OpenAI's viability called into question by reported spending with Microsoft (www.theregister.com)

Leaked Microsoft internal documents, reported by Ed Zitron and reviewed by the Financial Times, suggest OpenAI may have paid more than $12 billion to Microsoft for compute since 2024 and spent roughly $8.7 billion on inference on Azure in the first three quarters of 2025 alone — more than double its reported 2024 inference spend. The leaks also use Microsoft’s 20% revenue-share deal with OpenAI to infer sales: Microsoft received $454.7M from OpenAI in H1 2025 (implying ~$2.27B in OpenAI sales for that period) and $865.9M through Q3 (implying ~$4.33B year-to-date). Those figures clash with Sam Altman’s $13B annual revenue claim and earlier reports of a ~$9B cash burn, and they notably exclude training costs and other cloud suppliers (CoreWeave, Google), raising questions about completeness. For the AI/ML community the significance is twofold: technically, inference — running deployed models at scale — is extraordinarily expensive and may dominate ongoing operating costs, not just one-off training bills; commercially, opaque accounting (possible Azure credits, reseller revenue routing, and unbroken-out AI revenues across Microsoft products) makes it hard to judge whether current LLM deployment models are financially sustainable. If the leaked numbers are accurate, expect pressure for greater financial transparency, potential price increases for API/consumer access, slower infrastructure rollout, and renewed scrutiny of cloud‑LLM economics and partnership terms.