The AI Investment Gap: Why 95 Percent of Enterprise Projects Deliver Nothing

The number

Enterprise AI spending hit $30-40 billion over the past year. MIT found that 95% of those projects delivered no measurable business value.

I read that number twice. It seemed too high to be real.

The problem was not the models. MIT pointed to something else: a "learning gap" and "flawed integration strategies." Companies bought powerful AI systems and bolted them onto processes that could not support them.

Talk without investment

The St. Louis Federal Reserve analyzed 185,999 earnings calls from 7,047 companies between 2008 and 2024.

Before ChatGPT launched in late 2022, when executives spoke positively about AI, their companies followed through. Capital spending rose 3.1%. R&D spending rose 9.7%.

After ChatGPT, AI came up in nearly every earnings call. Executives promised big things. But the correlation between AI talk and actual investment dropped to zero. Companies talked. They did not build.

What failure looks like

Forward raised $650 million to build AI-powered medical kiosks that would diagnose patients and draw blood automatically. The blood draws failed regularly. Patients sometimes got stuck inside the pods. The company shut down in November 2024. $650 million gone.

Humane built a $699 AI Pin. It overheated. The interface was hard to use. The product failed.

AI-powered investment funds fared no better. Scientific American found that AIEQ returned 63% while the S&P 500 returned 108% over the same period. MIND returned -12% before shutting down in 2022. An index fund, left alone, beat the funds built to beat it.

Who is making the decisions

EDUCAUSE surveyed more than 900 higher education institutions about their AI policies. Only 23% had actual policies in place. 71% of AI initiatives were led by administrators with no practitioner input. 48% said their institutions lacked clear guidelines for AI decisions.

The people setting AI strategy often have not built anything with it. They have read the papers and attended the conferences. That is not the same as watching a system fail with real users and fixing it.

Before your company invests in AI, ask one question: who on this team has deployed an AI system in production and watched it work, or not work, with real users? If the answer is nobody, you are on track to join that 95%.

Why the failures repeat

Consultants bring slide decks. Vendors promise results. Executives want to avoid falling behind competitors. Nobody in the room has dealt with how a model trained on clean data behaves against a company's messy, inconsistent, real data. Nobody has explained to a user why the AI gave a confident, wrong answer. Nobody has handled the edge cases that make up 40% of real use cases but never showed up in training.

Companies also measure the wrong things. "60% of our employees use AI tools" sounds good until you ask what they are using them for. Rewriting emails that were fine already does not count as value.

One study found businesses lose 6% of annual revenue to decisions made on bad AI-generated data. Only 40% of IT leaders fully measure the ROI of their AI programs. The rest hope it works out.

The real cost

$30-40 billion is the direct number. Add the opportunity cost of what else that time and money could have funded. Add the fatigue after a third AI initiative fails. Add the credibility lost when a promise to the board does not hold.

When these projects fail, people blame AI. AI did not fail. The implementation failed. The strategy failed. The decision to let people without AI experience make AI decisions failed.

What the 5% did differently

The projects that delivered value shared a pattern: people who had worked with AI in production, not just data scientists who built models. People who had deployed systems, maintained them, fixed them when they broke, and learned how much messier reality is than training data.

They started small. They picked problems suited to AI. They measured real outcomes, not usage counts. They treated AI as a tool, one that is only as good as the person using it and how well it fits the job.

Before you spend

Put your decision-makers in a room with one question: can anyone here describe three specific ways AI systems fail in production, and how to handle those failures? If you get silence, do not spend the money yet.

AI is powerful. Power without understanding is expensive noise. $30-40 billion of it, so far.

Sources

MIT Sloan Management Review: "The $100 Billion AI Bubble." Finding: $30-40 billion enterprise investment, 95% zero measurable return. Published July 2025. Referenced in multiple analyses including Computer Weekly and financial media coverage.

St. Louis Federal Reserve Analysis: "AI Hype or Reality? Shifts in Corporate Investment after ChatGPT." Kalyani, A., Ozkan, S., Bass, M., & Dueholm, M. (October 3, 2024). Analysis of 185,999 earnings calls from 7,047 U.S. firms (2008-2024). https://www.stlouisfed.org/on-the-economy/2024/oct/ai-hype-reality-shifts-corporate-investment-chatgpt

EDUCAUSE 2024 AI Landscape Study: Survey of 900+ higher education technology professionals. Published May 23, 2024. https://www.educause.edu/research/2024/2024-educause-action-plan-ai-policies-and-guidelines

Computer Weekly: "AI hype hits reality roadblock." Vanson Bourne study of 550 organizations. Finding: 6% global annual revenue lost to AI-based misinformed decisions. https://www.computerweekly.com/news/366609259/AI-hype-hits-reality-roadblock

Scientific American: "Don't Trust AI for Important Things Such As Investment Decisions." Analysis of AI-powered investment funds (AIEQ, MIND, and others). Published February 19, 2025. https://www.scientificamerican.com/article/ai-makes-unreliable-investment-decisions/

Failed Ventures Documentation: Forward shutdown: multiple news sources, November 2024. Humane AI Pin: industry coverage of product issues and market reception (assets later acquired by HP, early 2025). Analysis: "The $100 Billion AI Bubble" report documenting high-profile failures.

Luke Paxton. Square Mile Design, May 2026.