last updated Nov 2025


🎯 The Bottom Line

To justify current AI infrastructure spend, inference token consumption needs to grow 9-12% monthly through 2026.

<aside>

Current trajectory: So far, so good 🟢

</aside>


📊 Actual Token Growth vs. Required Growth

Required Monthly Growth Rate

+9-12% CMGR (2025-2026)

Reported Token Growth

+13-17% CMGR range, in recent months.

Provider Period Token Consumption CMGR Source
Alphabet
(total) May 2025 480T Q2 earnings
Jul 2025 980T +43% Q2 earnings
Sep 2025 1,300T +15% Q3 earnings
Alphabet
(Gemini API only) Sept 2025 300T
(~23% total token volume) Q3 earnings
Microsoft
(Foundry APIs only) Full year - FY 2024
(Jul 2023-Jun 2024) 71T FY Q4 2025 earnings
Full year - FY 2025
(Jul 2024-Jun 2025) 500T +17% FY Q4 2025 earnings
Q3 - FY 2024
(Jan 2025-Mar 2025) 20T FY Q3 2025 earnings
Q3 - FY 2025
(Jan 2025-Mar 2025) 100T +14% FY Q3 2025 earnings
OpenAI
(API calls only) Oct 2023
(monthly run rate) 13T Source
Oct 2025
(monthly run rate) 259T +13% Source

Highlighted growth stats represent the latest figures. No public token consumption data for Meta, Amazon, or Anthropic.

Key Observations

🟢  All major providers reporting token metrics are currently exceeding +9% CMGR

🟡  Alphabet's monthly cumulative growth decelerated from 43% (May-Jul) to 15% (Jul-Sep). Alphabet's token increase partially driven by model inefficiency (Gemini 2.5 Flash uses 17x more tokens per request).


📈 The Model

This tracker is based on a simple framework that connects infrastructure spending to inference token demand.