The story everyone tells about cheap AI is that it makes large companies faster. The more useful story is that it has quietly removed the reason a company needs to be large at all.

Take the clearest case. Anysphere, the company behind the coding tool Cursor, went from one million dollars of annual recurring revenue to one hundred million in about twelve months, the fastest any software company has reached that line, and it did it with somewhere between forty and sixty people. By early 2026 the revenue was past a billion. Run the division and you get a figure with no real precedent: revenue per head in the millions, against the hundred and fifty thousand or so a normal software company manages. The headline writes itself as a story about growth. The interesting part is the denominator.

The headline writes itself as a story about growth. The interesting part is the denominator.

So here is the claim, stated plainly. For most of the last twenty years the binding constraint on starting a company in Southeast Asia was people and the money to pay them. You needed thirty engineers and a year of runway to find out whether the idea even worked. The tooling has not added talent to the region. It has made a small amount of talent reach further. The constraint has moved, from whether you can staff the thing to whether you have the taste to know what to build and a way to put it in front of customers.1

You can already see the shape of it locally. Level3AI, a Singapore company two people started in the middle of 2024, raised thirteen million dollars of seed money from Lightspeed last year to build enterprise customer-service agents; the cheque is larger than the entire company would have been a few years ago. Butterfly Effect moved the team behind the Manus agent out of Wuhan and Beijing and onto a single floor in Singapore, around forty core engineers, raised at a five-hundred-million-dollar valuation in April 2025, and was bought by Meta before the year was out at more than four times that. Neither is a thirty-person company. Both are the kind of outcome that used to need one.

The small team is a real advantage, and it is not a moat.

The honest part of the story is that the leverage is rented. The same subsidised models that let four people do the work of thirty are priced below what they cost to run, and the price will not stay there. A founder who builds a company whose margin depends on the model staying this cheap has made the operator’s mistake from the other chair. The small team is a real advantage, and it is not a moat. What the founder owns is the taste, the customers, and the workflow nobody else can see; the model underneath is a commodity that gets repriced the morning one of the four giants blinks.

Which leaves the regional question this magazine keeps circling. Singapore now routes most of the venture money in Southeast Asia, and the seed end has been thinning for two years; the ecosystem pays growth and starves origin. Cheap tools do not touch the money. They lower the cost of origin, the exact part the money stopped funding. The company worth watching next is the one with four names on the cap table and a product that already charges.

Footnotes

  1. The version that gets passed around is Sam Altman’s standing bet with other founders on the year someone builds a one-person billion-dollar company, and Dario Amodei putting it at 2026 with seventy to eighty percent confidence. The figure that should worry them is Medvi, a US health venture that booked four hundred million dollars of sales in its first full year on a headcount of two. The prediction is already quaint; the only open question is the postcode. ↩