Forty-eight percent of executives call their AI adoption a massive disappointment. Eighty-six percent are increasing their budgets anyway.
BCG surveyed 2,360 executives in January 2026 and found that forty-eight percent called their AI adoption a massive disappointment, up from thirty-four percent the year before. The finding was published the same week Axios reported that Microsoft had canceled most Claude Code licenses over runaway costs and that one unnamed company burned five hundred million dollars in a single month after failing to set usage limits.
The surveys behind the headline are worse. MIT studied three hundred public AI deployments and found ninety-five percent delivered zero measurable impact on profit and loss. S&P Global reported forty-two percent of companies abandoned most AI projects in 2025, more than double the prior year's seventeen percent. McKinsey found a seventy-three percent ROI failure rate across $665 billion in enterprise AI spending. Gartner estimated only twenty-eight percent of AI infrastructure projects fully pay off.
The stock market priced the disappointment before the surveys confirmed it. Salesforce dropped twenty-six percent after an earnings call where record Agentforce growth could not offset fears about a stagnating CRM core. ServiceNow beat consensus on nearly every metric and fell fourteen percent after hours. Cloudflare sank twenty-four percent and cut eleven hundred employees. Adobe slid thirty percent from its fifty-two-week high despite revenue growth. On a single April trading day, Cloudflare, Snowflake, ServiceNow, and Salesforce lost tens of billions in combined market capitalization. Wall Street labeled it the SaaSpocalypse. Two trillion dollars in software value has been erased since January.
Every technology cycle produces disappointment. What makes this one structural is what happens after the disappointment registers. In the BCG survey, only six percent of executives said they would pull back on AI spending even if their initiatives never paid off. Eighty-six percent are increasing AI budgets in 2026. Ninety-eight percent of technology leaders face board pressure to demonstrate returns, and seventy-one percent say they will cut only if first-half targets are missed. The data describes a ratchet: spending increases survive evidence of failure.
When an analyst asked Mark Zuckerberg on Meta's Q1 earnings call for visible evidence that $145 billion in annual AI spending was producing proportional value, he called it a very technical question. He did not answer because he did not need to. Meta's AI budget is not an investment with a return threshold. It is insurance against the scenario where a competitor builds something Meta cannot replicate without the infrastructure already in place. The same calculation runs through every boardroom where the CFO presents a negative AI ROI and the board votes to spend more.
MIT identified what separates the insured from the insurers. Companies that allocate more than five percent of their AI budget to training and workflow integration significantly outperform those that do not. The twenty-one percent of S&P 500 companies that Morgan Stanley found can cite a measurable AI benefit are concentrated in this group. They changed how work gets done. The seventy-nine percent who cannot cite a benefit purchased the option and deferred the exercise.
This distinction matters for capital allocation. Only fourteen percent of CFOs in a March 2026 survey reported measurable returns on third-party AI investments. The SaaS companies being punished most severely are those whose AI narrative runs on pilot announcements, integration partnerships, and feature labels attached to products generating the same revenue per seat as before. Salesforce's Agentforce crossed $1.2 billion in annual recurring revenue and the stock is still down thirty-three percent for the year. The market has learned to separate AI revenue from AI value.
The thesis is falsifiable. If enterprises cut AI budgets in the second half of 2026, the optionality framework breaks. Seventy-one percent of technology leaders say they will rationalize if first-half targets are missed. Watch third-quarter earnings calls for the word rationalization. Its absence would confirm that the insurance premium holds and that the gap between AI spending and AI returns will widen through 2027. The companies paying the premium are not irrational. They have decided that being wrong about AI is cheaper than being late.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)