- Enterprise AI adoption is rising but so are token costs
- Storyblok’s Sebastian Gierlinger warned businesses are overspending by defaulting to expensive models and using AI where cheaper automation would work
- A variety of factors could keep token costs high over the coming years
AI may be rapidly spreading across the enterprise – including throughout telcos – but the cost of using it is still poorly understood, Storyblok VP of AI and IT Sebastian Gierlinger told Fierce. He warned many companies are overusing expensive models and underestimating the true economics of token consumption.
As more business units experiment with generative AI, organizations are facing a new challenge: managing consumption in a way that aligns performance needs with cost. Gierlinger said a lack of internal education around model selection can quickly lead to unnecessary spending, particularly when employees default to higher-tier models for routine tasks.
“If there is no education on that, people will automatically – most of the time – opt for the higher model and then automatically [use] more tokens than they need,” Gierlinger said.
He added that enterprises also often overlook how token pricing works in practice. Tokens aren’t just consumed when a result is generated, but also when content is fed to a model. Gierlinger noted that different models also employ different pricing strategies, with some charging more for ingestion and some more for outputs.
“The more content you feed to AI, the more tokens you will burn by ingesting this content,” Gierlinger said.
AI overkill and ballooning costs
One key advantage of AI is that its natural language interface has made it possible for more people across organizations to automate tasks. But this also has a dark side. Some people, Gierlinger argued, are burning tokens for AI workflows that could be handled more cheaply by conventional automation.
“People have option to automate things using Claude and that sounds very nice but we already had the chance to automate things before AI,” he said. “There are automation platforms where you can save a lot of tokens by using automation that is doing things over and over again instead of asking Claude.”
For companies pursuing widespread AI deployment, Gierlinger offered a blunt warning: “Throwing AI at everything, in my personal opinion, is not the right path forward because with the costs that we see at the moment it will backfire.”
Gartner has predicted that by 2030, inferencing on a 1 trillion parameter model will cost GenAI providers 90% less than it did in 2025, citing a combination of chip and infrastructure efficiency improvements and higher chip utilization. But it added falling token costs likely “will not be fully passed on to enterprise customers.”
Meanwhile, Bain & Company this week noted that agent and token costs for some business areas have already surpassed the cost of offshore human resources.
“Agents consume significant tokens on multistep reasoning, error correction and context loading, which add up fast on complex workflows,” Bain’s team wrote. “The speed and quality gains are real, but the per-task economics don’t always pencil out, particularly when done for the wrong tasks or without the appropriate orchestration.”
Bain also pointed out three factors that could keep costs high: while earlier version models are cheaper, most companies want to use frontier models; tokens used per query are rising as tasks become more complex; and teams are expanding their use of AI across more workflows.
That’s not to mention that Bain found token consumption is climbing faster than token costs are falling.
Looking ahead, Gierlinger said infrastructure realities could keep AI costs elevated rather than pushing them down.
“I think that the costs might even rise because energy is not getting cheaper anytime soon,” he said. His solution? “I would suggest to use it like we introduced tools the last 50 years. Do a cost analysis where it pays off and try to balance out and figure out where I can really get value out of it.”