Revolut, the digital banking and financial services company, has moved to a self-hosted deployment of GLM, according to a social media post by guilleflorvs that was widely shared. The item reached 129 points and 95 comments on Hacker News, where the discussion focused on why a major fintech would choose to run a large language model on its own infrastructure rather than rely on an external API provider.
Self-hosting a model gives an organization direct control over where data is processed and stored, which matters for a regulated financial business handling sensitive customer information. It can also change the cost structure of AI usage, trading per token API fees for infrastructure and engineering investment. The post did not disclose the scale of the deployment or which specific GLM variant is in use.
Commenters debated the tradeoffs. Supporters argued that keeping model inference inside the company perimeter simplifies compliance and reduces dependence on third party vendors. Skeptics raised questions about operational burden, the pace at which self-hosted models fall behind frontier systems, and the difficulty of matching the reliability of managed services.
The move fits a broader pattern of large enterprises experimenting with running open and semi-open models on their own hardware. For Revolut, the decision signals that AI infrastructure is being treated as a core operational concern rather than a purely external service. The company has not issued a detailed public statement on the deployment.