Revolut cut AI coding inference costs ~90% by self-hosting GLM on 56 B200s instead of Claude
sebkrier · x · 2026-11-05
At AI Engineer NYC, Revolut's ops/infra lead David Abitbol said the fintech slashed AI coding inference costs by 90% in under a year by making self-hosted open models the default, replacing reliance on commercial services like Claude.
Key facts:
- Rolling AI coding tools to 2,000 engineers yielded a 60% improvement in coding time and 70% in lead time.
- Path: tried Gemini in early 2025, Claude Code in early 2026, then moved to GLM (5.1, later 5.3), now served on 56 Nvidia B200 GPUs.
- Motivation was cost skepticism: engineers were spending $1,000–$10,000/month with unclear ROI, so usage had strict budget caps, prompting the self-hosting investigation.
Takeaway: cheap open-model tokens let engineers use AI liberally (debugging, Jira tickets), improving overall economics.
Related event: Revolut cuts AI coding costs 90% by ditching Claude for self-hosted GLM(11 posts)→
More from coding & agent
- Turning concept art into playable 3D levels: a 7-step agent workflow with GPT Image — FinanceYF5 · 2026-11-05
- Building a Complete Original Game in 24 Hours with Opus, Flux, Tripo and Three.js — FinanceYF5 · 2026-11-05
- Team automates 100% of product design with Claude, hasn't touched Figma in months — manosaie · 2026-11-05
- AgentVQA From Princeton Offers Offline Agentic Visual Benchmark to Cut Costly Online Evals — Key_Recording9766 · 2026-11-05
- Non-game-dev ships AI-built game so good tinyfool says his own needs work — tinyfool · 2026-11-05
- BallotBench pits 9 AI agents against election markets with real money on Kalshi — karinanguyen · 2026-11-05