So far, the drumbeat of AI in the workplace has been about how technology will replace workers and cut costs.
But an executive at tech firm Nvidia recently revealed that the cost of AI computing “is far beyond the costs of the employees,” according to an Axios article. And Amos Bar-Joseph, CEO of Swan AI, recently said on LinkedIn that the company’s AI bill hit $113,000 for a single month of data for a four-person team.
In some ways, this feels like an alternate reality. Weren’t we promised that AI could replace various human work tasks — more productively, more efficiently, and (importantly) more cheaply?
What happens if the reverse is true?
» MORE: Are we in an AI bubble?
Misleading sticker price
At first glance, AI seems pretty affordable. Subscriptions start at about $20 a month. The problem is scale — because many companies aren’t stopping with one subscription.
“A $30-a-month tool usually becomes six or seven of them, plus API overages, plus a workflow platform tying it together,” says Scott Tobin, founder of Delegate, a tool that helps small businesses plan for AI hiring.
Additionally, when companies hit a certain number of subscriptions, they might move to a different payment model where they’re charged for actual data used. The jump can be startling.
“Costs can be going from something like $10,000 a month to $50,000 a month,” says Jesse Lipson, founder and CEO of marketing platform Levitate.
The result is a scenario where AI costs can eclipse the cost of human workers.
Where we go from here
Instead of the predominant anxiety that AI will replace millions of human workers, the real scenario may be that AI displaces some roles but also shifts which ones are considered essential. And companies will have to figure out how to balance the cost of AI with human workers.
“Some junior roles are getting automated almost entirely,” Tobin says. “But mid-level roles with judgment and stakeholder-management components are getting more valuable, not less.”
And even if AI boosts output, the labor savings may not materialize because human workers will still have to supervise, verify and coordinate the technology.
“You may be getting more productivity, but you may not be able to get it out of fewer people,” says Nils Gilman, editor-in-chief of the Berggruen Press at the Berggruen Institute, a think tank based in Los Angeles. “Everybody is becoming a middle manager of their team of bots.”
Meet MoneyNerd, your weekly news decoder
So much news. So little time. NerdWallet's new weekly newsletter makes sense of the headlines that affect your wallet.
