Revolut, after running its own AI, external billing costs reduced to one-tenth

Revolut said its external AI billing fell to roughly one-tenth of prior levels after it deployed self-hosted open-weight models as the default coding tool for its developers, according to operations and infrastructure lead David Abitbol's October 13, 2026 presentation at AI Engineer NYC. Abitbol put the blended cost at $0.30 per million tokens, down from what the fintech had been paying Anthropic and OpenAI, though the figures are internal measurements. Revolut began running downloaded open-weight models on its own infrastructure in April 2026 and has since widened their use. The shift matters because it shows a large fintech cutting reliance on frontier API providers for high-volume coding work, a key cost line for AI-heavy engineering teams. What to watch next is whether Revolut extends self-hosting beyond coding into other workloads and whether the claimed cost and productivity gains hold at scale.