OpenClaw Founder's $1.3M AI Bill Reveals Trend of Subsidized 'Tokenmaxxing'

A staggering $1.3 million monthly bill for OpenAI services, footed entirely by the AI lab itself, has revealed the extreme, cash-burning methods some developers are using to build software. The developer, OpenClaw creator Peter Steinberger, is exploring a future where token costs are irrelevant, but his spending highlights a growing disconnect between AI consumption and proven value creation that questions the sector's economic foundation.

The seven-figure spending drew sharp reactions online. “Bro, you better show something that $1MM worth of engineers’ couldn’t do or this might be the beginning of the advertising for the frontier lab bubble bursting,” one user wrote on X, formerly known as Twitter. The user noted the pricing is heavily subsidized, meaning the actual compute cost would be substantially higher.

Steinberger’s own tracking app, CodexBar, showed his project burned through 603 billion tokens in 7.6 million requests over 30 days, primarily using the gpt-5.5-2026-04-23 model. The total bill came to $1,305,088.81. In his defense, Steinberger noted, “I can disable fast mode and it’s 70% cheaper. So it’s more like one employee.”

The episode brings the AI industry’s uncomfortable "bubble" question into sharp focus. The technology works, but the current economics are propped up by AI labs subsidizing massive user consumption to gain market share. Spending $1.3 million on tokens is only rational if the output generates at least that much in revenue or cost savings, a metric that remains elusive for many projects and puts the sustainability of the current model in doubt.

The $1.3 Million Automation Engine

In response to the debate, Steinberger detailed what the massive token consumption achieves. His team of three people runs approximately 100 AI agents that continuously work on the OpenClaw open-source project. The agents review pull requests, find security vulnerabilities, deduplicate issues, and write their own code fixes. Some agents are designed to open new pull requests based on the project’s stated vision, while others monitor performance benchmarks and report regressions in a Discord channel. The system, which also uses tools like Vercel's Deepsec and Codex Security, allows a tiny team to manage a large-scale software project with a high degree of automation. Steinberger stated his goal is to answer the question: “If Token[s] no longer matter, how will we build software in the future?”

'Tokenmaxxing' Becomes a Performance Metric

Steinberger’s public spending is the most visible example of a growing trend in Silicon Valley known as “tokenmaxxing,” where developers and engineers maximize their consumption of AI tokens as a key performance indicator. The practice has been encouraged internally at companies like Meta and Amazon, which have reportedly used leaderboards to track employee AI usage. The trend has even spawned its own hardware, like the "Clawdmeter," a small open-source desktop device that provides a real-time display of a user's token consumption for Anthropic's Claude model. This gamification of AI usage underscores a cultural shift where token throughput is becoming a new form of productivity measurement.

The economics of this trend, however, remain a central concern for investors. While proponents point to massive productivity gains, such as Citadel CEO Ken Griffin’s claim that AI completed months of PhD-level work in days, the direct return on investment for most tokenmaxxing is unclear. The practice is fueled by the AI labs’ strategy of subsidizing costs to accelerate adoption. This raises questions for companies like OpenAI and Anthropic about the long-term path to profitability and whether the current high levels of consumption can be maintained if and when prices rise to reflect their true cost.

This article is for informational purposes only and does not constitute investment advice.