GenAI FOMO: Businesses Burn $40B with 95% Seeing Zero Return
US companies have poured an estimated $35 billion to $40 billion into Generative AI initiatives, yet a recent report from MIT’s Networked Agents and Decentralized AI (NANDA) initiative reveals a stark reality: 95 percent of enterprises have seen virtually no return on these substantial investments. Only a mere 5 percent of organizations have successfully integrated AI tools into their operations at scale, leading to what the study authors term the “GenAI Divide.”
This sobering assessment is based on a comprehensive analysis, drawing from 52 structured interviews with enterprise leaders, examination of over 300 public AI initiatives and announcements, and a survey of 153 business professionals. The report, authored by Aditya Challapally, Chris Pease, Ramesh Raskar, and Pradyumna Chari, attributes this significant gap not to a lack of infrastructure, training, or talent, but rather to a fundamental limitation of current AI systems: their inability to retain data, adapt, and learn effectively over time.
The “GenAI Divide” is most pronounced in deployment rates. Custom enterprise AI tools, specifically designed for internal use, struggle to move beyond pilot phases, with only five percent ever reaching full production. While consumer-facing chatbots might initially succeed due to their ease of trial and flexibility, their lack of memory and customization often renders them ineffective in critical business workflows. As one unnamed Chief Information Officer candidly put it in an interview with the authors, “We’ve seen dozens of demos this year. Maybe one or two are genuinely useful. The rest are wrappers or science projects.” This sentiment echoes other recent research indicating a decline in confidence among corporate leaders regarding their AI endeavors.
Despite the widespread underperformance, the NANDA report acknowledges that a small fraction of companies have found Generative AI beneficial, particularly in the Technology and Media & Telecom sectors, where the technology has had a tangible impact. However, for the majority of industries—including Professional Services, Healthcare & Pharma, Consumer & Retail, Financial Services, Advanced Industries, and Energy & Materials—Generative AI has so far proven inconsequential. An anonymous Chief Operating Officer at a mid-market manufacturing firm summarized this disconnect, stating, “The hype on LinkedIn says everything has changed, but in our operations, nothing fundamental has shifted. We’re processing some contracts faster, but that’s all that has changed.”
The evolving employment landscape is one area where Generative AI’s influence is becoming more evident, particularly within the Technology and Media sectors. The report notes that over 80 percent of executives in these industries anticipate reduced hiring volumes within the next two years. These workforce reductions are largely occurring in non-core business activities that are frequently outsourced, such as customer support, administrative processing, and standardized development tasks. These roles were already vulnerable due to their outsourced nature and process standardization even before AI implementation, with the report suggesting that between five and 20 percent of support and administrative processing roles in affected sectors have already been impacted. This trend aligns with broader industry observations, where companies like Oracle and IBM have reportedly grappled with balancing AI capital expenditures or have faced accusations of using AI as a pretext for offshoring jobs.
A critical finding of the report is the misallocation of AI budgets, with approximately 50 percent typically directed towards marketing and sales. The authors advocate for a strategic shift, urging corporate investment towards activities that yield meaningful business results. This includes front-end processes like lead qualification and customer retention, and back-end efficiencies such as eliminating business process outsourcing, reducing ad agency spending, and streamlining financial service risk checks.
Intriguingly, the study highlights that generic tools like OpenAI’s ChatGPT often outperform bespoke enterprise solutions, even when the latter utilize the same underlying AI models. The primary reason, according to the report, is user familiarity and accessibility, a consequence of “shadow IT”—employees adopting tools independently. A corporate lawyer interviewed for the study perfectly illustrated this, expressing dissatisfaction with a $50,000 specialized contract analysis tool. “Our purchased AI tool provided rigid summaries with limited customization options,” she recounted. “With ChatGPT, I can guide the conversation and iterate until I get exactly what I need. The fundamental quality difference is noticeable, ChatGPT consistently produces better outputs, even though our vendor claims to use the same underlying technology.”
The report concludes that companies successfully navigating the “GenAI Divide” approach AI procurement not as standard software-as-a-service clients, but more akin to business process outsourcing customers. They demand deep customization, foster adoption from the front lines, and hold vendors accountable to measurable business metrics. Ultimately, crossing this divide requires a true partnership, not merely a purchase.