The only AI moats in LatAm are local data and regulation

Investors keep asking the moat question about AI applications. Build it today, what stops someone from building the same thing tomorrow? Two layers of the same question, really:

1. Inside our region (LatAm), will the market atomize or consolidate?
2. If it consolidates, will US winners eventually come and eat the regional ones?

On the first question: I think the market will consolidate around a few regional winners per category, the same way SaaS did from 2000 to 2020. Distribution dynamics are heavy. Data dynamics are even heavier. In a CRM-like or operational-software-like product, more data → better models → fewer hallucinations → better outcomes for the customer → more data. That flywheel exists for AI more than it did for traditional SaaS. Your job, if you’re building, is to be one of the few regional winners in your vertical.

On the second question — will US winners eat the regional players — that’s where regulation comes in.

Look at oil. Look at telecom. Both started as commodities nobody cared about regionally. Then they became infrastructure. Once an industry becomes infrastructure, every government wants control. Wenceslao Casares told a story about going to a 1990s conference and saying “if I disconnect the internet right now, nothing changes.” He was right at the time. Try saying that today.

AI is heading the same way. Today, if every model in the world goes offline, only some workflows break — developers, customer service, a few others. In ten or twenty years, every white-collar process and a lot of blue-collar ones will run on this infrastructure. At that point, every government in LatAm will ask: where are the chips? Where do these decisions actually run? Are they here, or are they in Virginia?

That question forces regulation. Local data centers. Local compute. Local models. And regulation buys regional companies time. The Brazilian fund Atlantico called this pattern “Latam does it best” — same wave, better local results because the regulatory and distribution friction lets local players consolidate before US players can roll them over.

Mistral is the European version of this thesis. Their models aren’t the best in the world today, but as model capability commoditizes, “good-enough + local + sovereign” wins for a meaningful share of customers.

The same logic says: if you’re building an AI application company in LatAm, eventually you’ll want your own model — fine-tuned for your specific use cases, run on local infrastructure. Today that’s a bad bet. In three to five years it’s an obvious one. Intercom is already there in customer service.

So the moats in this region are: your data, your distribution, and the regulation that’s coming whether you like it or not. Plan for them.

▮