OpenClaw creator's AI token bill hit $10M in one month

Peter Steinberger, the developer who created the open-source AI agent OpenClaw and now works at OpenAI, said he spent about $10 million on AI tokens last month. He made the disclosure on stage at Cloudflare's Connect conference in San Francisco on Wednesday, and Thomas Claburn of The Register reported the number.

The figure is a reality check for anyone building products on large language models. Tokens are the units AI companies use to charge for compute, roughly mapping to pieces of words, and a single agentic task can consume thousands of them as the model reads, plans, and executes steps. Steinberger's bill is not a typo and not a server cost; it is the raw price of feeding frontier AI models at a scale most teams will never approach.

OpenClaw is an open-source AI agent, a class of software that does more than answer questions. It can operate a computer, browse the web, and take actions on behalf of a user, which means every action is another round of token-consuming inference. The project has become a popular tool for developers experimenting with autonomous agents, and Steinberger's own usage generated a bill that would strain the budget of many mid-size companies.

For CFOs and CTOs, the disclosure is a warning about unit economics. AI pricing is typically metered per token, and costs mount at every step of an agent's reasoning loop. A task that involves a dozen model calls can cost more than a traditional software request by orders of magnitude, and when agents run unattended, the metering never sleeps. The difference between a single query and an autonomous workflow is the difference between a rounding error and a line item.

The number also lands at an awkward moment for the industry. Token prices have fallen sharply over the past year as providers compete, yet the most capable models still command premium rates. Meanwhile, agentic workflows are expanding rapidly, and the market is learning that the total bill depends less on the listed price per token and more on how many tokens a task actually burns. Steinberger's month suggests that the biggest constraint on AI adoption may not be model quality but budget.

Steinberger's situation adds a wrinkle. He is now employed by OpenAI, one of the largest sellers of token-based AI services, and he built OpenClaw as an open-source project. That means his $10 million month is both a personal data point and, indirectly, a sign of how much demand the biggest AI labs are seeing from sophisticated developers who are willing to push agents to their limits.

For executives budgeting AI initiatives, the lesson is to instrument agents before scaling them. The same technology that makes agents powerful, their ability to take many steps, is what makes their costs unpredictable. Steinberger's number, shared without embarrassment on a conference stage, is the kind of reference point that should reshape how boards think about AI spending: not as a fixed software license, but as a variable cost that can scale with ambition.

The strategic stakes are clear. Any company deploying AI agents needs to understand that costs will scale with autonomy, and Steinberger's $10 million month is the vivid proof. The era of free experiments is over, and the teams that treat token spend as a first-class metric, not an afterthought, will be the ones who survive the agentic shift.