Thailand has roughly 14,000 7-Elevens. One for every 5,000 people. In Bangkok, you can stand in one and see another across the street. This is not a failure of planning. It is the plan.
The Western instinct is to build fewer, better locations. Bigger stores, wider aisles, more selection. Optimize the unit. The Thai 7-Eleven model inverts this completely: optimize for proximity. Each store is small, the selection is curated, and the experience is "good enough." But there is always one within a two-minute walk. Always.
Good Enough, Everywhere
This is a profoundly different infrastructure philosophy. Instead of asking "how do we make each node excellent?" you ask "how do we make sure no one is ever far from a node?"
The tradeoff is real. No single Thai 7-Eleven is impressive. The inventory is tight. The floor space is cramped. A Buc-ee's would eat it alive on any individual metric. But Buc-ee's requires a car, a highway, and a destination mindset. The 7-Eleven just needs you to be thirsty.
Density beats excellence when the problem is access, not quality.
The Infrastructure Parallel
I think about this pattern constantly because it maps directly onto compute infrastructure. The hyperscalers built cathedrals - massive facilities in cheap-power regions, connected by fat pipes, serving the world from a handful of locations. Brilliant engineering. Absolute unit economics.
But latency is physics. A packet from Bangkok to us-east-1 does not care how impressive your facility is. It cares about distance. And as workloads get more latency-sensitive - real-time inference, edge rendering, interactive anything - the cathedral model starts to crack.
The answer is not to build better cathedrals. It is to build more churches. Smaller, closer, "good enough" compute nodes distributed where the users actually are. Not as impressive individually. Devastating in aggregate.
The Catch
Density plays are operationally brutal. One cathedral is easier to staff, secure, and maintain than fifty churches. The management overhead scales with node count, not capacity. This is why most providers do not do it - not because they have not thought of it, but because the org chart rebels against it.
The companies that crack distributed operations at scale without distributed costs will own the next decade of infrastructure. Everyone else will keep building bigger barns in Virginia and wondering why APAC customers complain about latency.
The best infrastructure is not the most powerful. It is the closest.
Fourteen thousand 7-Elevens figured this out selling rice balls and iced coffee. The compute industry is still catching up.