OpenClaw creator spends a massive $10 million on AI tokens in a single month

OpenClaw's creator used a virtually unlimited budget to run AI workloads, giving us a rare glimpse into the sky-high costs of modern AI.

In Short

• OpenClaw creator Peter Steinberger spent a massive $10 million on AI tokens in just one month.
• The massive bill was entirely funded by investors through his position at OpenAI.
• This reveals the staggering and often hidden computing costs required to run advanced AI models.

AI is incredibly expensive, and nobody knows this better than Peter Steinberger. The creator of the popular open-source AI agent OpenClaw, who is now a member of the technical staff at OpenAI, recently revealed a jaw-dropping figure: he spent an estimated $10 million on AI tokens in last month.

He shared this massive number during a conversation with Cloudflare CTO Dane Knecht at the Cloudflare Connect 2026 conference in San Francisco on Wednesday.

For context, tokens are the basic units of data that AI models process. Processing billions of these tokens requires immense computing power, which is why they cost so much.

Who paid the $10 million bill?

Fortunately for Steinberger, this massive cost is being covered by investors. As Knecht pointed out during the event, Steinberger operates with a "virtually unlimited token budget", a luxury most developers do not have.

When the audience was shocked at the $10 million figure, Steinberger candidly replied, "What people don't always understand is not all of that is useful". This honest admission added to the unease, especially since OpenAI is reportedly trying to accelerate its funding ahead of rival Anthropic's expected IPO.

The Cloud vs Local Debate

Despite the high costs, Steinberger believes AI should run on cloud servers rather than personal devices.

"I feel like the main challenge a lot of people still have is that our infrastructure hasn't really caught up yet," he said. "Like why is this still using my computer? Shouldn't this all be in the cloud?"

However, cloud-hosted AI agents are still very expensive. This is a major reason why many developers prefer running AI models locally to save money and ensure privacy.

Is AI ruining open-source software?

Knecht highlighted a growing problem: many open-source developers have stopped accepting code contributions (pull requests) because they are flooded with low-quality, AI-generated code, often referred to as "AI slop".

When asked if this could mean the death of open-source, Steinberger suggested that the submission process itself needs to change. He explained that developers need to show how the AI reached its conclusion. "If I see your thoughts, how you got from A to B, my trust in the output is way higher than if you just throw the finished work across the fence," he said.

The Future: Quality over Quantity

Will everyday users be running hundreds of AI agents simultaneously in the future? Steinberger doubts it.

Instead, he predicts a more streamlined, proactive future. "I think everyone will have some sort of intelligence that runs and thinks and helps them throughout the day or even predicts what could help you in the future," he explained. "So instead of asking the question, maybe you already get the answer."

Trusting AI with your data

Giving an AI agent permission to act on your behalf can be risky. While models have become much smarter since the early days of OpenClaw, Steinberger admitted that deciding what data to share is still a hard question.

"You might not want to give your agent all the permissions," he warned.

His solution? Use a second AI agent to act as a security guard. "You want to have another agent that can give it permissions or not, and then this agent has to negotiate with another agent," he explained. He noted that these agent-to-agent negotiations often catch mistakes and find safer ways to complete tasks within boundaries. "Those two systems combined greatly increase my trust that I just can let the agent rip," he added.

As models become more efficient and infrastructure catches up, the dream of having a trusted, cloud-based AI assistant managing our daily tasks may soon become a reality for everyone, not just developers with an unlimited budget.

- Ends