The global AI divide is often pictured as a row of countries waiting for access to the newest model. That image begins too late. A model is not useful when electricity is unreliable, connectivity is unaffordable, the interface does not support a local language, a clinic has no secure workflow for using the answer, or a small business cannot reorganise work around the tool.
The World Bank estimates that 2.2 billion people remain offline and that low-income countries account for less than 1% of ChatGPT usage. Those figures do not measure every form of AI use, but they reveal the size of the access gap before questions of model quality even begin.[2]
AI readiness is a stack
The foundations beneath a useful AI service
| Layer | What must exist | Failure when it is missing |
|---|---|---|
| Power and devices | Reliable electricity and affordable hardware | The service cannot be reached or used consistentlySource 1Source 4 |
| Connectivity | Affordable data and usable networks | Cloud tools become slow, costly or unavailableSource 2 |
| Local fit | Relevant language, data and workflow design | An impressive model produces answers that do not match the settingSource 1Source 3 |
| People and firms | Skills, management capacity and organisations able to change work | Access remains a demonstration rather than a productivity gainSource 1Source 3 |
| Institutions | Privacy, safety, procurement and accountability | Adoption creates distrust, exclusion or unmanaged harmSource 1Source 4 |
This is why the World Bank’s development argument is deliberately less glamorous than a race to train national frontier models. It proposes a sequence: adopt tools already available, adapt them to local conditions and advance towards more sophisticated systems as foundations strengthen. Smaller models that run on ordinary devices can be more consequential than a distant frontier system when bandwidth, cost or data sovereignty are binding constraints.[1][2]
India shows how readiness can compound
India illustrates both the opportunity and the dependence on complements. The World Bank reports that private AI investment rose from $1.2 billion in 2024 to $4.1 billion in 2025, while employment in global capability centres increased from 1.9 million to 2.36 million. The country also has a large technical workforce, digital public infrastructure and an internationally connected services sector. Those assets make adoption easier; they do not guarantee that small firms, rural users or public institutions receive the same benefits.[3]
Two Indian growth indicators
| Measure | Value | Context |
|---|---|---|
| Private AI investment | $1.2 billion to $4.1 billion | Reported increase from 2024 to 2025.Source 3 |
| Global capability-centre employment | 1.9 million to 2.36 million | Reported increase from 2024 to 2025.Source 3 |
The distribution question is about organisations
A country can have model access and still lack the firms that know how to combine it with reliable records, redesigned processes and trained workers. The same is true in government. A chatbot attached to incomplete benefit data does not repair the registry; a diagnostic model does not create referral capacity; an agricultural assistant does not substitute for local weather, market and extension systems. AI amplifies the quality of the surrounding institution as often as it compensates for scarcity.
At the IMF–World Bank meetings, Kristalina Georgieva placed electricity and digital infrastructure among the preconditions for broadly shared gains. The warning is useful because ethical guardrails and physical capacity are not alternatives. A country needs both. The AI divide begins with who can switch the system on, but it widens or narrows according to who can turn an answer into a service, a job or a better decision.[4]
Sources & further reading
- AI offers lifeline to developing economies in an era of weak growth World Bank · 4 Aug 2026 · Accessed 16 Oct 2026
- Small AI, big impact World Bank · 2 Apr 2026 · Accessed 16 Oct 2026
- India Development Update — October 2026 World Bank · 6 Oct 2026 · Accessed 16 Oct 2026
- IMF chief says AI risks may exceed opportunities without safeguards and ethical framework Reuters · 16 Oct 2026 · Accessed 16 Oct 2026