Where AI Lands First Won't Be the Rich World

---

I want to flag something I keep noticing that the AI press isn't really writing about. The story everyone's telling about AI is a story about frontier capability. Bigger models, more agents, longer context windows, what GPT-6 will do that GPT-5 can't. That's a real story and I follow it like everyone else. But it's not the most interesting story.

The most interesting story is what happens when frontier capability becomes free, runs locally, and lands in markets that don't have the legacy process baggage that makes adoption slow in the rich world.

Here's the thing about absorptive capacity, the academic term for how fast an organization can take in new technology and put it to use. In rich-country businesses, absorptive capacity is the bottleneck I've been writing about. Operators have processes, hierarchies, vendor contracts, software stacks, trained employees, and a thousand small habits that all have to be touched before AI deployment can land. The absorption is slow because there's a lot to absorb against.

In a clinic in Nairobi running on paper and a single laptop, the absorption surface is different. There is no legacy CRM to integrate with. There are no compliance officers to convince. There is no SaaS stack to rationalize. There is, in many cases, no software at all. When a 31B-parameter model becomes free under Apache 2.0 and runs on a workstation, that clinic gets to skip three decades of enterprise software adoption and arrive directly at the current frontier. The same dynamic that has made mobile banking enormous in Kenya and Nigeria, while still niche in the United States, is going to repeat with AI.

This is technological arbitrage. The capability arrives at roughly the same moment everywhere, but the cost of using it is wildly different across markets. Where there's no incumbent process to displace, the deployment can move at the speed of the technology. Where there's an incumbent process, deployment moves at the speed of organizational change, which is much slower.

The humanitarian dimension is the part I think will be genuinely surprising over the next five years. A radiologist in a country with one specialist per 200,000 people benefits more from AI image triage than a radiologist in Boston. A small-claims court in a country where contracts are routinely unenforceable benefits more from AI document review than a courthouse with a functioning bar. A farmer who has never seen a county extension agent benefits more from AI advisory than a farmer with a USDA office down the road. The marginal value of frontier AI capability is highest exactly where the existing institutional capacity is weakest.

I want to be careful about how I'm framing this because I don't have firsthand experience deploying AI in emerging markets. I read the research, I watch the projects, I have opinions, but I'm an observer here, not a practitioner. So this is a hypothesis, not a report from the field. The hypothesis: the deployment opportunity in the developing world over the next decade is going to be the largest and least-noticed AI story of the cycle, and it's going to be powered by exactly the same shift that's making local models work for regulated industries in the United States. Capability got cheap. Compute got accessible. The legacy was never there to slow deployment down.

What I find interesting about this for businesses in the rich world is the implication. If absorption is the bottleneck, and emerging markets have less to absorb against, the global learning curve on what works with AI is going to be flatter than people expect. The case studies on what AI deployment looks like in five years won't all come from American Fortune 500s. Some of them will come from places that didn't have a Fortune 500 equivalent to compete with. Operators who pay attention to what's working in those markets will see the future earlier than operators who only read about American enterprise rollouts.

I don't have a sales hook for this post. It's an observation, not a service. I find this stuff genuinely fascinating and I think more operators should be watching it. The future of AI deployment is going to be more globally distributed than the press coverage implies, and the most interesting case studies of the next decade are not going to come from where the most interesting case studies of the last decade came from.