The soaring cost of running AI models has become one of the biggest threats to consumer AI startups.
This is forcing some of these young companies to slow growth, pivot to business products, or risk being copied by larger rivals with deeper pockets.
That’s the argument Kylan Gibbs, CEO of AI voice startup Inworld, is making as his company prepares to cut prices by more than 50% to help consumer-focused developers survive.
“Cost is the single number one problem,” Gibbs told me in a recent interview.
Ideally, as startups grow and usage increases, their unit economics typically improve, as costs either stay relatively flat or rise less than revenue. In the new world of consumer AI startups, this isn’t happening because running the underlying models that support their products is expensive, and these costs can increase as usage climbs.
“Their profitability goes through the floor every time they hit success,” Gibbs said. “We see that across lots of AI consumer applications.”
The issue centers on AI inference, the computing required every time a user interacts with a chatbot, voice assistant, or other AI application.
Large AI companies have their own infrastructure, can secure favorable AI chip pricing, and can spread costs across massive businesses. Meanwhile, startups often don’t have that power and end up paying more, per unit of intelligence, for access to AI models.
Gibbs said consumer AI startups frequently spend 70% to 90% of their operating budgets on inference costs. Many charge users just $5 to $10 a month, making profitability difficult as engagement rises.
“The irony is that users love these products,” Gibbs said. “Time spent keeps growing. But every time they become successful, their profitability falls.”
The economics differ sharply from enterprise software. A corporate customer might pay $1,000 a month for an AI tool and compare that cost with hiring an employee. Consumers, meanwhile, are far more price sensitive and sometimes more fickle when choosing which apps to use.
As a result, many startups hit a growth ceiling. Some stop investing in marketing, others pivot toward business customers, and some disappear altogether. Gibbs said that the dynamic ultimately benefits the largest AI companies, which can replicate popular features and distribute them through existing products.
The broader market reflects what Gibbs describes as inflated AI pricing. Inference providers and AI model developers often price services relative to competitors rather than their underlying computing costs, he said, generating revenue that can be many times the cost of compute.
Inworld can do something about this because it’s one of the leading providers of voice-based AI models.
Gibbs is responding by lowering prices and offering deeper discounts as customers scale.
The company, which has raised more than $117 million and says revenue has grown fivefold since the start of 2026, is betting that cheaper AI infrastructure will allow a new generation of consumer applications in areas such as education, therapy, health, and fitness to reach massive scale.
“If we want AI benefits to reach everyday people,” Gibbs said, “the economics have to work for the companies building those products.”
Sign up for BI’s Tech Memo newsletter here. Reach out to me via email at abarr@businessinsider.com.
