'We were highly skeptical about the cost of AI': Revolut reveals it cut AI coding costs by cutting out Claude entirely

Revolut has decided to replace Claude with self-hosted Z.ai models, resulting in a significant reduction in AI costs. The move has led to a 90% decrease in inference costs and a 70% improvement in lead time productivity, as well as a 60% improvement in coding time productivity. By adopting Z.ai's GLM models, Revolut can now operate its own inference infrastructure, although this comes with increased costs for hardware, maintenance, and electricity.

The company's Infrastructure Lead, David Abitbol, expressed that AI was initially met with skepticism due to high costs, but the self-hosted models have proven to be more cost-effective and offer greater control over data privacy and auditability. Despite the upfront investment in hardware, Revolut believes the economics are now favorable, with flat billing and no surprises at the end of the month.