The OpenAI partnership is simultaneously Microsoft's greatest cloud growth driver and most concentrated risk. OpenAI accounts for 45% of Microsoft's $625B commercial RPO and spent $12.43B on Azure inference between CY2024 and Q3 CY2025. Azure remains the exclusive provider for stateless OpenAI APIs -- even the $38B AWS deal routes API traffic through Azure. But Microsoft lost the right of first refusal as compute provider, and OpenAI has diversified across five or more cloud partners.
From exclusive partner to largest-but-not-only partner
OpenAI has signed infrastructure deals totaling $600B-$1.4T across Azure ($250B), Oracle ($300B), AWS ($38B), CoreWeave ($22.4B), Google Cloud, and custom Titan chip development. Microsoft is hedging via Anthropic ($30B Azure deal), MAI frontier models, and Maia silicon. The relationship has structurally shifted -- Azure's relative share of OpenAI compute is declining even as absolute consumption grows.
Can OpenAI fund its $600B+ infrastructure commitments? Expected $8B cash burn in 2025
OpenAI's compute exclusivity with Azure has eroded significantly since October 2025. Microsoft lost right of first refusal, and OpenAI signed infrastructure deals across five or more providers. The AWS deal alone provides hundreds of thousands of GB200/GB300 GPUs with AWS as exclusive third-party distributor for OpenAI's Frontier agent platform. The Stargate JV ($500B, 10GW) operates independently from Azure through a SoftBank/Oracle/OpenAI joint venture.
| Azure | $250B | ~6 years | Stateless API exclusivity preserved |
| Oracle (Stargate) | $300B | 5 years | ~400K GB200 GPUs, 4.5GW capacity |
| AWS | $38B | 7 years | Exclusive Frontier platform distribution |
| CoreWeave | $22.4B | -- | AI-native infrastructure |
| Custom Titan chip | TBD | H2 2026 | Broadcom/TSMC 3nm, Titan 2 on A16 |
Relative share declining, absolute volume growing
The key protection for Microsoft is that stateless API traffic remains Azure-exclusive regardless of where compute runs, and Microsoft receives 20% revenue share from all OpenAI partnerships. Azure's relative share of OpenAI compute is declining, but absolute consumption continues growing as OpenAI's total compute footprint expands. The Stargate JV's first UAE deployment (1GW) is operated by OpenAI/Oracle, not Azure.
Microsoft's financial relationship with OpenAI operates through four channels: Azure consumption revenue ($12.43B CY2024-Q3 CY2025), 20% revenue share from OpenAI's total revenue, 27% equity (~$135B), and Azure OpenAI Service revenue. OpenAI's rapid growth ($2B in 2023 to $20B+ annualized run rate by year-end 2025) drives growing payments to Microsoft but also creates substantial earnings volatility through equity method accounting.
Equity method creates EPS instability
Q1 FY2026 saw a $3.1B net income hit from OpenAI losses ($0.41 EPS), while Q2 FY2026 saw a $7.6B gain from recapitalization ($1.02 EPS) -- a $1.43/share swing across two quarters. Microsoft will exclude OpenAI from forward guidance. The mutual dependency is symmetric: OpenAI's pre-IPO risk factors flag Microsoft as a major business risk.
Microsoft receives 20% of OpenAI total revenue through 2032; $866M received through Q3 CY2025
OpenAI raised $110B at $730B pre-money in Feb 2026; IPO targeted 2026-2027
OpenAI expected to burn $8B cash in 2025 with ~40% gross margins; projected compute spending $17B (2026), $35B (2027)
OpenAI pre-IPO risk factors flag Microsoft as major business risk: partnership modification could adversely affect business