The Three Compute Workload Layers on Azure

Microsoft’s $120B+ CapEx infrastructure serves three distinct workload types. Understanding this allocation reveals strategic priorities and the tension between Azure growth metrics and overall value creation.

Layer 1: OpenAI Workloads

• Fairwater Clusters: 300MW+ GPU buildings
• 700M+ weekly ChatGPT users
• Frontier model training (GPT-5, o3)
• $250B Azure commitment
• Dedicated Wisconsin + Georgia facilities

⚠️ SHRINKING % as OpenAI diversifies to AWS/Oracle

Layer 2: Microsoft Internal

• M365 Copilot: 15M paid seats
• GitHub Copilot: 4.7M subscriptions (+75%)
• Security Copilot expanding
• Highest-margin AI consumption
• $30/user/month premium pricing
• Dragon + Bing Copilots expanding

↑ DELIBERATE GPU prioritization over Azure

Layer 3: Third-Party Enterprise

• Anthropic: $30B commitment
• Nebius: $17.4B through 2031
• 1,500+ multi-model customers
• Only cloud with GPT + Claude
• 80% of F500 on Azure Foundry

★ THE GROWTH FRONTIER

CFO Amy Hood’s Insight

“If I had taken the GPUs that came online in Q1 and Q2 and allocated them all to Azure, the KPI would have been over 40%.” Actual reported: 39%.

STRATEGIC INSIGHT: Microsoft deliberately prioritizes first-party Copilot products over raw Azure compute metrics.

This is part of a comprehensive analysis. Read the full analysis: Microsoft’s Frontier AI Dilemma on The Business Engineer.