Why we moved our entire stack to São Paulo

Leaving the Virginia region wasn't about speed. It was about sovereignty.

In February 2025, our engineering team sat in a windowless conference room in Buenos Aires and made a decision our investors called reckless. We would migrate every production service — payments, identity, the real-time order book — from a leased Virginia region to a colocated cluster in São Paulo. We budgeted six months. We did it in four.

The latency problem was never about milliseconds

Everyone assumes the argument for local infrastructure is speed. It isn't. Our p99 latency on the Virginia cluster was 87 milliseconds, and that was fine for a trading platform not chasing HFT margins. The real problem was something no North American cloud provider could solve: Brazilian banking regulation requires transaction logs to reside on servers physically within the country.

That compliance gap cost eighteen thousand dollars a month for two years. The São Paulo cluster eliminated it entirely. Clearing dropped from ninety seconds to four. That isn't a latency win. It's a different company.