      Tomasz Tunguz (Theory Ventures) · 2025-05-01
      · 399d
    

Microsoft's 100 Trillion Token Milestone: AI Infrastructure Economics and Azure Growth

Microsoft processed over 100 trillion tokens in Q1 2025, representing a 5x year-over-year increase, which could represent $327.6M to $4.6B in annualized AI revenue depending on model mix assumptions. The analysis reveals AI comprises 2-22% of Azure revenue, with non-AI database services (PostgreSQL, Cosmos DB) experiencing accelerated growth due to supporting infrastructure demands. Cost efficiencies from inference optimization and Jevon's Paradox demonstrate that improved performance drives broader ecosystem adoption.

9 metrics· Cited 0× in the knowledge base ·Open source ↗

Metrics in this report

AI Revenue as % of Azure

2-22%

range (low to high scenario)

Azure total revenue $21B annualized

Cost Per Token Reduction

50%

minimum

Microsoft inference operations

GPU Performance Improvement

30%

median

Microsoft Azure GPU fleet, ISO power

High Scenario Annualized AI Revenue

4,594.8million USD

mean

70% OpenAI, 20% Claude, 10% other

Input-to-Output Token Ratio

20:1ratio

assumption

Revenue modeling calculation

Monthly Token Processing Peak

50trillion tokens

maximum

Microsoft Azure, April 2025

Supply Chain Lead Time Reduction

20%

median

Dock to lead time for new GPUs

Token Processing Volume

100trillion tokens

quarterly

Microsoft Azure, Q1 2025

Year-over-Year Token Growth

5xmultiplier

median

Microsoft Azure