Revolut launches AI research unit for banking models

Yesterday, Revolut announced they're building their own foundation model—PRAGMA—with NVIDIA, instead of licensing someone else's AI.

For 80 million customers across 40+ markets, that's a statement about where banking infrastructure is heading.

But here's what should keep other financial institutions awake: they're not just deploying AI. They're treating it like core banking rails. Risk assessment, platform operations, product recommendations—all running on a model trained on their own transaction patterns, their own fraud signals, their own customer behavior.

While legacy banks are still negotiating enterprise licenses with OpenAI or Anthropic, Revolut just made those relationships optional. That's not an R&D flex. That's a margin play and a compliance moat.

The gap widens when you realize what this means for speed. Internal tests already show performance gains. No API latency. No rate limits. No shared compute with a hundred other enterprise customers. And critically—no customer data leaving their infrastructure.

Three things become clear:

→ AI infrastructure is becoming a competitive wedge, not a feature checkbox

→ Fintechs with data density can train models legacy players can't replicate

→ Build vs. buy just became build vs. irrelevance for regulated industries

If you're running a bank, a telco, or any platform business with proprietary transaction data—this is your horizon. The question is whether you're funding a research unit or a vendor relationship.

Follow for more on AI-native business models in regulated markets.

#AIinBanking #Fintech #EnterpriseAI #DigitalTransformation #EmergingMarkets

📩 The full weekly breakdown lands in AI — Daily Brief → https://amplyfy.app/wire/subscribe/1