What Cheaper AI Means For Business Leaders

Lindsey Witmer Collins is Founder and CEO of WLCM.AI, an Inc. 5000 company building AI solutions for enterprises, and Scribbly Books.

Lower AI prices can make more work affordable to automate. The business question is whether that work saves enough time or earns enough money to justify the cost.​

Every few months, another chart shows AI getting cheaper, and every few months a client asks me some version of the same question: Shouldn’t our bill be going down? Usually the answer is no—the bill tends to go up. The trick is knowing whether it went up for good reasons.

The latest version of the chart comes from a16z, using data Citadel Securities compiled through August 2026. Average prices per token—the chunks of text a model reads and writes—have been falling, while the cost of renting the chips that do the work has stayed stubbornly high. Cheap tokens are being sold on top of expensive compute, which suggests pricing will keep shifting under us.

The other thing to notice is what “average” means here. The average can fall without any provider cutting a price, simply because more buyers are routing routine work to cheaper models, the way the average price of a car drops when more people buy compacts. That’s largely what’s happening: Ramp’s data shows the share of business usage going to top-tier models fell from about 53% to 45% in a single month this summer. Your contract didn’t change. If you’re still paying what you paid last year, your team’s behavior probably did.

What you actually spend depends on things the price list doesn’t show: how hard you ask the model to work, how much material you hand it and how many tries it takes to get something usable. Here’s how I’d look at each:

1. Price the finished task, not the answer.

2. Pick the model, and then pick the effort.

3. Be stingy with what you send.

4. Expect to use more of it.

A lower token price might lower your spend, and it might increase it. Being smart about how you allocate them is the challenge worth rising to.​