Two recent studies have coincidentally sounded alarms: enterprise-level artificial intelligence (AI) applications are facing severe challenges.
Three weeks ago, a study released by the Massachusetts Institute of Technology (MIT) claimed that up to 95% of enterprises are receiving zero returns from their generative AI investments. Last Sunday, Torsten Sløk, Chief Economist at Apollo Global Management, pointed out that AI adoption rates among large US enterprises are showing a declining trend.
Both studies reveal significant obstacles in the transition from AI technology hype to practical application. Sløk, citing US official survey data, stated that enterprises with more than 250 employees are slowing down their AI adoption pace. This could signal a corporate reevaluation of AI technology's actual value.
The MIT report further analyzes the underlying causes of this phenomenon, pointing out that the problem lies not with AI models themselves, but with flawed internal integration strategies within enterprises. These two research findings have triggered strong market reactions, causing the Nasdaq index to record its largest single-day decline since August 1st, with AI-related stocks like Nvidia experiencing significant sell-offs.
**Large Enterprise AI Adoption Rates Fall into Decline**
Torsten Sløk's analysis is based on large-scale enterprise surveys conducted by the US Census Bureau under the Department of Commerce every two weeks. The survey covers 1.2 million enterprises, asking whether companies have used AI tools such as machine learning, natural language processing, virtual agents, or voice recognition in the past two weeks to help produce goods or provide services.
The chart above presents a six-survey moving average from the US Census Bureau. Survey data shows that AI adoption rates among large enterprises employing more than 250 people are declining. This trend indicates that while the market is enthusiastic about AI, large enterprises may be experiencing a "technology disillusionment period" at the practical application level, beginning to reassess the actual value and return on investment of AI tools.
This decline in adoption rates may reflect integration challenges enterprises encounter after initial attempts, as well as difficulties in converting AI tools into actual business value. For investors, this data signal suggests that the commercialization path of AI technology may be more complex than previously anticipated.
**MIT Study Reveals AI Investment Dilemma**
The MIT NANDA project report "The Generative AI Gap: The State of Business AI in 2025" released on August 18th provides deeper analysis. Based on interviews with 150 enterprise leaders, surveys of 350 employees, and analysis of 300 public AI deployment cases, the study found that only about 5% of AI pilot projects achieved rapid revenue growth.
Lead author Aditya Challapally pointed out that the core problem lies in "learning gaps" within enterprises and flawed integration strategies. Many enterprise leaders mistakenly attribute failures to regulatory environments or model performance, while ignoring organizational adaptation and integration issues.
For example, general-purpose tools like ChatGPT, designed for individual users, are popular due to their flexibility but often perform poorly in enterprise environments because they cannot effectively learn from specific workflows or adapt to enterprises' specific needs. This "one-size-fits-all" application approach has resulted in numerous AI projects failing to bring measurable financial impact to enterprises.
**Key Differences in Successful AI Implementation**
The MIT study also analyzed key differences between successful and failed AI deployment cases. A few successful enterprises, particularly some startups, adopted strategies of "focusing on single pain points, precise execution, and building intelligent partnerships." Challapally mentioned that certain startups led by young people achieved "revenue jumps from zero to $20 million within a year" through this approach.
The research found that over half of generative AI budgets are used for sales and marketing tools, yet the largest return on investment actually comes from back-office automation, such as applications that reduce business process outsourcing and external agency costs. This indicates enterprises may have misjudged their AI investment directions.
Another key finding is that "buying" outperforms "building." The success rate of purchasing AI tools from professional suppliers and establishing partnerships is approximately 67%, while the success rate of enterprises building systems internally is only one-third. This data poses a direct challenge to companies that have invested heavily in attempting to establish proprietary AI systems.
**Market Reaction and Investment Impact**
The MIT research results had significant market impact last month. The day after the report's release on August 20th, US tech stocks plunged, with the Nasdaq Composite Index falling 1.4%. Nvidia, a core beneficiary of the AI boom, dropped 3.5%, while Palantir and Arm fell 9.4% and 5% respectively.
A trader close to a multi-billion dollar US tech fund reportedly stated, "This story is causing people to panic."
This sentiment shift resonates with OpenAI CEO Sam Altman's recent warnings about "overly excited investors" potentially forming an AI bubble, further intensifying market skepticism about AI technology commercialization prospects.
The MIT report's release coincided with growing concerns about tech stock overvaluations, with the Nasdaq 100 index trading at a forward P/E ratio of 27 times, nearly one-third higher than its long-term average.
For investors, these two studies provide important risk signals, indicating the need for prudent evaluation of AI-related companies' valuations and actual implementation capabilities, rather than focusing solely on technological breakthroughs and market enthusiasm. The AI revolution may continue, but its commercialization path is more complex and lengthy than expected.