Revolut cut AI coding costs 90% by ditching OpenAI for self-hosted GLM
Revolut started its AI coding rollout the same way most companies do: Gemini, Claude Code, and a growing bill. Engineers were spending up to $10,000 a month per person, and the ROI was not obvious enough to justify it.
According to David Abitbol, operations and infrastructure lead at Revolut, the company responded by capping token budgets. That backfired. Engineers started hoarding their allocations for complex tasks and stopped using AI in their day to day work.
The Switch to Self-Hosted GLM
Revolut tested GLM models from Chinese company Z.ai, starting with GLM-5.1 before moving to GLM-5.3. The results Abitbol reported at a New York conference:
- ~90% reduction in inference costs
- ~60% decrease in coding time
- ~70% decrease in lead time
Deploying GLM-5.3 across the organisation required 56 of Nvidia’s B200 GPUs. But Abitbol noted the biggest gains did not come from throwing more hardware at the problem. They came from tuning the model’s hyperparameters and software optimisations.
The Security Question
Abitbol acknowledged the obvious concern: using a Chinese-developed model at a financial institution. His answer was that the risk is primarily tied to accessing these models through APIs running in China. Self-hosting removes that vector. The weights run on Revolut’s own infrastructure.
The Broader Pattern
Revolut is not alone in rethinking its AI cost structure. Santander moved from large general-purpose models to smaller specialised ones to cut costs and improve output quality. Goldman Sachs went the other direction, restricting most traders and bankers from advanced AI coding tools entirely after a productivity and risk review.
The operator takeaway: the default stack of frontier API providers is not the only path, and for teams with the infrastructure to self-host, the cost delta can be significant.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
