cross-posted from: https://hexbear.net/post/8624879
https://www.axios.com/2026/05/28/ai-spending-roi-enterprise-costs
Archive link https://web.archive.org/web/20260528114303/https://www.axios.com/2026/05/28/ai-spending-roi-enterprise-costs
Why it matters: Companies that rushed to embrace AI are now confronting ballooning IT costs, uncertain productivity gains and growing employee skepticism.
Driving the news: Microsoft canceled most of its Claude Code licenses, in part over costs, according to The Verge, and Uber’s COO said AI costs are getting “harder to justify.”
An AI consultant tells Axios one of their clients recently spent half a billion dollars in a single month after failing to put usage limits on Claude licenses for employees. Companies are citing AI's ability to automate jobs as a cause for layoffs, though Anuj Kapur, CEO of CloudBees, told Axios that workforce cuts may simply be "the only lever they can pull" to offset their AI bills. Consumer sentiment around AI is also nosediving, and employees are rebelling against the use of the technology at work.What they’re saying: The enterprise is undergoing a “healthy swing” away from AI overuse — or “tokenmaxxing,” the push to burn as many AI tokens as possible — Ali Ansari, CEO of model training firm Micro1, told Axios.
Ansari hopes this correction will push companies toward more efficient AI use. While the market views these tools as working equally well across the enterprise, Ansari says "the reality of AI right now is that it only works for coding." That disconnect can drive up IT bills without leading to high return on investment in agents, he said.Friction point: Corporate AI adoption is running into four unique problems.
Use cases: "Most people default to automating tasks they dislike rather than tasks most valuable to the company," Sophia Velastegui, CEO of Velastegui Ventures and former chief AI officer at Microsoft told Axios. Instead, they should focus on using AI to drive revenue. Costs: One CTO told Axios that employees were using AI models to check the weather. That gets expensive fast: Enterprise AI plans are not truly 'all you can eat,' and even simple chatbot queries can carry heavy token costs.
One CTO told Axios that employees were using AI models to check the weather.
Is that not what every AI proponent has been telling everyone to do?
Why pull up your calculator app to find out what 12+17 is? Just ask the AI. Why use Ctrl+F to find keywords in a document when you can upload the document to the AI and have the AI find those words in it for you? Why look out the window when you can just ask the AI what the weather is like outside?
We’ve been telling them all along that this kind of shit is incredibly stupid and wasteful, both in monetary, but also in social and environmental costs. Are they just now finally starting to catch on?
Mind you, the people who are pushing to replace everything with LLMs are not the people who spend half a billion bucks on tokens in a month. They’re the ones who charge half a billion bucks and the ones who charge them for the hardware. As far as they are concerned, that half a billion dollar bill is a rousing success they’d like to see repeated as often as possible.
Okay, and then there’s the useful idiots who vibecode a 50 kLOC basic CRUD application with broken auth in two days and conclude that LLMs can craft arbitrarily complex applications instantly at near-zero cost. And then proceed to shill the stuff every chance they get because these days the internet is all about hyping yourself up and they can pretend that their finely-honed 1337 prompt crafting skillz will make them as god-kings among peasants when vibecoding will inevitably subsume all other forms of development, nay, all forms of creative work entirely!
While remaining cheap, of course, because nobody has ever offered a service for cheap and then made it more expensive.
We do this with every single new revolutionary technology.
Computer boom and busts
Even fucking railroads. They were building tens of thousands of kms of railroad to literal nowhere. “Towns will settle as long as you build a station and run a service there!”
Which is partially true in land that was a PITA to get to, and couldn’t access lucrative markets. I.e. a lot of prairie in the Midwest, which only truly settled around rail due to grain elevators and a more structure co-op system.
Not only that, but when your employees are loading information/documents into the AI, they’re often giving away company secrets and parts of their IP.
Half a BILLION in one Month is MUCH Cheaper then $100000 a YEAR for a Human!
-CEOS!
I don’t even want to eat them. They’re hardly even worth being hog feed and I wouldn’t want to feed too many of them to my hogs. There’s an obviously high level of heavy metals in them.
Mealworms > Hogs > CEO Bacon
Yeah that’s a good strategy. Good for chicken feed too then they can be the new CEO of breakfast
i bet it’s even more fatty than pork bacon. lol
Just to see half Billion in Numbers:
500,000,000.
That is 5,000 x 100,000a month
You know… Bigger question is: how many employees racked up that bill? In a single month?
Even if its 100k employees that means every one of them used $5k tokens - in a month.
But just think, someday you could replace all of your $100k/yr employees with $500m/mo AIs!
Corporations to workers: “Use AI for everything you do!”
Workers use AI.
Corporations to workers: “Not like that!!”
“Only do it in the way that makes me more money.”
“Don’t worry, boss. I type ‘make money for my boss, please’ into the AI prompt every hour on the hour. So you should have tons of money rolling in by now.”
¯\_(ツ)_/¯
The business version of a kid using their parents credit card to buy hundreds of loot boxes.
I had more fun opening loot boxes than using AI at work though
Except that the parent gave them explicit permission and instructions to do so, because some marketing guy told the parent that loot boxes are the future.
C suite douche: “Most people default to automating tasks they dislike rather than tasks most valuable to the company”
Also C suite douche: “AI makes workers more productive, therefore we don’t need as many workers!”
Can’t have it both ways, ya twats.
This wasn’t an ai problem, hate to say it. They failed to understand their user base in their business case and have no apparent controls over licenses or access. A whole lot of management heads would roll at my company if they failed their basic IT duties this hard.
Isnt that the whole area of AI though. Its not based on fundamentals. Its not based on proven productivity gains. Sure, anecdotally, some engineera are increasing their output dramatically. But also many people are feeling prodictice while producing slop.
AI has the potential to reduce so many monotonous tasks and streamline processes that currently need a human.
Companies and managers are rushing in based on that potential, trying tonfigure out how to meet it as its new. However, the big driver seems to be a fear of missing out. Thats a bubble.
The facr that antropic seems to have overtaken openai and google has overtaken them in the consumer space says that maybe nobody will be ledt that far behind, if they just wait for a use case.
Sure, anecdotally, some engineera are increasing their output dramatically.
And it’s questionable whether they really are.
A) Are they actually producing more? Some studies on the subject have found that coders who use AI think they’re more productive that way, but their productivity actually goes down when objectively measured.
B) Yes, they’re churning out lots and lots of code, but is the code any good? Does it even work? Is it riddled with bugs that will have to be fixed later?
C) Is all that code they’re churning out maintainable? Does anyone working on it actually understand how it works? Will they be able to make updates and changes to it over time? … Or is all this ‘productivity’ coming at the cost of piling up huge amounts of technical debt in the future that will have to be paid when future devs have to wade through and fix all the AI slop code?
The amount of “what does this regex do?” - “dunno, ChatGPT wrote that for me” I see in code reviews these days doesn’t inspire much confidence in either B or C. I don’t have direct observational evidence for or against A but at least there’s maybe a claim for people who restrict its use to very specific use cases.
We all know the answer to C. In a couple of years it will be a good time to open a “tech debt reduction” consulting company.
Ah the age-old question: do you pay for tokens, or engineers?
They used to say something along the lines of “you don’t buy a drill, you buy a hole.”
Any tool, whether a widget, software, AI, or labour, is a means to an end. This might be something coherent, like “we bought sheet metal to make tortilla presses, which we can sell for more than they cost to make” or a a rejected Bond-villian vision like “if we acquire all the potassium in the world, they’ll be forced to declare our boss the next king of eSwatini.”
Specifically, right now, the “end” seems to be “if we vapourize enough cash, investors will buy our stock becsuse we’re the most True Believers in the current trend.” I’m not sure, but the potassium plan might make more sense.
Depends, are you trying to impress customers or investors?
As a reminder, these things run on normal computers, and they are giving them away for free.
Does your company have computers, y/n?
