OpenAI's Sam Altman Forecasts Exponential Surge in AI Token Usage, Casting Intelligence as a Metered Utility
Key Takeaways
- •OpenAI's most intensive internal user processes approximately 100 billion AI tokens per month as of mid-2026, representing a million-fold increase from the late 2019 peak of roughly 100,000 tokens.
- •A single external user consumed 603 billion tokens in one month, generating a $1.3 million bill that exceeded the usage of OpenAI's own top internal consumer.
- •Major technology providers including Microsoft, Google, and Amazon have transitioned from flat-fee software licensing to consumption-based pricing models where customers are charged per token processed.
- •Altman projects individual users could average 500 billion tokens monthly by approximately 2032 to 2033, with some leading-edge users potentially consuming quadrillions of tokens on a monthly basis.
- •Altman identified 'constant running proactive AI'—systems that operate continuously in the background without human-initiated queries—as a primary catalyst for future growth in token consumption.
In June 2026, OpenAI CEO Sam Altman highlighted a striking metric that underscores the rapid scaling of artificial intelligence usage. The company's most intensive internal user now processes approximately 100 billion AI tokens each month. By comparison, peak monthly token consumption in late 2019 stood at roughly 100,000 tokens—a one-million-fold increase over a span of about six and a half years.
AI Tokens as the New Metered Utility
AI tokens serve as the fundamental units of text and data that large language models handle when generating responses, composing code, or analyzing documents. Altman and other industry leaders are increasingly describing AI-driven intelligence as a metered utility, drawing comparisons to electricity rather than traditional software. That framing reflects a broader shift across the AI industry, where major providers—including Microsoft, Google, and Amazon—have moved away from flat-fee software licensing toward consumption-based pricing models in which customers pay per token processed.
The 100,000-token ceiling that represented the outer limit of AI usage in late 2019 now corresponds to the global per capita average. Looking forward, Altman projects that individual users could average 500 billion tokens per month by approximately 2032 to 2033. At the leading edge, some users may consume quadrillions of tokens on a monthly basis.
A $1.3 Million Monthly Bill
The scale of external consumption is already surpassing internal use. One external user reportedly consumed 603 billion tokens in a single month, generating a bill of $1.3 million. That figure surpasses the 100 billion tokens processed by OpenAI's own top internal user, indicating that the most aggressive usage is occurring beyond the company itself.
Throughout 2026, enterprises across multiple industries have been racing to deploy consumption controls, usage caps, and metering strategies to prevent unchecked AI-related spending. The surge in token volume has also intensified scrutiny of the physical infrastructure—data center capacity, semiconductor supply, and energy consumption—that underpins AI delivery at scale.
"Constant Running Proactive AI" as a Growth Driver
Altman identified a concept he termed "constant running proactive AI" as a major catalyst for future growth. Under this model, AI systems would not merely react to user prompts but would operate continuously in the background—monitoring systems, producing insights, and executing actions without requiring a human to initiate a query. The concept aligns with the emerging category of agentic AI, in which autonomous systems are designed to perform multi-step tasks with minimal human initiation—a focus area that multiple AI labs and enterprise vendors have publicly prioritized.
