OpenClaw Creator Peter Steinberger Spent $10 Million on AI Tokens in One Month

OpenClaw Creator Reveals Massive Token Spending at Cloudflare Connect

Peter Steinberger, creator of the open-source AI agent OpenClaw and current OpenAI employee, disclosed he spent approximately $10 million on AI tokens last month during a session at the Cloudflare Connect conference in San Francisco2

2

1

1

2

2

. The revelation came during a conversation with Cloudflare CTO Dane Knecht, who noted that Steinberger operates with "a virtually unlimited token budget, which not all of us in the room have." The audience reacted with visible surprise to the staggering figure, highlighting the growing concern among executives about AI token economics and the real costs of AI adoption2

2

1

1

2

2

.

The Reality Behind Unlimited AI Token Budgets

Steinberger was quick to temper expectations about his massive token spending, acknowledging that "what people don't always understand is not all of that is useful"1

1

2

2

2

2

. This admission underscores a critical challenge in AI agent development: distinguishing between necessary and wasteful token consumption. The contrast with other companies is stark—Accenture caps its staff's Claude spending at just $29 per month2

2

, illustrating the vast divide in corporate AI strategies. Steinberger's position at OpenAI clearly affords him advantages most developers lack. He previously told an agentic engineering meetup in San Francisco in June that he had solved his token constraints by joining OpenAI, stating, "Last year I was constrained by token. Now I fixed that"2

2

.

Infrastructure Challenges and Cloud-Based Solutions

Steinberger identified infrastructure as the main obstacle facing AI digital assistant adoption, questioning why AI agents still rely on local computing power. "Why is this still using my computer? Shouldn't this all be in the cloud?" he asked1

1

2

2

2

2

. This push for cloud-based solutions raises important questions about privacy and cost management that many CFOs are already grappling with. The timing is notable given Nvidia's recent launch of RTX Spark laptops designed specifically to run AI agents locally2

2

, suggesting competing visions for how AI agents should operate. Steinberger's comments about adding cloud costs to the equation did little to reassure those concerned about the financial implications of widespread AI adoption1

1

.

Agent-to-Agent Permission Systems

When asked about data permissions for AI agents, Steinberger proposed an innovative approach: using one agent to manage permissions for another. "You want to have another agent that can give it permissions or not and then this agent has to negotiate with another agent," he explained2

2

. He noted that agents often back down or find simpler routes within established guardrails, and this dual-system approach makes him more comfortable letting agents work autonomously. This vision of layered AI agent development addresses security concerns while enabling greater automation, though it also suggests additional token spending as multiple agents interact.

The Future of Open-Source AI and Token Economics

Steinberger joined OpenAI in February and committed to keeping OpenClaw open-source under a foundation2

2

. At Y Combinator's Startup School in August, he identified cost as the primary barrier to always-on AI agents, noting that few people are willing to spend that many tokens. "That's not so much a tech problem. It's more a token problem," he stated2

2

. Looking ahead, Steinberger doubts everyone will run hundreds of agents. Instead, he expects each person to have one intelligence that assists throughout the day and may even predict their needs. His massive token spending may be music to OpenAI's ears as the company prepares for its upcoming IPO and reportedly races to close a $30 billion funding round before Anthropic's IPO1

1

2

2

2

2

.