Revolut Cut AI Costs 90% and Started Shipping Faster (linas.substack.com)

Revolut has dramatically reduced its AI coding expenses by 90% while significantly accelerating its engineering output. At a recent AI Engineer conference in New York, David Abitbol, head of AI operations at Revolut, revealed that the company transitioned approximately 2,000 engineers from pay-per-use AI APIs from Anthropic and OpenAI to Z.ai’s open-weight GLM model, hosted on 56 Nvidia B200 GPUs through Together AI. This strategic shift lowered the cost to $0.30 per million tokens and decreased the lead time for feature implementation from planning to merge-ready code by 70%. This transformation is notable for the AI/ML community as it exemplifies an effective model of cost management and productivity enhancement in AI operations. Revolut’s approach reflects an increasing trend in the fintech sector toward utilizing self-hosted models to achieve better performance and lower costs. With its substantial revenue growth and a focus on AI through the development of its foundation model, PRAGMA, Revolut serves as a case study for other companies in the industry to reevaluate their AI purchasing strategies and operational efficiencies. This move not only influences its competition and AI suppliers but also provides valuable insights for investors and industry leaders on how to leverage AI effectively.