The honeymoon is over. After months of aggressive AI spending, major corporations are pumping the brakes, confronting eye-watering bills and questioning whether their seven-figure bets on the technology have actually paid off.
Microsoft recently cut most of its Claude Code licenses due to costs, while Uber's chief operating officer acknowledged that artificial intelligence expenses have become hard to defend. The problem is stark: one consultant told Axios that a client burned through half a billion dollars in a single month after failing to cap employee access to Claude licenses. When the bills arrive, companies are reaching for the same solution: layoffs.
Some executives are openly suggesting they're using job cuts as a tool to offset AI spending rather than because the technology genuinely replaced those workers. Anuj Kapur, CEO of CloudBees, told Axios that workforce reductions may simply be "the only lever they can pull" to fix their balance sheets.
The backlash extends beyond the C-suite. Consumer trust in AI is collapsing, and employees are openly resisting the technology's introduction at work. What started as transformational promise has become a cost containment problem.
Several structural issues are creating this deadlock. First, companies are automating the wrong things. Rather than targeting high-value business functions, most organizations default to eliminating tasks that annoy employees. Sophia Velastegui, a former chief AI officer at Microsoft, told Axios that leadership is making a strategic error. "Most people default to automating tasks they dislike rather than tasks most valuable to the company," she said. The focus should be on driving revenue.
Second, token economics are brutal. Enterprise AI plans sound unlimited until you see the bill. A CTO reported that employees were using advanced AI models to check the weather. Simple chatbot queries carry surprising costs, and the math breaks quickly when multiplied across an entire workforce.
Third, human behavior is slowing adoption. Leadership is often not strategic: throwing licenses at problems and hoping something sticks, Velastegui said, isn't producing tangible returns. Employees and IT teams are still learning how to use these tools effectively, and that learning curve has a price tag.
Finally, data constraints are strangling AI usefulness. When companies refuse to give AI agents full access to proprietary information, those agents become less effective, according to Josh Pantony, CEO of Boosted.ai, which builds AI tools for finance teams. The security and liability concerns are real, but they limit what AI can actually accomplish.
Ali Ansari, CEO of AI model training firm Micro1, says the market is undergoing a healthy correction away from what he calls "tokenmaxxing," the race to burn through as many AI tokens as possible. The dirty secret nobody wants to admit: AI, right now, actually works well for one thing: coding. Everything else is still experimental and expensive.
The question now is whether companies will get disciplined or overcorrect entirely. Either way, the age of "move fast and spend heavily" on AI is ending fast.
Author James Rodriguez: "Corporate America bet billions that AI would solve everything and is now discovering that throwing money at the problem doesn't work if you don't know what problem you're solving."
Comments