For the past couple of years, tech companies have treated AI features like a mandatory checkbox. If your app didn’t suddenly gain a chatbot, an image generator, or some sort of “AI assistant,” investors and customers started wondering if you were already falling behind. The problem is, many businesses rushed into AI before figuring out whether the economics actually made sense.
A new report from DigitalRoute suggests the industry may finally be running into reality. According to the company’s “AI State of Monetization 2026” study, only 8 percent of organizations say they fully understand the true cost of serving AI features. Nearly half of respondents also said rising AI-related costs are now a major challenge.
That sounds about right.
Unlike traditional SaaS products, where adding more users can become increasingly profitable over time, AI behaves differently. Every generated image, every chatbot response, every API call, and every AI-powered workflow consumes compute resources that cost real money. Folks using AI casually may not think about it, but behind every prompt sits expensive infrastructure powered by GPUs, storage, networking, electricity, and cooling.
This is one of the reasons so many AI companies keep pushing subscriptions, usage caps, premium tiers, and enterprise pricing. The “free AI for everyone” phase was never likely to last forever. Somebody has to pay the cloud bill eventually.
The report claims only 23 percent of organizations feel highly accurate at forecasting AI-related usage, costs, and revenue fluctuations. Meanwhile, 61 percent say revenue predictability has become harder. That is not exactly comforting if you are a company trying to build an entire business around AI services.
What makes this especially interesting is that AI pricing itself still feels chaotic. Some companies bundle AI into existing subscriptions. Others charge separately for advanced features. Some meter usage by token counts or API calls, while others experiment with outcome-based pricing. There is no standard yet because the industry still appears to be figuring things out in real time.
To be fair, this report comes from a company that sells monetization and billing infrastructure, so it clearly benefits from highlighting these problems. Still, the concerns themselves feel legitimate. Even outside this study, there are already signs that AI costs are becoming uncomfortable across the industry.
OpenAI reportedly continues spending enormous amounts on compute infrastructure. Microsoft keeps pouring money into AI data centers and GPU expansion. Google is embedding AI into nearly every product it offers while also trying to convince users to pay for premium AI subscriptions.
At some point, all of these companies need AI to become sustainably profitable rather than simply impressive.
That is where things could get messy. Consumers have already been trained by streaming services and SaaS apps to expect flat-rate subscriptions. AI does not always cooperate with that model because heavy users can generate dramatically higher backend costs than casual users. An “unlimited AI” plan sounds great in marketing copy until power users start hammering expensive inference models all day long.
There is also an uncomfortable possibility hanging over the industry: many AI features may not actually generate enough real-world value to justify their ongoing infrastructure costs. Plenty of people still use AI mostly for novelty, experimentation, or convenience rather than mission-critical work.
The AI race is clearly not slowing down anytime soon. But as businesses move from hype to actual financial reporting, investors will probably start caring less about who added AI first and more about who can make money from it without setting cash on fire.
