Exclusive: Here’s How Much OpenAI Spends On Inference and Its Revenue Share With Microsoft

Source: Where’s Your Ed At Date: 2025-11-12 URL: https://www.wheresyoured.at/oai_docs/

Summary

Based on internal documents reviewed by Ed Zitron, OpenAI’s Azure inference spend was nearly triple previously reported figures: $3.77B in CY2024, $5.02B in H1 2025, reaching $12.43B combined through Q3 2025. Microsoft’s 20% revenue share payments implied OpenAI’s actual H1 2025 revenue was roughly $2.27B — against a $4.3B figure previously reported publicly, a ~$2B gap. The piece raises pointed questions about cost structure sustainability and whether public revenue figures reflect the same definition of revenue as the internal documents.

Implications

• The inference cost trajectory ($3.77B → $8.67B annualised run rate by Q3 2025) contextualises why Anthropic’s parallel compute diversification strategy (Akamai, multiple cloud sources) is competitively significant — Azure concentration at that cost level is a structural risk.
• The revenue-definition discrepancy feeds the broader economics bifurcation thread: if OpenAI’s unit economics are materially worse than reported, pressure on pricing and product pivots intensifies.
• For teams evaluating OpenAI vs. alternatives, the inference cost signal matters indirectly: high infrastructure costs at the provider level tend to flow into pricing floors and model deprecation cycles sooner than lower-cost operators.