How much does it cost to implement digital public infrastructure?

Brazil built Pix, its instant payment system, for about $4 million. It now costs around $14 million a year to run.

Read that twice. The build was the cheap bit.

The short answer: implementing an open source digital public good (DPG) on top of a country's existing digital public infrastructure typically costs $250,000 to $500,000 for a pilot and $1 million to $2 million or more for a national rollout. Building the core DPI itself from scratch costs far more. The number moves with scope: event volumes, integrations, whether ID is included and how much of the government stack it has to talk to. Then add support, which most budgets forget.

Key numbers:

• Pilot on an open source DPG: $250k to $500k (my field ballpark)
• National rollout on an open source DPG: $1m to $2m+ (my field ballpark)
• A "conversion" building block on existing systems: around $750k, plus or minus 15%, over six months (CDPI)
• Pix in Brazil: about $4m to build, about $14m a year to maintain (DIAL)

A quick disclosure before we go further. I lead commercial partnerships at OpenCRVS. The ranges here are industry ballparks from my work across the sector, not OpenCRVS pricing.

Why does nobody publish what DPI costs?

Because almost nobody does. In June 2026, Ameya Naik at the Digital Impact Alliance wrote that in five years working on DPI, he couldn't recall a single publication on what any DPI actually cost to set up or maintain.

That silence has consequences. A ministry can find a dozen reports on DPI benefits. India's Aadhaar took the cost of an e-KYC check from $12 to six cents. Lovely. But when the finance ministry asks what their own system will cost to build and run, there's no benchmark to point at.

So governments either overpay or underbudget. Usually the second.

My take: the opacity suits anyone selling a proprietary system on a ten-year contract. If nobody knows what a national registry should cost, nobody can tell you you're paying three times too much.

What does a DPI implementation actually cost?

Two layers get lumped together here. DPI is the core infrastructure: the identity, payments and data exchange rails. DPGs are the open source products built on top of those rails to deliver services, such as civil registration. The table covers both, and the costs are very different.

These aren't like-for-like. PhilSys covered door-to-door enrolment teams and biometric kit, not just software. But the spread is the point. There is no single price for DPI, and anyone who quotes you one before asking about scope is guessing.

What pushes the cost up?

Four things drive most of the number in the projects I see.

1. Event volume. A registry handling a few thousand births a year is a different animal from one handling a few million. Volume shapes hosting, training and support.
2. Integrations. Every system it has to talk to is its own piece of work: health facilities, the statistics office, courts, social protection.
3. Whether ID is in scope. Civil registration and identity often come hand in hand. Bring ID into the project and you've changed its size.
4. Interoperability with the wider stack. Plugging into payments, data exchange and other government services adds design, testing and politics.

Scope creep does the rest. Birth and death registration is one thing. Add marriages, divorces, adoptions and name changes, and you're into fairly complex workflows.

Then there's the paper. Most of the registration offices I've visited hold decades of hard-copy records, some of them quietly rotting. Digitising and migrating those is a separate cost line, and if you get the digitisation wrong, the data migration becomes a nightmare. More in digitising historical civil registration records.

If the software is free, where does the money go?

"Is it really free?" is the question I hear most from governments. It sounds almost too good to be true.

It is free. A digital public good is open source, carries no licence fee and has no vendor lock-in. The money goes into making it work in your country: configuration, integrations, data migration, training, hosting and security. That work is usually delivered by a systems integrator, either a global firm or a local partner trained on the product.

So you're not paying for software. You're paying for people who know how to deploy it. That's a very different conversation from a licence negotiation, and a much better one. See is open source government software really free?

What do most DPI budgets forget?

Support.

Implementation gets the attention, the launch event and the press release. Support and annual maintenance contracts (AMCs) get treated as an afterthought, when they matter almost as much as the build.

Go back to Pix. Its running cost is roughly three and a half times its build cost, every year. Not every system scales like a national payment switch. But the direction is universal: the build is a one-off, the running is forever.

Before you sign an implementation contract, get answers to these:

• Who provides support after go-live?
• Where is it delivered from: a local partner in-country, or cross-border?
• What is the annual cost, and whose budget line is it on?
• Who handles upgrades when the DPG releases a new version?

My view: a DPI budget without a support line isn't a budget. It's a down payment. More in what happens after go-live.

Is cloud cheaper than a government data centre?

Sometimes, but cost is rarely what decides it.

Most DPGs can run on bare-metal servers in a government data centre. There's a growing push towards cloud, because it's faster and removes the dependency on government hardware. But civil registration and ID hold the most sensitive personal data a state has.

So there's a natural friction between keeping data in-country on your own servers and trusting it to a hyperscaler like Amazon, Google or Microsoft. Data residency rules settle it in some countries before the finance ministry gets a say. Model both options, but don't let a hosting quote make a sovereignty decision for you.

What I've seen in practice

Funders often set the number. In many of my conversations, it's the development funders who pull DPGs into the room: the World Bank, the Asian Development Bank, the Pacific Community. That means the budget often starts from the funder's envelope, not the ministry's cost model. Worth knowing before you plan. See how DPI is funded.

Time is money, literally. A pilot can go live in about three months. A full national implementation usually takes six to 12 months. Every month of delay is a month of team costs.

Lock-in is the most expensive line item nobody budgets. I've been pulled into conversations with several governments sitting on unpaid bills with proprietary vendors, disputes that have gone legal. They're locked in, getting nothing, and they end up moving because they have no choice. The cheapest DPI is the one you never have to rebuild.

How to budget a DPI implementation

1. Define scope first: which life events or services, how many per year, which regions.
2. List every integration, including the ones that are "phase two".
3. Decide whether ID is in or out, and write that down.
4. Price digitisation of historical records separately from the new system.
5. Add support and AMCs for at least three years, before anyone approves the build.
6. Choose hosting on data law first, cost second.

Sources

• How much does it cost to build DPI?, Centre for Digital Public Infrastructure
• We need an open conversation about the costs of digital public infrastructure, Ameya Naik, Digital Impact Alliance, June 2026

If you're budgeting a DPI or DPG implementation and want a second pair of eyes on the numbers, let's talk.