Revolut shifts 2,000 engineers to self-hosted GLM, cutting AI costs to a tenth of previous spend

·2026.11.03 07:11

Key point

The bank deployed 56 Nvidia B200s to run GLM internally, reducing token costs to 30 cents per million and improving engineering lead time by 70%.

Revolut moved approximately 2,000 engineers to self-hosted GLM for coding, restricting Claude Code and Codex to an exception list.
The infrastructure consists of 56 Nvidia B200s, achieving a blended cost of 30 cents per million tokens, roughly one-tenth of previous spending on Anthropic and OpenAI.
Engineering lead time improved by 70% as engineers stopped rationing AI usage for difficult tickets and applied it to all tasks.
The shift positions open-weight models as the default for high-volume coding, challenging the pricing power of frontier model providers.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.