Microsoft and Anthropic play invoice tennis with startup's $17,600 Claude bill

If you use cloud credits for third-party AI models, a hidden billing distinction could lead to unexpected charges.

Norwegian startup Vegalabs AS incurred a $17,600 bill for using Anthropic's Claude model through Microsoft Azure Marketplace. Vegalabs had received $25,060 in Azure credits through Microsoft for Startups, expecting these to cover their AI workloads. However, Microsoft's sponsorship rules exclude third-party models like Anthropic's when purchased via Azure Marketplace. Vegalabs claims the deployment interface failed to clearly distinguish between credit-covered services and separately billed Marketplace charges, leading to the unexpected bill. Microsoft attempted to charge Vegalabs' card, but the issuer declined it for suspected fraud. The startup deleted the deployment within an hour of discovery and had $21,168 in unused Azure credits expire. Vegalabs reports that both Microsoft and Anthropic support directed them to the other for resolution, with support communications, including from Anthropic, appearing to be AI-generated and misidentifying their account type.

This incident highlights a critical disconnect in how cloud providers bill for third-party AI services, particularly when bundled with promotional credits. Vegalabs' experience suggests that the distinction between Azure-sponsored credits and Azure Marketplace charges for models like Claude needs far greater clarity at the point of deployment. The startup's assumption that all AI usage would be covered by provided credits, despite monitoring usage, points to a user interface and billing transparency issue that could affect many startups relying on similar promotional offers to experiment with advanced AI.

The back-and-forth between Microsoft and Anthropic support, with each deferring responsibility, also underscores a potential gap in vendor accountability for integrated marketplace offerings. This situation creates significant friction for customers encountering billing discrepancies and raises questions about the effectiveness of support channels when complex, cross-vendor charges arise. As more startups leverage these powerful AI models, clearer communication from cloud providers about credit exclusions and more streamlined support processes for marketplace issues will be essential to avoid similar costly misunderstandings.