The Compute Divide: AI's New Geography and the Global South

A new fault line is forming in the global economy, and it runs along access to computing power. The chips, data centres, and energy needed to train and run frontier AI are concentrated in a handful of states and firms. Everyone else is becoming, in effect, a renter — dependent on infrastructure they do not own and standards they do not set. For the Global South, including Pakistan, this “compute divide” may prove as consequential as the digital divide was a generation ago.

Why compute is the new oil — and isn’t. Compute resembles oil in that it is scarce, strategic, and unevenly distributed. But unlike oil, you cannot simply extract it from the ground you happen to sit on. It requires advanced semiconductors (controlled by a few firms and gated by export rules), enormous capital, reliable power, and cooling — a stack that most developing economies cannot assemble. Pakistan, with its chronic power deficit and tight fiscal space, faces every one of these constraints at once.

The dependency trap. The likely path for countries like Pakistan is to consume AI as a service — renting access from US or Chinese cloud providers. That delivers benefits, but it also deepens the very dependency this site has flagged in the chip war: the underlying capability, data governance, and pricing power sit elsewhere. A national economy increasingly run on rented intelligence is exposed to decisions made in boardrooms and capitals far away.

What realistic ambition looks like. Pakistan will not build a frontier AI lab. But there is a viable middle path: investing in the layers it can own — talent, data, applied AI in agriculture, health, and public services, and the energy and connectivity that make any of it usable. The countries that benefit from the AI era will not all be the ones that build the models; some will be the ones that deploy them well.

The takeaway. The compute divide is a structural feature of the next decade, not a passing trend. For Pakistan, the strategic question is not whether to compete at the frontier — it can’t — but how to avoid being permanently relegated to the consuming end of someone else’s revolution.