There's the pattern across the last six months of the 2026 earnings season, once you notice it: functioning companies are being punished for spending on AI, struggling companies are being punished for not disrupting fast enough, and almost nobody is being rewarded for actually using the technology at a profit yet. Nobody running these companies could tell you what a good AI outcome looks like right now, and the market is scoring them anyway.
On July 23, 2026, Alphabet beat on revenue and Cloud and then lost 7.1%, or roughly $300 billion, in a single session. $195 to $205 billion in 2026 capex did what a beat on revenue and Cloud couldn't. Free cash flow turned negative for the first time in the company's history. The story matches the size of the selloff.
What spooked investors was the size of the AI cheque the company said they'd keep writing. Duolingo swung the other direction and back again within a year, down 38% after backlash to its "AI-first" announcement, then recovering as investors decided the fear had been overstated, then down again on renewed competitive worry.
IBM had its worst single trading session on record on July 14, 2026, after an unscheduled letter from CEO disclosed preliminary Q2 revenue and earnings below consensus, tied to delayed deals and a late June client shift into AI hardware. The stock closed down 25.21%, erasing close to $68.8 billion in market value, and dragged Salesforce and ServiceNow down with it in the same session.
None of these companies changed what they actually do between one trading day and the next. What changed was a headline about what AI might do next, and the market treated the headline as the fact. Consulting was supposed to be the industry AI helped, not the one it hollowed out. Firms like Accenture, Gartner and McKinsey spent three years selling AI transformation to everyone else. Now the market is asking who transforms them.
Accenture posted a beat and raise adjacent quarter on June 18, 2026, earnings per share up 9%, margins expanding, and still lost 18% of its market value in a single session, its worst day as a public company. Gartner had already shown the same pattern in February. It beat on earnings while its Consulting segment contracted 13 to 15%, and the stock fell nearly 21% in a day. McKinsey, Bain, BCG, Deloitte, KPMG, Accenture and PwC have each also cut headcount or slowed hiring over the same period, concentrated in the junior research and back office roles where generative AI has compressed delivery timelines that used to require analyst hours.
Forrester's 2025 professional services research found firms using AI tools report roughly 40% productivity gains, translating directly into lower demand for the junior staff who used to do that work by hand.
The IT sector is experiencing a similar shift since the launch of ChatGPT. The first rung of the career ladder is disappearing, with U.S. IT employment among 22–25-year-olds down 23%. The pyramid that built the modern consulting firm was never really about expertise. It was about billable bodies, and AI is the first technology that has made the bodies optional.
Set against all of this, the return on investment data tells a more sobering story than the disruption headlines do, and it cuts the other way. MIT's Project NANDA studied 300 public AI deployments and found that 95% of enterprise generative AI pilots delivered no measurable P&L impact, against $30 to 40 billion in enterprise spending. S&P Global found that 42% of companies abandoned most of their AI projects in 2025, up from just 17% the year before, and the average enterprise scrapped 46% of its AI proofs of concept before they ever reached production.
Most of that money bought a model. It did not buy the distribution path, the workflow, or the customer relationship a model needs to actually move a company's bottom line. That gap is the real reason the layoffs and the stock swings are arriving well before any project has proven it can pay for itself.
The odd part is that all of this spending, hiring and firing is happening around a target every major AI lab defines differently. There is still no agreed upon definition of artificial general intelligence. OpenAI defines it as a system that outperforms humans at most economically valuable work. Meta is pursuing "superintelligence" without specifying a finish line at all. Even the tone has shifted.
Amodei’s October 2024 essay on powerful AI, "Machines of Loving Grace," was largely optimistic about curing disease and extending lifespans. His follow up, "The Adolescence of Technology", warned the same technology could be "the single most serious national security threat we've faced in a century, possibly ever." Anthropic's official position, submitted to the U.S. Office of Science and Technology Policy, is that powerful AI systems will likely emerge in late 2026 or early 2027.
Google DeepMind's Demis Hassabis defines AGI more narrowly, as a system that can exhibit the full range of human cognitive capabilities, and puts the odds at roughly 50% by the end of the decade. As of May 2026 he had narrowed that window further, to 2029 or 2030.
Samotsvety's superforecasters, who hold no financial stake in the answer, put the same probability at 28% by 2030, a sharp compression from the 32% by 2042 they forecast just three years earlier. None of this disagreement has slowed the spending down. If anything, it has done the opposite. An undefined target is easier to fund than a defined one, because nobody can be proven wrong yet.
That budget decision is now large enough to carry systemic weight. Amazon, Alphabet, Meta and Microsoft are on track to spend roughly $725 billion combined on AI infrastructure in 2026, up about 77% from roughly $410 billion in 2025. That spending is better understood as a bet on owning the compute chokepoint than as a bet on any single model proving itself, since whoever controls the infrastructure gets paid regardless of which model wins on top of it. That is a rational bet for the hyperscalers to make. It is a much riskier one for everyone financing it on their behalf.
The Bank for International Settlements said as much in its Annual Economic Report 2026, published June 28, warning that hyperscalers are increasingly financing data centers through joint ventures and special purpose vehicles capitalised by private credit, keeping the debt off their own books, with the risk that the same underlying assets get pledged more than once. Private credit loans to AI related companies grew from roughly $3 billion in 2010 to over $40 billion in 2025. That's the kind of structure that shows up as a $68.8 billion single day loss once a warning like IBM's lands.
Underneath all of it, moats built on switching costs, the workflow lock-in that keeps customers into a platform rather than any real product advantage, are eroding at a pace the companies losing them did not plan for. Morningstar reevaluated moat ratings for 132 companies in its coverage in March 2026 specifically to test for AI disruption risk. Twenty-two wide moats were downgraded, twenty to narrow and two to none, concentrated in enterprise software, IT services and payroll, hitting names like Workday, Adobe, Salesforce and ADP.
Only two companies were upgraded, both for infrastructure positioning rather than application-layer stickiness. Chegg, the American ed-tech company, remains the clearest casualty of that shift. It lost nearly half its stock value in a single trading day in May 2024 after telling investors that ChatGPT was pulling students away from its homework platform, and it never recovered. Its market capitalisation fell from a 2021 peak of $14.7 billion to just over $100 million by April 2026, and the company cut its workforce twice, first by 22% in May 2025, then by another 45% that October. Chegg's mistake looks obvious only in hindsight. It had spent years assuming convenience was a moat. AI made convenience free, and there was nothing underneath it.
What connects the stock swings, the layoffs, the ROI numbers and the spending is a market pricing certainty into a technology whose own builders cannot agree on what it is or when it arrives. That is not how mature markets are supposed to work, and it will not last. Markets have priced technologies ahead of their definitions before. Usually the correction comes from the definition catching up to the money, not the other way around.
The harder problem belongs to the companies that already cut staff. Choosing to rebuild a workflow around AI before anyone knows what the technology can actually do is one decision. Rebuilding that workflow around what the technology turns out to do, once the people who could have done the rebuilding are gone, is a different one. Most of these companies are finding that out the hard way.