A new report from MIT’s NANDA initiative, The GenAI Divide: State of AI in Business 2025, has revealed a striking reality: while generative AI promises to reshape business operations, most corporate attempts to harness it are falling short.
According to the study published by Fortune, which draws on 150 leadership interviews, a survey of 350 employees, and an analysis of 300 public deployments, only about 5% of generative AI projects are achieving rapid revenue acceleration. The overwhelming majority either stall or deliver minimal measurable impact on profitability.
Speaking on the findings, Aditya Challapally, the lead author and research contributor to MIT’s NANDA project, said a small fraction of organizations—often startups—are excelling because of their sharp focus and strategic partnerships.
Register for Tekedia Mini-MBA edition 19 (Feb 9 – May 2, 2026): big discounts for early bird.
Tekedia AI in Business Masterclass opens registrations.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab: From Technical Design to Deployment (next edition begins Jan 24 2026).
“Startups led by 19- or 20-year-olds have seen revenues jump from zero to $20 million in a year,” Challapally explained. “They pick one pain point, execute well, and partner smartly with companies who use their tools.”
For the remaining 95%, the gap is less about the AI models themselves and more about how businesses integrate them. MIT describes this as a “learning gap”—both on the side of the tools, which often fail to adapt to enterprise workflows, and the organizations, which struggle to embed AI meaningfully into processes.
While executives often cite regulatory barriers or imperfect model performance, MIT’s findings suggest the real issue lies in enterprise adoption strategies. Off-the-shelf AI models such as ChatGPT work well for individuals because of their flexibility, but inside businesses, they often underperform without workflow-specific adaptation.
Misaligned Investments
One of the most striking insights from the report is that more than half of corporate AI budgets in 2025 are flowing into sales and marketing tools. Yet, MIT’s data shows the highest returns are being realized in back-office automation, where AI can cut outsourcing costs, reduce reliance on agencies, and streamline repetitive operations.
This mismatch between investment and impact underscores why so many initiatives fail to deliver revenue growth.
“Almost everywhere we went, enterprises were trying to build their own tool,” Challapally noted.
But the report found that companies that purchased AI tools from specialized vendors succeeded 67% of the time, while internal builds had only a one-in-three success rate.
Sectoral Implications
The divide is particularly visible in financial services and other regulated sectors, where companies are racing to build proprietary AI models. MIT’s findings suggest this strategy is backfiring: instead of establishing a competitive advantage, many firms are wasting resources on systems that struggle to scale or adapt.
Workforce Shifts
The study also points to workforce disruption already underway. While fears of mass layoffs have not materialized, businesses are quietly reshaping their labor forces by not replacing workers in administrative and customer support roles as they leave. Much of this work was already outsourced, making AI’s entry a direct replacement for offshore contractors rather than core employees.
Shadow AI
Another concern is the rise of “shadow AI”—unsanctioned employee use of tools like ChatGPT to bypass internal restrictions. This poses compliance and security risks, particularly for industries handling sensitive data. Meanwhile, companies still struggle to measure AI’s real impact on productivity, making it difficult to justify investments internally.
What Works: Lessons from Success Stories
MIT’s research highlights several common features of successful deployments:
- Empowering line managers—not just central AI labs—to lead adoption.
- Choosing tools that can integrate deeply and adapt over time.
- Partnering with vendors instead of attempting costly solo builds.
Looking ahead, the most advanced firms are already experimenting with agentic AI systems—models that can remember, learn, and act autonomously within defined boundaries. These tools could mark the beginning of AI as a proactive decision-making partner in business, rather than just a passive assistant.
Challapally concluded that the next frontier will depend less on raw model power and more on organizational agility: “Companies that adapt their workflows and decision-making structures to AI will pull ahead. Those that don’t risk being stuck in endless pilots that never deliver.”