OpenAI's top researchers are burning $140K on tokens per month — Cathie Wood calls it a sign of massive productivity

In a Sept. 8 X post, Wood said soaring AI usage among OpenAI’s researchers, who are writing more code, conducting more experiments and deploying multiple AI agents at once, points to the potential for massive productivity.

“Since the ChatGPT moment, token usage inside OpenAI has scaled astoundingly: to a median $12,000 per researcher per month and $140,000 per top-level researcher per month,” Wood said, citing an OpenAI blogpost. “Their agents seem very busy, and productive. Could be a leading indicator of massive productivity gains!”

That $140,000 figure appears high

Despite impressing business executives like Wood, experts say the $140,000 monthly token spend figure isn’t what a normal OpenAI researcher spends.

“It’s the top of their range, and their own numbers put the middle of the pack about ten times lower,” Nathan Thornhill, owner at 3Rivers WebTech in Fort Wayne, Indiana, told Moneywise.

While OpenAI says researchers using AI agents are writing substantially more code and running more experiments, tech experts, like Thornhill, say they’d need evidence to conclude those increases represent genuine productivity gains rather than simply more activity.

He advises business leaders deploying AI agents to confirm whether the work done is actually finished and measured by something other than lines of code and token spend.

Is there a world where $140,000 per month is cheap?

Some AI analysts say it’s possible that $140,000 worth of AI tokens a month could be considered budget-friendly one day, but only if it replaces months of work otherwise done by a team of people.

And you can’t tell that just by looking at the token bill alone, Arjun Jaggi, an applied AI researcher and executive advisor in Santa Monica, California, told Moneywise. “You have to compare cost per validated outcome against what the realistic alternative would have actually cost in salary, time, and missed opportunity.”

At the very least, OpenAI’s top researchers are well-equipped to use AI agents, seemingly justifying the expense in 2026.

“They understand how these models behave, know how to specify tasks precisely, and can spot subtle errors most people would not even notice,” he said.

That’s exactly the skill most companies don’t have yet.

“For this kind of gain to actually show up in corporate profits, wages, or GDP, most organizations would need employees who can direct and critically judge AI agents at something close to that level, which is a hiring and training problem, not a token pricing problem,” Jaggi added.

Where does the token trend leave U.S. workers?

A larger question for businesses and workers is whether consuming vastly more AI computing power actually translates into more valuable work. And whether what’s happening inside one of the world’s leading AI labs can eventually spread to the broader economy, which could use a productivity boost.