But what does the report say to unleash such sales wave? For what he says Sheryl Estrada In Fortune, American companies have invested between 35,000 and 40,000 million dollars in generative projects (Genai), but most efforts are stagnant in the pilot phase. According to the report of the so -called “Nanda” of the MIT, only about 5% of the initiatives generate rapid growth of income; Most have little or no impact, that is, they are failing. Apparently, the main problem is not the quality of the models used, but the lack of integration, learning and alignment with corporate workflows. Companies often invest in sales and marketing solutions, but the greatest benefits seem to be in the automation of the back-effect and the optimization of internal processes, says Viktor Eriksson of Computerworld.
The Nanda of the MIT media lab, “The Genai Divide: State of Ai In Business 2025“ He found that successful companies tend to buy specialized solutions and build associations, while internal development projects fail more frequently, and that despite the billions of dollars invested in generative, 95% of organizations do not obtain any commercial return. The report authors point out that “The results are so clearly divided between buyers (companies, middle market, SMEs) and developers (startups, suppliers, consultants) that we call it the Genai gap “ and argue that “Only 5% of integrated AI pilots are extracting millions in value, while the vast majority remains stagnant without a measurable impact on profits and losses.” The aforementioned report recognizes although the generative AI is promising for companies, most initiatives to boost rapid income growth are failing.
Estrada explains, according to the report that, despite the rush to integrate new powerful models, about 5% of the Pilot Programs of IA achieve a rapid acceleration of income; The vast majority stagnate, offering little or no measurable impact on profits and losses. The research, based on 150 interviews with leaders, a survey of 350 employees and an analysis of 300 public implementations of AI, describes a clear division between success stories and stagnant projects.
The main author of the report and research collaborator of the Nanda Project at MIT, Aditya Challapallyhe spoke with Fortune and told him that: “Some pilots of some large companies and new younger companies are really protruding with the generative AI; that startups led by 19 or 20 years old, for example, have seen how their income is fired from zero to 20 million dollars in a year, which is due to the fact that they identify a problem, execute it well and are intelligently associated with companies that use their tools”.
However, for 95% of the companies in the data set, the implementation of the generative AI is being short. The main problem? It is not the quality of AI models, but the “learning gap” for both tools and organizations. While executives usually blame the regulation or performance of the model, MIT research points to poor business integration. Generic tools such as Chatgpt stand out for people due to their flexibility, but they stagnate business use because they do not learn from workflows or adapt to them, Challapally explained.
The data also reveal a misalignment in the allocation of resources since more than half of the generative the budgets are allocated to sales and marketing tools; However, the MIT found the greatest return on investment in administrative automation, eliminating the outsourcing of business processes, reducing the costs of external agencies and optimizing operations.
So, what is behind the successful implementations of AI: the way companies adopt AI is crucial. The acquisition of AI tools to specialized suppliers and the creation of alliances are successful in approximately 67% of cases, while internal implementations only succeed in a third of cases.
“This finding is particularly relevant in the financial services sector and other highly regulated sectors, where many companies are developing their own generative systems patented in 2025. However, MIT’s investigation suggests that companies experience many more failures by acting alone,” Estrada comments.
According to Challapally, companies surveyed were often reluctant to share failure rates, almost all places where they were, companies tried to develop their own tool, but the data showed that the acquired solutions offered more reliable results. Other key factors for success include empowering line managers, not only the central laboratories of AI, to boost adoption, and select tools that can be deeply integrated and adapt over time.
On the disruption of the workforce, they indicate that it is already underway, especially in customer service and administrative functions: instead of mass layoffs, companies less and less cover vacancies; Most of the changes are concentrated in positions that were previously outsourced due to their low perceived value. The report also highlights the generalized use of “AI in the shadow” (unauthorized tools such as chatgpt) and the constant challenge of measuring the impact of AI on productivity and profits.
All specialized and non -specialized media echoed the report, above all, for its impact on the US stock market, highlighting the conclusions about the vast majority of failures. However, Tim Fries From The Tokenist, he argues that the MIT report has been misunderstood, scaring investors and dropping the technological markets.
“It was supposed that a new MIT research article on generative would shed light on the difficulties of companies for adoption. However, once traditional media such as fortune spread the findings (reducing nuances to a catastrophic head of AI markets as the risks of bad informative coverage on niche technologies ”Fries said that the report detected not only that 5% of AI pilots achieved rapid acceleration of income, but many others generated cost savings or operational efficiency without explosive growth.
For Fries the articles, not only of Fortune, but even of the Financial Times, which in addition to amplifying the expansive wave on the markets, simplified the conclusion that 95% of the generative the IA projects failed with a scarce or zero measurable impact, erasing the distinction between losing extraordinary income and not providing any value. That is, Fries argues that the media analysis turned a limited statistic into a generalized failure narrative.
On Sam Altman Openai had launched a warning on an AI bubble. This confluence of signals created the perfect storm for a sales wave in the technological sector, Considers FRIES who emphasizes that, given the repercussions, the MIT modified access to the report by a restricted access version that required a application form. Although no explicit explanation was offered, the time of publication suggests concern about the way in which research had been used as a weapon in the media and markets. The incident may underline two truths: first, that the markets of AI are hypersensitive to negative holders; Second, that the main economic media are still badly prepared to responsibly interpret the dense technical research. Where MIT spoke of a “learning gap” and integration challenges, Fortune heard failure. That distortion, amplified on a large scale, spread through global markets, wreaked havoc in the midst of other market forces, said Fries. We will see what the market says.
For now David Ramel Virtualizationreview commented, on the report, that companies execute the largest number of pilot programs, but convert the least amount; Middle market organizations pass the pilot faster to complete implementation (approximately 90 days) than large companies (nine months or more). On the other hand, it deepens that the general purpose tools are widely explored, but their impact is limited: more than 80% of organizations have explored or proven (chatgpt/co -ilot), and almost 40% report having implemented them. However, they mainly improve individual productivity, not the performance of the results account. Meanwhile, 60% of organizations evaluated business level systems, but only 20% reached the pilot phase and only 5% reached production. Hence they conclude that the root cause is the learning gap.
The central explanation of the report is that the main barrier is learning, more than infrastructure, regulation or talent: “Most Genai systems do not retain feedback, do not adapt to context or improve over time”. The report summarizes this gap succinctly: Chatgpt’s limitations reveal the central problem after the Genai gap: forget the context, it does not learn and cannot evolve. For complex and long -lasting tasks, humans remain the main option.
Source: Ambito
