The Advantage of Being Late
The most valuable thing about building in Southeast Asia right now is that you’re late.
This sounds like a weakness, so let me argue it properly, because the whole point turns on it being a strength.
When people talk about the AI race, they talk about it as a race to build the biggest model. There’s a leaderboard, and a handful of labs in California and China are at the top of it, and everyone else is somewhere below, presumed to be losing. If that’s the game, then yes, a startup in Jakarta or Manila or Kuala Lumpur has already lost, and so has almost everyone reading this.
But that’s not the game. It’s just the part of the game that’s easiest to measure, which is why it gets all the attention. The model is an input. What matters is what the input is used for, and that question is barely being asked yet.
Consider what actually happened with the last platform shift. The company that built the best web browser didn’t win the internet. The company that laid the most fiber didn’t win the internet. The winners were businesses that took the infrastructure as a given, as plumbing, and built something people wanted on top of it. Amazon didn’t invent the internet. It figured out what to sell over it, and how, before its competitors took the medium seriously.
AI is going to rhyme with that. The models are becoming a commodity faster than almost anyone expected. A capable model is now something you rent by the token, and the price of that token has been falling on a curve that should terrify anyone whose entire business is selling tokens. When the powerful thing becomes cheap and universal, the advantage moves. It moves to whoever knows exactly what to build with it and for whom.
This is where being late helps.
A company in a mature market that wants to use AI has to fight its own past first. It has thirty years of software, of processes, of people whose jobs are defined by the old way of doing things. Every improvement is a renovation, and renovations are slow and political and expensive. You can’t just install the future. You have to demolish the present to make room for it.
Now add the thing the leaderboard can’t see: the region is not one market but a dozen, with different languages, different regulations, different ways money moves and trust is earned. To a lab optimizing a global benchmark, that fragmentation is noise to be averaged away. To someone building on the ground, it’s a moat. The company that does the unglamorous work of making a model actually function for a Vietnamese small business, or a Filipino logistics operator, or an Indonesian lender, in their language, inside their rules, wired to how they already work, builds something no global model provider will bother to replicate. It’s too specific to be worth their time and too hard to be copied by the next founder who skipped the work.
I want to be careful here, because this is exactly the kind of claim that’s easy to make and easy to get wrong. Being late is only an advantage if you use it. Plenty of markets are late and stay late. The clean install means nothing if you spend it importing someone else’s playbook a beat behind. The opportunity isn’t to become a regional copy of an American company. It’s to build the thing that couldn’t have been built anywhere else, the application that only makes sense given this language, this regulation, this way people here already live.
So the question isn’t which country wins AI. Countries don’t win AI any more than countries won electricity. When electricity arrived, factories that just bolted a big motor where the old steam engine sat saw almost nothing change. The gains came decades later, once someone rethought the entire factory around what electricity made possible. The technology was never the point. What you built on top of it was.
The model will be cheap and everywhere sooner than you think. The advantage was never in owning the current. It’s in being the one who knew what to plug in.
