Generative AI has dominated boardroom agendas, but the latest data from MIT’s NANDA initiative offers a sobering truth: 95 percent of corporate pilots aren’t delivering measurable business impact.
A gap is widening between companies experimenting with AI and those translating it into profit, according to the GenAI Divide: State of AI in Business 2025 report, based on interviews with 150 leaders, surveys from 350 employees, and analysis of 300 public deployments (MIT, 2025).
The “GenAI Divide” is defined as a structural gap between model intelligence and organizational intelligence. The former is advancing exponentially; the latter, linearly at best. This is a subtle but crucial distinction. Generative AI thrives on iteration: it improves through exposure to data and feedback loops. But most corporations still treat AI as a plug-in technology rather than a co-evolving system and the result is what MIT calls a learning gap: tools that can learn, embedded in organizations that can’t.
MIT found that over half of AI budgets flow into sales and marketing tools, even though the highest ROI comes from automating procurement, finance, and operational workflows. Back-office AI doesn’t make headlines, but it quietly eliminates outsourcing costs and speeds up processes that drive profitability.
“The core barrier to scaling is not infrastructure, regulation, or talent. It is learning. Most GenAI systems do not retain feedback, adapt to context, or improve over time.” the report said. “A small group of vendors and buyers are achieving faster progress by addressing these limitations directly.”
Those buyers that “succeed demand process-specific customization and evaluate tools based on business outcomes rather than software benchmarks,” the report explained. “They expect systems that integrate with existing processes and improve over time. Vendors meeting these expectations are securing multi-million-dollar deployments within months.”
The few companies that are winning with AI follow consistent patterns:
To outsiders, a 95 percent failure rate might suggest the AI bubble is bursting. But history tells a different story. The report draws parallels to the railroad boom of the 1840s and the dot-com era of the late 1990s: both periods in which massive speculative failures ultimately laid the groundwork for lasting infrastructure.
Even as enterprises stall, employees aren’t waiting. 90 percent of employees use AI tools personally, from ChatGPT to Claude, while fewer than half of companies have formalized adoption plans. This “shadow AI economy” represents the fastest bottom-up tech adoption in corporate history, quietly reshaping work beneath IT’s radar. Rather than suppress it, leaders who formalize and secure this grassroots momentum could close the learning gap faster than any new model release.
The numbers sound quite grim: almost 100 percent failure rates, vanishing ROI, and cooling investor sentiment. Yet history tells us otherwise. The railroad mania of the 1840s and the dot-com bust both produced infrastructure that reshaped the world. Generative AI may be following the same curve: a short-term wave of failed experiments paving the way for long-term transformation.