How Revolut Cut AI Coding Costs 90% With Self-Hosted Open Models - David Abitbol, Revolut
Revolut cut AI coding costs 90% by self-hosting open-source models on dedicated GPUs
“We've seen engineers spend a thousand a month, ten thousand a month, without much return.”
Revolut's AI operations team self-hosted open-source models (GLM for coding) on dedicated GPUs starting April 2026, cutting AI coding costs by 90% compared to commercial APIs like Claude and Gemini. The shift also delivered 60% improvement in coding time and 70% improvement in feature execution time across 2,000 engineers. The core motivation was enabling unlimited token usage without per-engineer budget caps that were constraining adoption.