Don’t use Swiss Army knives for surgically precise work; use scalpels.
That’s what Manos Koukoumidis and other players in the AI cost-saving business say is the biggest problem with how companies use AI. They’re using the most powerful, versatile, and expensive AI models — from frontier labs like Anthropic, OpenAI, and Google — to handle niche tasks within their organizations.
Koukoumidis, the CEO of the Washington-based AI company Oumi AI, called this practice “widely irrational and inefficient,” resulting in staggering AI bills.
Founded in 2024, Oumi lets individuals vibe-code their own niche AI models within minutes, tailored to their tasks and goals. It raised $10 million in its 2024 seed round.
Oumi is one of the many companies that have sprung up in the last two years to solve the biggest enterprise problem of the day: reducing highly inflated enterprise AI spending and increasing its returns on investment. As companies move away from the tokenmaxxing trend — where employees were encouraged to burn as many AI tokens as they could — these cost-saving services are becoming increasingly sought after.
Some of these companies are consulting firms and measurement platforms that advise clients on integrating AI into their workflows in a smart way. Others, like Oumi, are building new products, such as model builders and inference platforms, to tackle the problem from the source.
The coaches
The first category is the coaches: the consultants and advisors lighting up the path to AI success.
Sydney-based business consultancy Adaptovate has been operating for more than eight years, with over 100 consultants across offices around the world. But over the past three years, the company has zeroed in on helping others get the most bang for their AI spending buck.
Michael Murphy, a partner in the company, said Adaptovate works with clients at all stages of their AI adoption journey — from a snacks manufacturer that’s just figuring out how to use AI in its supply chain, to a 30,000-employee-strong professional services company that wants to restructure all its teams around GenAI products.
“It always starts with the strategy,” Murphy said. “Figuring out how decision-making changes. How does your talent model change? How does your organization’s structure change when you’re beginning to scale AI across the full organization?
The measurers
Then there are the measurers, who use software tools to help companies pinpoint where AI is working for them and where it isn’t.
One example is the San Francisco-based Larridin. CTO Ameya Kanitkar told Business Insider that the company acts as a measurement layer across its clients’ AI tools, employees, agents, and spending.
One of Larridin’s products is a data analysis tool that shows how employee productivity varies with token spend, helping companies identify the sweet spot for the token budget they allocate to their staff.
Kanitkar, who has worked at LinkedIn, Coinbase, and Groupon, said Larridin got off to a slow start because companies weren’t particularly concerned about AI spending in the early days of adoption. But that has shifted after they started pouring millions into the resource, and he’s seeing Larridin’s traction double every quarter since the start of the year.
“It just makes sense that you need a measurement layer to make sure that all those hundreds of millions of dollars you’re spending are actually justified and are properly spent,” he said.
Larridin raised $17 million in seed funding from Andreessen Horowitz, Bloomberg Beta, Google Ventures, and other investors and launched the platform in 2025. It now serves clients from data center construction companies to biosciences and financial services firms.
The builders
The third category is the builders: the startups like Oumi that are creating products to directly help companies slash AI costs.
Another example is Runware, a company cofounded by serial entrepreneur Ioana Hreninciuc. Runware provides the inference infrastructure companies need to run AI models quickly, cost-effectively, and at scale. This makes it much easier for companies to scale up their AI products without worrying about AI outages.
For those buying their own GPUs, there’s Tensormesh. The startup focuses on Key-Value Cache, cofounder Junchen Jiang said, something that “never appears in people’s token bills” but certainly affects their costs.
Tensormesh’s caching system helps businesses “accept more queries on fewer GPUs,” Jiang said. The company raised $20 million from big-hitters in hardware like AMD and NVentures, Nvidia’s venture capital arm.
And even more are gestating, ready to hit the market. Conifer is still participating in Y Combinator’s summer 2026 batch. Its cofounder, Charles Muehlberger, said the company had already raised $1.3 million, about 20% of its desired raise, before Demo Day. The company’s product will synthesize queries, break them into pieces, and feed the pieces to different models (including local ones).
“We think the main customer is one who is focused on cost,” said Michael Jeffords, Conifer’s other cofounder.
Bottom line
While the coaches, measurers, and builders are cutting the AI-cost-saving pie in different ways, their advice is largely the same: stop using frontier AI models for every menial task, start using lighter, open-source models, and build your own models to reduce costs and reliance on the big AI labs.
They’re also proposing new ways to think of ROI. Kanitkar from Larridin said AI costs should be treated as capital expenditures, not operating expenses, and companies shouldn’t expect to see gains immediately, but rather over a period of a year or two.
They’re all united on one belief — that companies jumped on the AI hype train without proper planning, and most don’t really know what they’re doing.
“Everybody’s pretending their AI spending is paying off,” Hreninciuc from Runware said.
“The incentive right now is for people to say it is paying off and to find a way to justify it, and because of this, we don’t have an accurate view of the market, as nobody wants to be the canary in the coal mine,” she said.
Read the full article here
