The world is splitting into two AI economies.

What the "AI divide" means

A three-part gap, and it runs deeper than "who has a chatbot"

Economists and policy bodies increasingly describe AI's rollout not as a single global rising tide, but as three overlapping fault lines: a gap between nations that command AI's capital and compute, a gap between the individuals equipped to use AI and those who aren't, and a gap between the countries that write AI's rules and the ones excluded from doing so.

Each of those fault lines is measurable, each is documented by a different institution — Stanford's AI Index, the OECD, the IMF, Microsoft's AI Economy Institute, and the United Nations among them — and every figure that follows survived independent, adversarial fact-checking against its original source before it was allowed onto this page. Claims that didn't hold up were dropped; the methodology and what got cut are laid out at the bottom.

1. I.The National Divide — who owns the investment, the compute, and the chips.
2. II.The Human Divide — who has the connectivity, literacy, and income to use what gets built.
3. III.The Governance Divide — who gets a vote in the rules, and who doesn't.

I.The National Divide

AI's capital and compute are concentrating in one country — and one factory

Private AI investment has never been more lopsided. In 2024, U.S. private AI investment reached $109.1B, nearly twelve times China's $9.3B; U.S. generative-AI investment alone outspent the combined total of China, the EU, and the UK by $25.4B.¹ By 2025 the gap had widened again: U.S. private AI investment hit $285.9B against China's $12.4B, inside a global corporate AI investment pool that had itself more than doubled year-over-year to $581.7B.³⁴

That growth wasn't even. Greater China's organizational adoption climbed 27 points and Europe's climbed 23, the fastest gains of any region — yet North America still holds the largest absolute share of adopting organizations, a lead built on the investment numbers above.¹

And the money doesn't track neatly with everyday use. The United States, for all its spending, ranks only 24th in the world for consumer AI adoption at 28.3% — well behind Singapore (61%) and the UAE (64%). Worldwide, generative AI reached 53% adoption within three years of release, faster than the PC or the internet did, and adoption still tracks closely with a country's GDP per capita.³ Capital concentrates before use does — which is exactly why the next fault line matters as much as this one.

II.The Human Divide

You can't prompt a chatbot in a language you can't read

Underneath the national numbers is a more basic constraint: literacy. The IMF projects that AI's contribution to productivity growth in low-income countries could run at roughly half the rate seen in advanced economies — a projection that is explicitly conditional on how adoption unfolds, not a guarantee — and traces much of that gap to schooling. An estimated 59% of the population in low-income countries, and 32% in lower-middle-income countries, has no formal schooling at all; literacy sits at just 63% and 78% respectively. Large shares of people in those countries cannot read well enough to hold a conversation with a text-based AI system, let alone benefit from one.⁵

Microsoft's own researchers put it plainly: "uptake in the Global North grew nearly twice as fast as in the Global South."⁷ That figure is drawn from Microsoft's Windows-ecosystem telemetry rather than an independent census, so it should be read as the company's own best estimate rather than a settled count — but it's the most current, granular adoption data publicly available, and every corroborating data point points the same direction: the gap is not closing. It's widening, and it's widening because the wealthier half of the world is still accelerating.

III.The Governance Divide

118 countries have no seat at the table where AI's rules get written

The imbalance isn't only economic — it's procedural. Of the world's countries, only seven (Canada, France, Germany, Italy, Japan, the UK, and the U.S.) are signed on to every one of the major non-UN AI governance initiatives sampled by the UN's own advisory body. One hundred and eighteen countries, overwhelmingly in the Global South, are party to none of them.⁸

Funding follows the same pattern. Of the grants awarded to AI projects addressing the UN's Sustainable Development Goals between 2018 and 2023, only 10% went to organizations based in low- or middle-income countries. A further slice of private capital — nominally 25% — is recorded as flowing to those countries, but more than 90% of that slice went to China alone, leaving the rest of the developing world just 3% of total funding despite representing 7% of the firms in the underlying sample.⁸

The answer, so far

The UN just built a room for this conversation. It opens this week.

In August 2025, all UN member states adopted General Assembly Resolution A/RES/79/325, creating the Global Dialogue on AI Governance — the first standing multilateral venue built explicitly to close these gaps rather than simply debate AI safety.⁹

The Dialogue's mandate includes a thematic cluster titled, in its own words, "Bridging AI divides: capacity-building, access and digital foundations" — aimed squarely at computing infrastructure, applications access, and workforce skills in lower-income countries, alongside a push toward open-source and open-data AI models as a partial remedy.⁹ Its first substantive sessions on that specific cluster are scheduled for July 6–7, 2026, in Geneva — this week, as this page goes live.⁹

Why it's your problem too

This isn't a story about other countries. It's a preview of your labor market.

The same divides that separate nations run inside them. A firm with capital adopts AI-driven tools years before a smaller competitor can justify the spend; a worker with the literacy, language, and connectivity to use AI compounds a productivity advantage that a colleague without those things cannot close by working harder. The 10.6-point Global North/South adoption gap and the 24th-place U.S. consumer-adoption ranking documented above aren't abstractions — they're the same mechanism (capital and skill compounding faster than access spreads) playing out at every scale, from continents down to a single office.

None of this is fixed yet. The Global Dialogue on AI Governance is one week into its mandate to bridge these divides, not one year — and the experts surveyed by the UN's own advisory body don't expect the payoff for lower-income countries for the better part of a decade. Whether that timeline holds, and whether the 118 excluded countries gain real influence rather than a seat with no vote, is the open question this page will keep tracking.

Follow the UN Global Dialogue →

Method

How this was checked

Every statistic and quotation on this page passed a three-vote adversarial review against its original source: each claim was independently checked three separate times, and a claim needed a majority to survive. Anything not traceable to a specific, checkable sentence in a primary or credible secondary source was cut — including several strong-sounding numbers that didn't hold up.

Several figures are time-sensitive by nature. Microsoft's adoption percentages are a snapshot from H2 2025 / Jan. 2026 and rest on the company's own Windows-ecosystem telemetry rather than an independent census. The IMF's productivity-gap estimate is an explicit projection, conditioned on how adoption unfolds — not a settled outcome. And this dataset, while strong on macro statistics, is thin on named, ground-level case studies from individual Global South countries; that remains an open gap in the reporting, not just in the world it describes.