The Signal Gap

On July 2, in a joint announcement from Kigali and San Francisco, Rwandan President Paul Kagame, Salesforce chair and CEO Marc Benioff, and ITU Secretary-General Doreen Bogdan-Martin stood up a new institution with an old, unglamorous mandate: make sure the next technology revolution doesn't leave a quarter of the human race behind. They called it the AI for Good Global Commission. Forty founding members signed on — heads of state, tech CEOs, and the leaders of UN agencies — and their opening line of business was to say, plainly, what a lot of people in the industry have been reluctant to say out loud: 2.2 billion people are currently offline, cut off not just from artificial intelligence but from the internet that carries it.

That number is the whole story in miniature. It means roughly one in four people alive today has no realistic path to the tools reshaping medicine, education, and work for everyone else. The commission's founding language is careful to frame AI as a potential leveler — a way to leapfrog decades of underdevelopment in health diagnostics, crop forecasting, and literacy. But the same week it launched, a companion report from the United Nations delivered the harder verdict: the window to prevent AI from widening global inequality is closing, and it may already be narrower than anyone wants to admit.

A gap with a name, and a history

The "digital divide" is not a new phrase. It described who had electricity in the 1950s, who had a telephone line in the 1980s, who had broadband in the 2000s. Each time, the gap eventually narrowed — not because markets solved it on their own, but because governments, aid agencies, and international bodies treated access as infrastructure, not luxury. What is different about artificial intelligence is the speed of the divergence and the size of the entry ticket. Building or renting frontier AI capability requires not copper wire but some of the most concentrated, capital-intensive infrastructure ever built: specialized chips, gigawatt-scale power, and engineering talent clustered in a handful of metropolitan areas on two or three continents.

That concentration means the AI divide is compounding faster than the divides that came before it. A country that missed the broadband rollout of the 2000s could still catch up within a decade once costs fell. A country without sovereign compute, local-language models, or a functioning grid today risks missing not one wave of AI capability but several in succession, since each new generation of frontier models is trained on the last.

Where the intelligence actually lives

To understand why the commission's founders feel urgency rather than triumphalism, it helps to look at where frontier AI is physically built. The training runs behind today's most capable models are estimated to cost well over a hundred million dollars each, and the compute behind them is concentrated in a small number of hyperscale data center campuses owned by a handful of companies, sited overwhelmingly in the United States, China, and a few wealthy allied nations. Entire continents — most of Africa, much of South and Southeast Asia, large parts of Latin America — currently host essentially none of the infrastructure that trains frontier models, even as their populations increasingly use AI products built somewhere else, in languages that are not always their own.

This is the deeper problem the Commission is trying to name: access to a chatbot is not the same as access to AI capability. A farmer in Rwanda who can query a crop-disease model over a basic smartphone is meaningfully better off than one who cannot. But the country hosting no compute, training no models on its own languages and problems, and educating no AI engineers of its own remains a consumer of someone else's intelligence rather than a participant in building it — permanently downstream of decisions made elsewhere.

What the commission actually proposes to do

The AI for Good Global Commission is not a regulator and has no enforcement power; it is closer to a standing coalition, modeled loosely on public-health partnerships that pool money, expertise, and political will across borders. Its stated priorities are threefold: expand meaningful access to compute and connectivity in underserved regions, strengthen public trust in AI systems through shared standards, and accelerate the use of AI against concrete problems like disease surveillance and climate adaptation. It arrives alongside — and is meant to feed into — the UN's own Global Dialogue on AI Governance, a separate diplomatic track convening in Geneva to hash out international norms for the technology's development and use.

Bogdan-Martin, whose agency has spent decades trying to close the older, plainer gap in basic internet connectivity, framed the stakes in terms that echo that history directly: AI's benefits, she has argued, will only reach everyone if the infrastructure underneath it — power, spectrum, fiber, skills — is built out deliberately, not left to follow investment wherever it is already cheapest and most profitable to expand.

The skepticism in the room

Not everyone greeted the launch as good news. Commentary in the days after described the Commission through the lens of "splintered sovereignty" — the observation that global AI governance is fracturing into competing blocs (an American approach built around private labs and light-touch rules, a Chinese approach built around state-directed champions, a European approach built around binding regulation) even as a new commission tries to paper over those divisions with shared language about trust and access. Voluntary coalitions of this kind have a mixed track record: they can genuinely redirect money and attention, but they rarely bind anyone to anything, and the hardest resource to reallocate — compute itself — remains squarely under the control of a small number of private companies and the states that host them.

There is also a harder question the commission's own materials gesture at but do not resolve: is the goal to give the rest of the world access to systems built elsewhere, or to help the rest of the world build its own? Those are very different projects. The first is a subscription. The second is sovereignty. Rwanda, notably, has spent recent years positioning itself as a hub for African AI and cloud infrastructure — which makes the country's leadership of a global-access coalition also, not coincidentally, a bid for regional leadership within it.

What actually closing the gap would take

Set against the scale of the problem, what would meaningful progress look like within a few years, rather than a few decades? Researchers and development economists who study the digital divide point to a familiar, unglamorous list: reliable electricity and connectivity built out as public infrastructure rather than left to the market; regional or shared compute facilities that smaller countries can access without hosting their own gigawatt-scale data centers; language and open datasets that make AI systems actually usable in Swahili, Bengali, or Quechua rather than only in English and Mandarin; and a pipeline of local engineers trained to build and maintain systems rather than only to operate someone else's dashboard. None of it is exotic. All of it is expensive, slow, and politically unglamorous compared to a summit announcement — which is exactly why coalitions like this one exist, to keep the unglamorous work funded once the cameras leave.

A test case already underway

Rwanda is a useful place to watch precisely because it is not a neutral bystander in this story. Over the past several years, Kigali has positioned itself as a regional hub for African cloud and AI infrastructure, hosting data-center investment and courting international partnerships in an explicit bid to become the continent's on-ramp to advanced computing rather than merely a market for it. Kagame's co-leadership of the new commission is, in that light, both a genuine act of coalition-building and a fairly transparent piece of national strategy: if the rest of Africa is going to gain compute access through international partnership rather than building everything domestically, Rwanda would very much like to be the country those partnerships flow through.

That dual motive does not discredit the effort — national interest and global benefit are not mutually exclusive, and plenty of past infrastructure gains (vaccine manufacturing hubs, undersea internet cables) have arrived exactly this way, through a mix of altruism and self-interested positioning. But it is a useful reminder that "closing the gap" is not one project with one beneficiary. It is dozens of overlapping national projects, each trying to make sure that when new intelligence infrastructure gets built, it gets built somewhere they can reach.

The AI for Good Global Commission will be judged less by its founding roster of forty prominent names than by what gets built in the countries that were not part of the announcement. If compute gets cheaper and more available in Kigali, Jakarta, and Lagos over the next several years — if models start speaking more of the world's languages, and more of the world's engineers start training rather than only using them — the July 2 launch will look, in hindsight, like the moment institutions caught up to a crisis they saw coming. If it does not, it will look like one more group photo taken at the edge of a widening gap.