Anthropic’s mind-boggling IPO today (expected to be valued at $1.4 trillion or higher) shows how valuable investors view companies expected to dominate the AI landscape.
It also begs the question: How will these companies — and their new-found investors — make money?
A recent panel at the SAX Symposium had an idea. A tech blast-from-the-past to the beginning of the cell phone era.
Remember when cellphone plans gave you free nights and weekends but charged 10 cents for every text?
Maria Koller told a panel audience that’s where she thinks AI is headed.
Koller, Vice President of AI & Edge Infrastructure Growth at Verizon Business, said AI pricing will come down to tokens: tiny units of computing that get used up every time an AI tool produces an answer. She made the point at a fireside chat on AI in the workplace at the SAX Symposium, a few weeks back at MetLife Stadium.
“That’s how you think about tokens,” she said. “In my mind, that’s kind of what’s going to end up happening.
“You start to pay per text, start to pay per word, per token.”
Leon Grassi, a SAX Partner who moderated the session, said tokens were the one thing he wanted everyone to understand before they left. Then he asked for a show of hands from people who understood how tokens are used.
“See, that’s not enough,” he said.
Here’s the short version of what they are talking about.
Every time you ask an AI tool a question — even “What’s the weather today?” — AI writes its answer word by word. Each piece uses tokens, Koller said. Those tokens are the work being done by the chips in data centers.
Rob Owen, Chief Information Officer and a Partner at SAX, offered a simpler picture from a simpler time — playing games in an arcade.
“You can’t just use the games for free,” he said. “You got to put coins.”
So, why will the bill change?
Building AI is enormously expensive. Koller cited a recent Goldman Sachs report that puts AI capital investment at $1 trillion this year, covering data centers, chips and models. Data centers and the power to run them are limited, which is what makes tokens a commodity.
Companies like OpenAI and Perplexity will want that money back, Koller said. And she sees nothing wrong with that.
“They’re going to start to monetize it in a different way, and I think that’s a very natural way of doing commerce,” she said. “You build something and you figure out how to monetize it.”
Right now, most people barely notice. Koller said she’ll sometimes tell her kids they just used 20 tokens, and it doesn’t matter yet.
“But I think you’re going to start to see that bundling change,” she said.
Some models already charge less. Owen said Chinese models such as DeepSeek can do a lot of work for a fraction of the token cost, but users “sort of lose control of data privacy.”
For businesses, Owen said, the lesson is to understand how they are using AI now — and determine if it’s still worth it if you were being charged by the word.
Asking AI to rewrite an email is nice, he said, but the real value is in automating routine work at the lowest possible cost.
“If you can control when you use tokens and when you don’t … the movement of data, the control of data, and getting the repeatable process that follows a logic tree done through automation and orchestration in the least amount of cost that you can is really what this is all driving towards,” he said.
In other words, the free nights and weekends won’t last forever. Be ready to prioritize your usage when the pay structure changes.


