Small AI, big foundations: what the World Bank tells us about jobs, and what it changes for ID30

🤖 This version is an AI-assisted translation. Read the original in French →

Source: World Bank Group, “Small AI. Big Impact”, Global AI and Digital Summit (Seoul, 20 October 2026), and “Small AI: More and Better Jobs”, IMF and World Bank Annual Meetings (Bangkok, 12 October 2026).

In ten days, the President of the World Bank Group delivered the same message twice. In Bangkok at the Annual Meetings, then in Seoul at the first Global AI and Digital Summit, as shown in the session streamed by the World Bank, Ajay Banga put artificial intelligence at the service of one priority: jobs. I followed both sessions closely, because they say a lot about what lies ahead for the governments we support. Here is what I take from them, and what it changes for ID30.

Ajay Banga’s vision: AI as a job creation engine

The starting point: jobs first

Ajay Banga has made jobs the common thread of his presidency. In Bangkok, he recalled that the private sector creates between 75 and 85 % of jobs, and that the role of the state is to make that creation possible. He put it in one line: “It is private capital that creates jobs. Governments enable.”

AI exposure depends on income level

His most original point is about the geography of risk. In his view, large language models could affect about a third of jobs in advanced economies. That share drops to single digits in middle income countries, and the immediate effect would be close to zero in low income countries. The IMF, on the same Bangkok panel, uses a broader estimate: 40 % of jobs affected worldwide, 60 % in advanced economies, 23 % in low income countries.

The real risk: jobs that will never exist

Hence the reversal that struck me most. For the poorest countries, Ajay Banga’s first concern is not the loss of existing jobs. It is the missed opportunity: the jobs that will never exist if these countries stay away from AI. He also warns against a “new divergence” in growth and jobs between advanced and developing economies.

The answer: “Small AI”

Large models need five ingredients at scale: computing power, electricity, data, skills and capital. Very few countries hold all five. So Ajay Banga proposes another path: practical tools that solve one precise problem, with local data, local languages and modest hardware. His examples speak for themselves: women farmers in Uttar Pradesh who diagnose crop diseases on a basic phone, and tools in El Salvador that assess students’ language skills and support teachers in class.

He insists on one last point: a tool that works should move beyond the pilot stage and be taken to a hundred countries. And he recalls that digital lowers the cost of entering a market, so a small business can compete without building a large organisation.

The Group’s answer: a first digital strategy

The arithmetic of jobs

The figure behind the whole speech is stark. About 1.2 billion young people will reach working age in developing countries over the next 10 to 15 years. Their economies are on track to create around 420 million jobs.

1 billion people by 2030

In Seoul, the World Bank Group presented its first digital strategy, with one goal: to open digital opportunity to 1 billion people by 2030, including 600 million on broadband and 400 million using digitally enabled services. Four priorities frame it: expanding connectivity, making devices and services affordable, strengthening digital ecosystems, and preparing countries to adopt AI.

Physical foundations and trust

Two announcements complete the picture. The AI Infrastructure and Readiness Initiative aims to give emerging economies affordable computing through shared infrastructure, starting with Africa and Latin America. And Anna Bjerde, Managing Director of Operations, recalled the basics: around 2.2 billion people remain offline, no digital economy grows without reliable electricity, and Mission 300 has already connected 59 million people in Africa. Above all, she stressed trust: people adopt digital systems when they know their information and their money are protected.

My reading: small in the model, never in the foundations

A respite, to turn into a head start

The low exposure of low income countries comes from the structure of their economies. It is a window of time. It gives governments a few years to build the foundations before the pressure reaches service jobs.

Three layers under every use case

To reach 400 million people, a Small AI tool stands on three layers:

• Physical infrastructure: electricity, networks, and a deliberate choice of where data and computing live, between national data centres, shared regional capacity and cloud.
• Trust: identity, authentication, consent, data protection and cybersecurity. Trust is what turns a pilot into a public service.
• Governed AI: models, data and use cases running on these rails, under rules the state sets and controls.

We made the same point in June, reading the IFC and World Bank handbook on AI investment: digital public infrastructure sits at the base of the AI stack, and sovereignty now means owning the architecture underneath the model (our reading of the handbook).

AI, ally and risk for DPI

What AI already solves

For years, the hardest constraint on open source digital public goods was the team: large, skilled, hard to retain. AI reduces that burden. It helps document, test, maintain and evolve platforms that smaller national teams can now carry. It gives local ownership a real chance.

What AI exposes

AI also makes systems more vulnerable. Attacks get cheaper, faster and more convincing: tailored phishing, document fraud, voice and face impersonation. Every automated service opens a new door. Nepal’s identity security scare reminded us of it (a warning for every DPI rollout).

Three questions at the heart of our programmes

• How to reduce exposure: collect less, centralise less, open only what a service needs.
• How to protect what is exposed: security by design, trust frameworks and public key infrastructure, monitoring and incident response.
• How to build sovereign and resilient infrastructure: hosting, data centre and cloud choices, networks, and continuity when a link or a supplier fails.

These questions are already on the table in our projects. They will be our daily work in the years ahead.

ID30’s trajectory: one layer at a time

ID30 has always grown by widening its coverage (ID30 turns five):

1. Digital identity first: legal identity, civil registration and national ID systems, from Lomé to Apia.
2. Digital transformation of public services next: architectures, portals and the GovStack building blocks that turn an identity into a service.
3. Institutional and legal frameworks then, built with key partner law firms such as Jones Day: the laws, decrees and governance that make a system enforceable.

The next layer is the one Small AI stands on. Our LRP2030 strategic plan extends ID30’s scope to physical infrastructure (hosting, data centres, networks), cybersecurity and AI. We build this offer through strategic partnerships that increase the reach, scope and size of our 2027+ assignments (One day in Brussels). Our position stays the same: independent advice, on the side of the state.

Small AI, built first in our own house

We practise what we advise. Small AI is what we are building inside ID30 itself, for our own work:

• Sovereign infrastructure: our tools run on infrastructure we control.
• Locally hosted data: our knowledge, documents and registries stay with us.
• A modular approach: every building block can be replaced, and we keep our dependency on cloud AI models to a minimum.

This experience feeds straight into our advice. The trade offs we recommend to governments, between sovereignty, cost and performance, are the ones we make ourselves, every day.

What we recommend to governments

1. Start from a use case tied to jobs: one trade, one problem, one population, and an income or jobs indicator to track.
2. Build on existing rails: identity, payments and data exchange, rather than one isolated platform per project.
3. Map exposure before deploying: which data, where it lives, who accesses it, what happens if it leaks.
4. Decide on hosting with eyes open: sovereignty, cost, resilience, and a possible exit from every supplier.
5. Write security and reversibility into contracts: audit, data and model portability, and continuity clauses.
6. Train the national teams who will run these systems after the consultants leave.

Changing the question

The question for ministers is shifting. It used to be “which AI model should we buy?”. It is now “which infrastructure will carry our AI, and who owns it?”. The countries that answer the second question will turn small AI into many jobs. That is the road we want to walk with them.

Jaume Dubois, Founder and President of ID30

Sources: Small AI. Big Impact, World Bank (YouTube) · Small AI: More and Better Jobs, Annual Meetings 2026 · Aju Press, World Bank unveils first digital strategy · Asia News Network, Bangkok IMF World Bank talks · National Herald, AI can empower India’s farmers · Pattaya Mail, Bangkok summit

Want to comment, ask a question, or join the discussion? Continue the conversation on LinkedIn.