed important, the interface changed. SDIs aren't less relevant; they're much more relevant — but the primary consumer becomes an agent. If agents answer questions about planning, climate, infrastructure, mobility, risk and environment, they need trusted SDIs underneath — machine-readable catalogs, metadata, provenance and standards matter more than ever. The future SDI is trusted infrastructure for agents that serve everyone.

Second requirement: cheap and fast. Agents query far more data than humans ever did; a server-side GIS service per interaction gets expensive fast. Old model: database → server → service → portal → tile caches → computation inside the SDI. New model is much simpler: object storage + CDN, cloud-native files, compute pushed to the analytical engine. With GeoParquet, Apache Iceberg, PMTiles, COG and cloud-native rasters the cost of serving the next SDI drops dramatically — not a custom API per dataset, not a GIS server per operation. APIs don't disappear — they stay critical for discovery, governance, access control, interoperability — but for large-scale analytical access, cloud-native files change the cost model. Here the data layer is hosted in Finland on UpCloud — the same architecture that makes SDIs cheaper makes them more sovereign.

Third requirement: interoperability — and I'll be direct: we need to break the GIS data silo. Geospatial data is data. It should live where the rest of analytics happens — usable from SQL engines, lakes, notebooks, AI agents, warehouses and GIS platforms — using formats the broader data industry already understands: GeoParquet, Apache Iceberg, columnar cloud-native files, SQL, DuckDB, BigQuery, Snowflake, Databricks. NOT about abandoning OGC — about making OGC and geospatial interoperability converge with the broader analytics ecosystem. OGC APIs remain essential for discovery, access, governance and machine-actionable interfaces; but the data itself must be computable by the modern stack. That's how we expand who can use spatial data. The next OGC client isn't only a desktop GIS or web map — it's an AI agent, and that agent needs standards, metadata, semantics, quality and trusted access.

Fourth requirement: sovereignty. Everything you saw runs on European infrastructure. Storage — UpCloud, hosted in Finland. Data — open cloud-native formats. Compute — DuckDB, an open engine. AI — Mistral, a European model, or self-hosted. GIS platform — CARTO, working toward SEAL-4-level sovereignty. Sovereignty in the AI era isn't just where data is stored — it's which model sees it, which engine processes it, which cloud hosts it, who controls the platform, can I audit it, move it, interoperate without giving up control. It's not about walls — it's controlling your infrastructure while staying interoperable. And it matters more in the AI era: whoever controls the agent controls the interface to the data, and whoever controls the interface increasingly controls the inference. Sovereign geospatial infrastructure is the whole stack: data, catalogs, compute, AI, GIS platform, governance. (Note: "not subject to the US CLOUD Act" is a strong legal claim — confirm with legal before the talk.)

This is where CARTO comes in — and as sponsors, let me be direct. Public-sector GIS is in a once-in-a-generation transition. The old model: portals, specialist users, proprietary workflows, expensive infrastructure. The new model must be cheaper, AI-ready, interoperable with the modern data stack, and sovereign by design. That's what CARTO is building — the GIS platform for the AI era: spatial data discovered by agents, queried directly from cloud-native formats, OGC APIs and open standards for discovery/governance/interoperability, modern engines for analysis, CARTO to visualize, analyze, govern and operationalize. Sovereignty is central: working toward SEAL-4-level sovereignty. The ambition: your data, your cloud, your AI, your engine, your GIS platform, your governance. If you're modernizing your SDI, geoportal or public-sector mapping platform, we'd love to work with you. The SDI vision was right. The interface has changed. The next interface is an agent. Thank you. — SHORT VERSION: SDIs were built for experts; now they must serve agents; agents serve everyone. That makes SDIs more important, not less — but they must become AI-ready, cheap & fast, interoperable, and sovereign by design. CARTO is here to help.