The OpenClaw Burn: A Case Study in Venture-Backed Token Bloat

In the creator economy, we often talk about 'scaling' as a virtue, but the current trajectory of AI agent development suggests a dangerous pivot toward token-circularity—where value is inflated by spending venture capital on compute that doesn't necessarily yield a proportional return in utility.

**Opinion:** The OpenClaw burn rate is a textbook example of the 'token-circularity' trap. When creators inflate ecosystem value using house money that will never convert to sustainable revenue, they aren't building a business; they are running a subsidized experiment. This creates a distorted reality where the cost of innovation is decoupled from the value of the output.

As first reported by *The Register*, this disconnect became evident during a conversation at the Cloudflare Connect 2026 conference in San Francisco. Peter Steinberger, who created OpenClaw and serves as a member of technical staff at OpenAI, estimated that his AI workloads cost roughly $10 million in tokens last month.

While the figure is staggering, *The Register* notes that the costs are borne by investors. This financial cushion was underscored by Cloudflare CTO Dane Knecht, who observed that Steinberger operates with a "virtually unlimited token budget," a luxury not shared by most developers. Even more telling was Steinberger's own admission to the audience: "What people don't always understand is not all of that is useful."

This admission reveals the inefficiency inherent in the current AI agent gold rush. While Steinberger argues that the primary hurdle is that infrastructure has not yet caught up—questioning why workloads are still running on local computers rather than in the cloud—the reality is that cloud-hosted agents are prohibitively expensive for anyone not "setting venture funds alight," according to *The Register*.

Beyond the burn rate, the conversation between Knecht and Steinberger touched on the systemic friction AI is introducing to the creator ecosystem. Knecht pointed out that many open source maintainers are abandoning pull requests due to frustration with "AI slop." Steinberger suggested that the process for handling pull requests must evolve, arguing that trust increases when contributors explain the reasoning behind their code changes rather than simply delivering a finished product.

As OpenAI reportedly seeks to "accelerate funding" to compete with the expected IPO of rival Anthropic, the OpenClaw model serves as a warning. When the cost of the 'intelligence' is subsidized by venture capital, the incentive to optimize for efficiency vanishes, leaving a fragile ecosystem that may struggle to survive once the house money runs dry.