The question is no longer whether enterprise AI has failed. It is simply at what scale, and at whose expense.

According to a 2025 MIT study analysing 150 interviews with executives, 350 employee surveys, and 300 public AI deployments, approximately 95% of generative AI pilots fail to deliver measurable return on investment. This is not a margin of error. This is an indictment.

American enterprises have invested an estimated $40 billion in artificial intelligence systems over the past 18 months, yet, as one data analyst noted, "95% of companies are seeing zero measurable bottom-line impact from their AI investments." Simultaneously, 42% of companies abandoned most AI initiatives entirely in 2025, up from just 17% in 2024. The acceleration of retreat is as striking as the breadth of failure.

What makes this particularly damning is not merely that projects fail—all innovation carries risk. What matters is that the failure pattern reveals something worse than incompetence: it reveals that corporations have systematised the illusion of progress whilst extracting nothing of value from it.

The MIT research identifies the root causes with surgical precision. Enterprise AI stalls because generic tools like ChatGPT, however sophisticated, do not learn from or adapt to bespoke workflows. A company's data lives in departmental silos. Budget allocation favours sales and marketing deployments (over 50% of AI spending) despite better potential returns in operations and finance. Large firms lead in pilot volume but lag in successful deployment—a paradox suggesting that resources translate to risk-taking without translating to returns.

Most damning: internal AI builds succeed only one-third as often as tools purchased from specialist vendors and built in partnership. Yet companies continue to bet on internal development, presumably because it feels like control, even when the data says it means failure.

The "GenAI Divide" thesis—that roughly 5% of pilots achieve rapid revenue acceleration whilst 95% stall—suggests something subtly different from random failure. It suggests that success requires specific conditions so rare that the industry cannot reliably replicate them. The fact that success is statistically anomalous, not normal, means that AI has not become business infrastructure. It has become a lottery disguised as strategy.

But the story grows darker. Whilst enterprises pour capital into failing systems, the industry simultaneously engages in systematic deception about what AI can accomplish.

The FTC has now filed twelve AI-washing cases since 2024, with four in 2025 alone. In February 2026, the agency resolved a case against Growth Cave, which had misrepresented that its "AI software" would "automate nearly 100% of the process" of setting up online education courses. In August 2025, Air AI was sued for falsely claiming its product could replace human customer service representatives. Workado's "AI Content Detector" promised 98.3% accuracy in identifying human versus AI-written text; testing revealed 74.5% at best, and 53.2% for non-academic content.

These are not edge cases. They are symptoms of a market where vendors have zero incentive to distinguish between pilot success and actual commercial value because both generate revenue in the short term. A company that spends $4 million on a failed AI project still enriches the vendor. The vendor's incentives point toward volume and visibility, not outcomes.

Meanwhile, the industry continues to manufacture false credibility elsewhere. A February 2026 analysis of over 150 claims made by major tech companies and institutions about AI's climate benefits found 74% completely unproven. Thirty-six per cent of those claims cited no evidence whatsoever. The analysis examined corporate websites, press releases, and sustainability claims from OpenAI, Google, Microsoft, and others—and found not a single verified example of consumer generative AI systems delivering material, measurable, substantial emissions reductions.

ChatGPT. Gemini. Copilot. None have demonstrably reduced carbon emissions. Yet the narrative persists that AI will solve climate problems, embedded in investor decks and sustainability reports across the Fortune 500.

What ties these failures together is a market structure that rewards confidence over evidence. Vendors succeed financially whether their products succeed operationally. Enterprises blame "poor data governance" or "lack of AI readiness" rather than the technology itself. Investors treat pilot volume as a proxy for progress. Regulators, despite nascent efforts via Operation AI Comply, remain vastly outnumbered by the marketing machinery.

The 95% failure rate is not an anomaly to be solved through better implementation. It is the market revealing its true nature: that most of the $40 billion was never going to produce value. It was always going to produce *the appearance* of progress, strategic positioning, and executive confidence.

The uncomfortable question nobody in the boardroom wants to answer is simple: if 95% of your pilots fail, if 42% of your peers have already quit, and if you cannot name a single concrete financial improvement from your AI spending, at what point does continued investment become not strategy but capitulation to industry narrative?

The answer, evidently, is not yet. But the pile of failed projects grows higher each month.