AI stocks have lost about $5 trillion as investors ask when all the spending pays off. Boards will soon ask the same of every AI line in next year's budget, and the answer depends less on the technology than on what organizations build around it.
Since early October, the biggest AI and chip stocks have lost about $5 trillion in market value. Alphabet had its worst day as a public company on Oct. 29, falling 12.6% after search growth slowed and it signalled capital spending of about $200 billion this year and about $300 billion in 2027. Oracle is raising $20 billion in new stock and debt to keep building. Then came the midterms. The morning after Democrats won both chambers of Congress, the broader market rallied while Nvidia and several data-centre operators slipped, on expectations of closer oversight of the AI buildout.
The message from investors is plain: show us the return. That question won't stay on Wall Street. Boards, CFOs and public sector budget offices are going to put it to every AI line in their 2027 plans.
This is the dip before the payoff
Investors are right to ask, but the slowdown isn't proof that AI doesn't pay. Economists have a name for this stage: the productivity J-curve. When a major new technology arrives, productivity often dips before it rises, because organizations first have to spend on retraining people, redesigning processes and reorganizing work. Those costs show up right away, while the gains arrive later.
What makes AI harder than past technologies is the pace of change. The models keep improving, often faster than an organization can absorb the last version. Leaders are being asked to redesign how work gets done and show results at the same time, on a foundation that keeps shifting underneath them. That makes the dip longer and harder to sit through.
The payoff isn't automatic, though. McKinsey's 2026 State of AI survey found that only 6% of companies attribute 5% or more of their operating earnings to AI. Those companies aren't waiting for better technology. They've done the organizational work, and that's what separates them from everyone else.
Why pilots stall
About 90% of function-specific AI use cases in McKinsey's survey are still in pilot mode. Pilots stall for a familiar set of reasons, and most of them have little to do with the technology itself.
The value was never defined. Many pilots start because a tool is available, not because there's a clear problem worth solving. Without a measurable target and real demand from the business, there's nothing to justify scaling. The leading performers in the survey are more than twice as likely to have a defined way of measuring AI's impact.
The work around the tool didn't change. A pilot can succeed with a handful of enthusiasts, but scaling means changing roles, handoffs and decision rights across a whole function, and that's where adoption breaks down. Workers report the gains, with 80% of AI users in the survey saying they're personally more productive. Those gains rarely add up at company level when the workflow stays the same. Nearly three-quarters of high performers redesigned workflows around AI, compared with about a quarter of everyone else.
The cost of keeping up wasn't budgeted. One in five respondents said AI operating costs had limited their use. And every time a model changes, something has to be retested, retrained or rebuilt. Pilots funded as one-time projects tend to run out of money just when they need to grow.
Budget for the change, not just the tool
The 2027 AI budget shouldn't read like a technology purchase order. Leaders should change three things.
Fund training as a core cost. If people don't know how to use the tools well, or don't trust what they produce, adoption stalls. Training should be a standing line that grows as use grows, not a one-time onboarding session.
Set aside money for version upgrades. The models will keep changing. Budget for the work of testing new versions, updating integrations and retraining staff, so each upgrade improves results instead of disrupting them.
Tie each dollar to a workflow and a measure. Before scaling, name the process that will change, who owns it and the number that will show it's working. If no one can answer those questions, the pilot isn't ready to grow.
For public sector organizations, the stakes are higher. Agencies work within fixed budgets and have to explain any dip to taxpayers. That's an argument for funding the redesign and the training up front, so the payoff arrives within a budget cycle rather than after one.
The market will settle its own argument about AI valuations. Inside organizations, the question is simpler: did you buy a tool, or change how the work gets done?
Planning your 2027 AI budget, or trying to get a pilot to scale? Let's talk.
Sources
UPS · Reuters · AP · Motley Fool · 24/7 Wall St. · TradingKey · Yahoo Finance