Altman Reveals OpenAI Token Consumption Spike

According to Axios, OpenAI CEO Sam Altman said during a livestream on enterprise adoption that "The token leader at OpenAI uses about 100 billion tokens a month," then added, "To my embarrassment, that's not the token leader in the world. We found someone that used even more." Axios reports Altman contrasted that figure with OpenAI's top user roughly six and a half years ago, who consumed about 100,000 tokens a month, implying a near 1,000,000-fold increase in per-user token consumption.

Business Insider has documented OpenAI's internal "token leaderboard," where employees track and sometimes publicize how many tokens they burn. The most-cited example is OpenClaw creator Peter Steinberger, whose roughly 100 Codex agents ran up about 603 billion tokens across 7.6 million requests in 30 days, a bill near $1.3 million that OpenAI, his employer, covers, as reported by The Next Web and Tom's Hardware. Business Insider, citing The New York Times, separately reported one employee using about 210 billion tokens in a single week.

Altman framed cost as becoming "a huge issue" for customers, according to Tom's Hardware, which quotes him saying it is "kind of a meme now" for companies to report that they "spent my entire 2026 budget in Q1." Axios reports Altman described cost as among the most common issues customers now raise, behind simplifying workflows, and pointed to a growing infrastructure challenge as always-on, proactive AI agents push usage and spending higher.

Rapid growth in tokens per user magnifies two engineering pressures, sustained inference compute and state storage for long-lived agent sessions, pushing operators toward quantization, caching, distilled models, batching, and tiered serving to control unit costs. The cost theme is not unique to OpenAI: Tom's Hardware and other outlets reported that an unnamed company accidentally spent about $500 million on Anthropic's Claude in a single month after failing to cap employee usage, a case widely cited as a cautionary tale about how agentic workloads can escalate spending.

Key indicators over the coming months include vendor pricing or volume-discount changes for high-token accounts, product-level rate limiting and server-side token caps, the spread of token-cost-optimization tooling such as profilers, budgets, and proxy caches, and whether competing clouds publish per-customer spend data. For practitioners, these are the operational levers most likely to curb surprise bills and improve ROI as per-user consumption scales into the tens or hundreds of billions of tokens per month.