Charted: How Many Tokens Are Major AI Brands Processing? (2022–June 2026)

The metric nobody publishes uniformly

Every major AI lab reports revenue, MAU, or GPU counts — but almost none publish a clean "tokens sold" or "tokens processed" figure. Yet tokens are the atomic unit of AI economics: they drive inference cost, API pricing, and datacenter power draw.

The interactive chart above tracks 352 monthly records across 8 major brands from November 2022 (ChatGPT launch) through June 2026 — the most recent full month. Each row flags whether the figure is disclosed (partial public metric) or estimated (model-derived).

June 2026 snapshot

Aggregate estimated token volume across tracked brands reached roughly 3.8 quadrillion tokens per month (~3,787T) in June 2026 — up sharply from June 2025 as enterprise API and Copilot/Gemini deployments scaled.

Share shifted materially:

• OpenAI still leads but its share compressed as hyperscaler bundles scaled
• Google and Microsoft gained fastest on enterprise Copilot + Vertex/Gemini ramp
• Anthropic API growth accelerated on Claude 4 enterprise adoption
• xAI went from zero to meaningful share in under 18 months post-Colossus

Why tokens grew faster than revenue

Three forces explain the divergence:

1. Price per token fellcompetition and model efficiency gains (speculative decoding, distillation) pushed effective $/MTok down 60–80% from 2023 peaks
2. Free tiers expandedChatGPT, Gemini, Copilot, and Meta AI absorbed billions of consumer tokens without direct revenue
3. Enterprise bundlingtokens shipped inside Office 365, Workspace, and AWS contracts where revenue is recognized separately from inference volume

Methodology

Estimates blend:

• Disclosed partial metrics where available (e.g., Google I/O 2025 Gemini throughput peaks, OpenAI DevDay API scale comments, xAI Colossus capacity disclosures)
• Revenue ÷ blended $/MTok for API-first providers (Anthropic, Cohere)
• MAU × avg tokens/session for consumer products (ChatGPT, Meta AI)
• Copilot/Azure attach rates × tokens/user for Microsoft
• Bedrock/AWS AI revenue share × throughput proxy for Amazon

No row should be read as audited financial data. The dataset is designed for relative trend analysis and share-shift visualization, not precision accounting.

What to watch

• Enterprise vs consumer mixAPI tokens carry revenue; free chat tokens do not
• Model efficiencyfewer tokens per task compresses volume even as usage grows
• Regulatory disclosureEU AI Act and SEC climate rules may eventually force throughput reporting
• Custom siliconGoogle TPU and Amazon Trainium shift cost curves independently of token counts