Not long ago, a young startup built around a single generative AI tool shot from zero to millions in revenue within a year. At the same time, a well-funded corporate AI pilot in a Fortune 500 company fizzled out quietly, shelved after months of work and millions spent with nothing to show. These two stories capture the reality of the AI era: for a few, it is a rocket ship; for many others, it is an expensive dead end.
The AI revolution is not a future to anticipate. It is already here, unfolding in real time. New tools and models appear almost daily, promising to reshape industries and the way we live and work. But this is also a period of enormous risk. A single win can deliver billions, while a wrong bet can drain everything. A recent report by MIT’s NANDA initiative, titled “The GenAI Divide: State of AI in Business 2025,” published on Fortune, reveals just how stark the odds are: despite the explosion of platforms and research, only about 5% of generative AI pilots succeed. The vast majority stall, producing little to no measurable impact on business results.
Based on 150 interviews with industry leaders, a survey of 350 employees, and an analysis of 300 public AI deployments, the research reveals a sharp contrast between the rare breakthroughs and the many projects that falter.
READ: Foxconn, SoftBank team up on Stargate data center project in Ohio (
Reflecting on the findings, Aditya Challapally, the lead author of the report, said, “Some large companies’ pilots and younger startups are really excelling with generative AI. Startups led by 19- or 20-year-olds, for example, “have seen revenues jump from zero to $20 million in a year,” he stated. Challapally further added, “it’s because they pick one pain point, execute well, and partner smartly with companies who use their tools.”
But for the majority: 95% of companies in generative AI is failing to deliver. The problem is not the AI itself, but a “learning gap” affecting both the tools and the organizations using them. While executives often point to regulations or model performance, MIT’s research highlights deeper issues in how AI is integrated within enterprises. Generic tools like ChatGPT perform well for individual users because of their versatility, but they struggle in corporate settings where they don’t learn from or adapt to existing workflows, Challapally explained.
Challapally also emphasized how companies are channeling their AI spending. According to him, more than half of the budgets for generative AI are being poured into sales and marketing tools. The report suggests that if businesses redirected these investments by reducing dependence on outsourcing, trimming agency costs, and tightening internal operations, they could address many of the challenges slowing adoption.
READ: Amazon App Store to shut down on Android within 48 hours (
The report also pointed out that the way companies approach AI makes all the difference. Working with specialized vendors and forming partnerships has a success rate of roughly 67%, while in-house builds succeed only about a third as often. This contrast is especially striking in financial services and other tightly regulated industries, where many firms are trying to develop their own proprietary generative AI systems in 2025. But MIT’s research shows that going solo more often leads to setbacks than wins. “Almost everywhere we went, enterprises were trying to build their own tool,” noted Challapally, but the data showed purchased solutions delivered more reliable results.
The report also notes that success with AI often depends on giving line managers which is not just central AI teams. But the power to drive adoption, and choosing tools that can integrate smoothly and evolve with business needs.
At the same time, AI is already reshaping the workforce. The impact is most visible in customer support and administrative roles. Instead of large-scale layoffs, many companies are quietly reducing headcount by not replacing employees when they leave. Most of these changes are happening in roles that were often outsourced in the past because they were seen as low-value. Yet as companies try to formalize their AI strategies, “shadow AI” use is widespread, with employees relying on tools like ChatGPT. Adding to the challenge, many firms still struggle to measure AI’s real impact on productivity and profitability.