---
title: "Who Pays for Intelligence?"
slug: who-pays-for-intelligence
canonical: "https://narain.io/writing/who-pays-for-intelligence"
author: "Narain Jashanmal"
description: "Rich markets could finance AI for the rest of the world. Whether that distributes power—or merely access—depends on how the technology diffuses."
topics: [AI, Political Economy, Technology, Infrastructure]
date: "2026-09-02"
last_modified: "2026-09-02"
---

# Who Pays for Intelligence?

*Rich markets could finance AI for the rest of the world. Whether that distributes power—or merely access—depends on how the technology diffuses.*

In the 1990s, Motorola promoted [Iridium](https://en.wikipedia.org/wiki/Iridium_satellite_constellation), its satellite-phone network, as a way to connect everyone on Earth, including remote communities neglected by existing infrastructure. Technically, its satellites could reach almost anywhere. Economically, the service was impractical. The expensive handset and high call charges made Iridium a product for a small number of wealthy travellers and governments.

The poor supplied the project's moral legitimacy. The rich were supposed to supply its revenue.

This pattern is now familiar. A company develops a technology around the economics of affluent markets, then describes its theoretical availability as a universal human benefit. Technical reach becomes practical access; practical access becomes social progress; corporate expansion becomes synonymous with making the world better.

Meta retained the rhetoric but solved Iridium's economic problem. WhatsApp functions as something close to a public utility across India and much of South America: family telephone network, small-business operating system, marketplace, school bulletin board and interface to public services. Users pay Meta nothing. Meanwhile, most of Meta's revenue comes from advertising in richer markets, where a user may be worth many times more than one in India. Advertisers pay because consumers in those markets have greater purchasing power, and the spending those advertisements generate ultimately supports their advertising budgets. The economic chain therefore runs from affluent consumers, through advertisers and Meta, to WhatsApp users elsewhere.

The result is a strange form of international redistribution. Western consumer spending indirectly finances a communications service producing enormous consumer surplus elsewhere. More precisely, the financing crosses borders; the surplus is created locally. And it is not charity. Low-revenue users contribute network effects, strategic dominance and future commercial possibilities. Everyone gets utility, but Meta retains ownership and control.

**Consumer surplus is not sovereignty.**

Bill Gates now uses similar language about artificial intelligence. AI, [he argues](https://www.gatesnotes.com/home/home-page-topic/reader/a-turbulent-ai-era-and-critical-choices-to-make), could become the greatest equalizer ever invented. The emblematic beneficiaries are again the rural patient, the underserved student and the low-income farmer receiving advice previously available only to the rich.

These benefits are plausible. But so is the underlying commercial structure: American and European companies will eagerly pay for AI that can replace or augment expensive labour. Their spending could finance cheap or free assistants across the rest of the world. The cloud-AI version of this future looks much like WhatsApp: immense global utility, funded disproportionately by rich markets and governed by a handful of foreign companies.

AI, however, may permit a different outcome.

A large frontier model can potentially teach or be distilled into many smaller, task-specific models. These models might diagnose crop disease, translate a local dialect, guide a community-health worker or help someone navigate a government bureaucracy. Combined with local databases and small real-time feeds for weather, prices or medical guidance, they could run cheaply on phones, computers and regional servers.

If that works, intelligence need not remain a continuously metered service rented from a distant cloud provider. It can become a locally possessed capital good: copied, adapted and operated without requiring permission for every use.

Rich-market demand would still finance much of the costly frontier research. But instead of indefinitely subsidizing API calls for poorer users, that research could produce cognitive tools whose marginal cost approaches zero and whose ownership travels with them. This would resemble technology transfer rather than platform welfare, as users would acquire the means to provide a service instead of receiving it for free.

That possibility should not be confused with inevitability. Written knowledge is not all knowledge; much expertise is tacit, local or absent from the internet. Small models may remain unreliable precisely where errors are most consequential. Hardware, electricity, maintenance, evaluation and live data are not free. An offline model may still depend on proprietary tooling, locked hardware or externally controlled updates.

Most importantly, possessing a model does not mean a society can productively diffuse it.

As Jeffrey Ding argues in [*Technology and the Rise of Great Powers*](https://press.princeton.edu/books/paperback/9780691260341/technology-and-the-rise-of-great-powers), technological leadership depends less on spectacular inventions than on the capacity to adapt a general-purpose technology throughout an economy. That requires ordinary engineers, capable institutions, complementary infrastructure and organizations willing to redesign how they work.

We therefore need to distinguish four things: access to AI, ownership of AI, the capacity to adapt AI, and the ability to translate AI into broadly shared prosperity. Each is harder than the one before it.

Iridium distributed theoretical reach. Meta distributed practical access while retaining centralized control. AI might distribute something more consequential: productive intelligence that communities can possess and compound for themselves.

Or it might become another globally indispensable utility whose benefits are widely shared while its ownership, profits and governing power remain concentrated elsewhere.

**The question is not merely whether AI will be available to everyone. It is whether AI will distribute the consumption of intelligence—or the capacity to own, shape and deploy it.**