Millions of people use free versions of AI tools like ChatGPT, Gemini, and Claude every day. Behind the scenes, companies like Microsoft, Google, and Anthropic have spent hundreds of billions of dollars building the technology that powers these tools.
To recover some of that money, these firms sell paid versions of their AI products. These paid versions often include extra features for tasks like coding or business use.
Other companies are also building services on top of these AI models. Many of these services use AI agents, which are trained to complete specific jobs on their own.
Setting a fair price for these services has turned out to be harder than expected.
"Trying to tie someone into a cost model for the next 12 months, two years, three years, it doesn't make any sense, honestly, because we don't know," said Simon Gooch of identity management firm Saviynt.
How Tokens Drive the Cost of AI
Every time someone asks an AI tool a question, that request gets broken down into small pieces called tokens. The AI's response is also made up of tokens.
This process is not always predictable. Small changes in wording can lead to very different answers, even when nothing else changes.
When businesses use several AI agents together to complete a task, token use climbs even faster. This makes total costs harder to predict.
The price of each token has dropped over the past few years. But the number of tokens being used has grown at a much faster pace.
Goldman Sachs expects business token use to rise 24 times between 2026 and 2030. That would bring monthly use to about 120 quadrillion tokens.
Some companies have already been caught off guard by their AI bills. Microsoft has reportedly limited how much its engineers can use certain third party coding tools.
Uber reportedly used up its entire AI coding budget for the year in just a few months.
Will Venters, from the London School of Economics, said many firms are struggling with this problem. "People are finding it really hard to manage that cost… it's a non-deterministic output, so it's a non-deterministic value," he said.
Businesses Look for Ways to Manage the Cost
Some smaller companies have found ways around high costs. Oliver King-Smith, founder of smartR AI, said some firms use flat fee personal accounts to avoid higher charges.
He said this approach has limits. "This has to end at some point in time, because the big guys are taking a bath on those accounts," he said.
He expects larger AI companies to eventually tighten these rules once shareholders push for profits.
King-Smith said businesses should think more carefully about which AI models they choose to use for each task.
Rob Steele, finance chief at accounting software firm iplicit, said clearer instructions can help control costs. He compared it to grocery shopping without a list.
Venters said costs can grow fast once a company builds AI into a product used by thousands of people. Extra tokens may be needed for testing, security, or safety checks, not just the main task.
He added that adding more AI agents is simple, unlike hiring more staff, which involves planning and cost review.
Bill Peterson of Sumo Logic said his company is still deciding how to charge for new AI based security services. Options include raising prices, charging per result, or selling bundles.
Peterson said pricing plans could shift again if the large AI companies change how they charge for their own models.
"You get into variable pricing, and it's changing every couple of months," he said. "Customers don't like that. That's not how anybody builds a budget."
