The investment world's love affair with artificial intelligence is hitting a hard reckoning. After years of pouring billions into AI systems with little proof they'd deliver value, major corporations are now facing the reality that the technology is expensive to run and often fails to justify its costs.
The shift is dramatic because it's coming from the companies that spent the most. Uber burned through its annual Claude Code budget in just four months, with executives struggling to connect the massive spending to better customer features. Amazon discovered employees were gaming an internal AI leaderboard with throwaway tasks, prompting executives to tell staff to stop using AI "just for the sake of using AI." GitHub moved its Copilot coding assistant to a pay-per-use model, forcing millions of developers to confront actual costs they'd never seen before.
A Bain survey of 951 large companies found projected AI savings falling far short of expectations, even as most firms planned to increase spending. The firm's blunt conclusion: "The technology worked. The value didn't arrive."
What's striking is where this skepticism is coming from. Three years ago, doubts about AI came from outside critics and short sellers betting on a crash. Today's skeptics work inside the companies that led the charge into AI deployment. Even OpenAI CEO Sam Altman acknowledged the concern is legitimate, calling questions about whether AI spending translates to revenue "the most fair criticism" of the current moment.
The market delivered its own warning Friday. The Nasdaq dropped 4.2% in its worst day in over a year, while the semiconductor index plunged 10.3%, its worst performance in more than six years. The catalyst was Broadcom, a chipmaker that reported massive AI revenue growth but disappointed investors by failing to raise its long-term AI revenue projections. Wall Street read the signal instantly: demand may not be accelerating as promised.
The paradox is that AI itself works. The technology genuinely boosts productivity when deployed with precision. Chipmakers, AI labs, and specialized power users are capturing real value. The problem is the broader business model: companies bet they could spray AI across their entire operations, plug it into every workflow, and watch productivity soar automatically. That assumption was the real bubble.
Most of the broader economy hasn't felt these shocks yet. The early adopters absorbing the cost explosions and token-burning inefficiencies are the same companies that moved fastest. Latecomers may learn from these mistakes, or they may repeat them. Either way, the next phase of AI deployment will look fundamentally different from the rush of the past two years.
Author James Rodriguez: "The gap between what AI can do and what it actually does for a company's bottom line is where the real story is now, and that's where the money will either be made or lost."
Comments