MIT Report Finds 95% of Generative AI Research Concentrated Among Few Companies

MIT's NANDA initiative released the "GenAI Divide: State of AI in Business 2025" report, showing that only about 5% of enterprise generative AI pilots deliver rapid revenue growth while the remaining 95% stall with little or no impact on the P&L. The study, based on 150 leader interviews, a 350-person survey, and analysis of 300 public deployments, draws a stark line between a few high-performing pilots and the vast majority that fail to move the needle.

Success stories come from startups that zero in on a single pain point, execute quickly, and partner with tool providers, generating revenue jumps from zero to $20 million in a year. In contrast, large firms suffer from a "learning gap"-the tools aren't integrated into workflows and executives blame regulation or model quality instead of flawed integration. Generic models like ChatGPT work for individuals but stall at scale because they don't adapt to corporate processes.

The report also uncovers a budgeting mismatch: over half of generative AI spend goes to sales and marketing tools, yet the highest ROI appears in back-office automation that cuts outsourcing, agency costs, and streamlines operations. Companies that purchase AI solutions from specialized vendors succeed about 67% of the time, while internal builds succeed roughly one-third as often, especially in regulated sectors like financial services.

Key levers for leaders include empowering line managers rather than relying solely on central AI labs, selecting tools that can learn and integrate deeply, and managing the rise of "shadow AI" where unsanctioned tools proliferate. Workforce disruption is already visible in customer support and admin roles, with firms opting not to backfill positions rather than mass layoffs.

Looking ahead, the most advanced organizations are experimenting with agentic AI systems that can learn, remember, and act autonomously within defined bounds, hinting at the next phase of enterprise AI adoption.