The ESRI Trap

When the Vendor Leaves, So Does the Map

In 2022, after the Sindh floods submerged a third of Pakistan, we had satellite damage assessment data sitting in a project folder. The imagery was current. The damage polygons were accurate. Provincial government, federal disaster management, the relief coordination bodies: all three needed it.

The master agreement with the satellite provider didn’t allow resharing with clients.

We worked without the data. The government worked without the data. The contract had been signed to produce exactly this kind of evidence for exactly this kind of emergency, and the licensing clause that nobody had flagged at procurement made it unavailable to the people it was procured for.

Nobody had done anything wrong. The vendor had delivered what the contract specified. We had signed what the contract required. The clause was standard. That’s the problem. A standard clause, in a standard master agreement, during one of the worst climate disasters of the decade, made the evidence base inaccessible at the moment it mattered most.

I have seen versions of this in almost every country I have worked in across Asia and Africa. The specifics vary. A city planning department in East Africa whose parcel data lives in a geodatabase that requires proprietary software to open, and whose license expired when the project closed. A coastal risk assessment in South Asia where the flood model outputs were delivered in a format the counterpart ministry couldn’t read without the vendor’s platform. A land administration project where the title records were digitized into a system the government technically owned but practically couldn’t operate without the vendor’s ongoing support.

The data exists. The government paid for it. It can’t leave.

This is what vendor lock-in looks like in geospatial infrastructure. It arrives quietly, embedded in procurement documents that nobody reads carefully enough, and it reveals itself at the worst possible moment.

The Map Africa Initiative

In July 2025, ESRI, Space42, and Microsoft announced the Map Africa Initiative: high-resolution base maps across all 54 African countries, covering 1.4 billion people over five years. The problem the initiative addresses is real. African cities are making planning decisions, collecting taxes, managing disaster risk, and allocating services without accurate spatial data. The deficit compounds every year.

The initiative’s public announcements say that data will be licensed to national governments, enabling ownership and long-term updating through National Mapping Agencies. That’s the right answer to the ownership question. It’s also the only question the announcements answer.

What the public record doesn’t address is portability and format. Data licensed to a government that can only be accessed through an ESRI platform is owned in a narrow legal sense and captive in every practical one. A national mapping agency that receives its country’s parcel data as an ArcGIS geodatabase, stored in Microsoft Azure data centers managed by G42 and Microsoft, and updated through ESRI’s production workflows, has legal title to an asset it cannot freely use elsewhere. The ownership claim is real. The operational dependency underneath it is also real.

This is the subtler version of the Sindh problem. In Pakistan, the licensing clause was blunt: no resharing. In Map Africa, the dependency is architectural. A government that builds its planning infrastructure on a proprietary stack, in vendor-managed cloud facilities, cannot easily move when the commercial relationship changes, the pricing shifts, or the political conditions that made the partnership attractive no longer hold. The map and the infrastructure dependency arrive together.

The questions that would resolve this are straightforward: What formats will the data be delivered in? Can governments export it to non-ESRI systems without additional licensing? What are the terms for accessing data that sits in vendor-managed data centers? None of these questions are answered in the public documents. That matters before any African government commits its national spatial data infrastructure to this architecture.

When the Platform Changes

On February 27, 2026, Anthropic was removed from USAi.gov, the GSA platform that centralizes AI procurement for U.S. federal agencies. By March 26, a preliminary injunction had restored access.

GSA’s OneGov platform had saved $1.1 billion in its first year. That concentration created the efficiency. It also created the single chokepoint. When vendor access changed, agencies using Anthropic through the centralized vehicle lost it overnight. The same architecture that produced the savings produced the vulnerability.

The Sindh case and the USAi.gov case look different on the surface. One is a satellite data contract in a disaster context. The other is AI procurement inside a high-income federal agency. The structural failure is the same: centralized procurement, insufficient portability provisions, and a dependency that reveals itself at the moment you most need continuity.

For geospatial infrastructure in Africa, the architecture risk runs deeper than in the U.S. case. A federal agency that loses access to an AI model can, with difficulty, find an alternative. A national mapping agency that has built its entire spatial data infrastructure on a proprietary platform, in vendor-managed cloud facilities, with workflows integrated into one vendor’s software ecosystem, has a much harder problem. The switching cost isn’t inconvenience. It’s years of data work and institutional capacity that doesn’t transfer.

When You Can’t Negotiate Anymore

The moment at which a government has most leverage in a geospatial technology relationship is before adoption. After adoption, the leverage has already transferred.

A government negotiating its first base map contract can insist on open data formats, interoperability requirements, and explicit exit rights: what the government owns at project close, in what format, and what it can do with that data without restriction. Those terms cost nothing to ask for when the contract is being drafted. They become very expensive to extract later, when the government’s planning systems, tax databases, and disaster response workflows all depend on the incumbent platform’s continued cooperation.

This is the procurement decision that determines the next decade of operational options, and it gets made once, by officials who are usually focused on the map, not the contract. The technical capability of the platform, the resolution of the imagery, the accuracy of the base map: these are what the procurement conversation is about. The portability clause is in the appendix. The exit rights are in the boilerplate. By the time the dependency is visible, it’s too late to negotiate.

Seoul’s experience with geospatial data offers a partial illustration of what the alternative looks like. When Seoul publicly disclosed its geospatial data and introduced ordinances allowing exchange between public and private sectors, it created conditions where multiple service providers could build on the same data layer. The data became infrastructure rather than a proprietary asset. That required a deliberate policy choice, made before any single vendor’s platform became the de facto standard. The window for that choice closes at the moment of adoption.

The EAC Fund Question

In April 2026, the East African Community announced a Regional AI Fund: eight governments pooling resources for shared compute infrastructure, framed explicitly as a sovereignty hedge. The instinct is correct. A shared, collectively governed compute facility removes the single-vendor chokepoint that USAi.gov demonstrated. It creates a negotiating position that no individual African government has alone.

The governance question is unanswered. Who decides what runs on the shared compute? Who sets data governance rules for the outputs? What happens when member governments disagree on acceptable use or data residency? The EAC fund has the right architecture. Whether it becomes a sovereignty instrument or a shared dependency depends entirely on whether those governance questions are settled before the technical infrastructure locks in.

A regional compute facility with no data portability requirements is a larger version of the same trap. The scale changes. The mechanism doesn’t.

Three Conditions Before Signing

The Sindh episode was not a procurement failure in the conventional sense. The contract delivered what it specified. The failure was in what the contract didn’t specify, and what nobody thought to ask for before signing.

Three conditions would have changed the outcome, and apply equally to any geospatial data initiative, including Map Africa. First, format portability: outputs must be deliverable in open, non-proprietary formats that the government can use in any system, not only the vendor’s. Second, interoperability: the system must be capable of exchanging data with other platforms without requiring vendor involvement or additional licensing. Third, exit rights: the contract must specify what the government owns at project close, in what format, stored where, and what it can do with that data without restriction.

These are procurement minimums. A government that insists on them still needs the institutional capacity to act on them. Open formats don’t help if there’s nobody trained to use them. Portability clauses don’t matter if there’s no alternative platform to move to. The conditions are necessary, not sufficient. But a government that can’t get these terms from a vendor should ask why, and what the refusal implies about the relationship it’s entering.

What the Map Doesn’t Show

The satellite imagery from Sindh in 2022 was accurate. The damage polygons were current. The analysis was real. It existed in a folder somewhere while the government tried to coordinate a response without it.

That data is probably still there. The floods are over. The clause that made the data unavailable is still in the standard master agreement.

The next generation of geospatial AI tools is more capable than anything that came before. Foundation models trained on global satellite imagery can detect buildings, map flood extent, estimate crop damage, and assess infrastructure condition with little local calibration. Sentinel-1D, now fully operational, gives cloud-proof radar coverage over every city on the continent. The tools are better. The procurement frameworks are the same vintage as the ones that failed in Sindh.

Map Africa will produce detailed, accurate base maps. That matters. Fifty-four countries that currently make major decisions without reliable spatial data will have an evidence base they don’t have now. The question is on what terms, and what those terms mean for the decade of infrastructure decisions that will be built on top of the maps.

A map that a government can’t freely use is not a public good. It’s an asset on someone else’s balance sheet.