Stock markets around the world took a hit this week after a new MIT study raised serious questions about whether artificial intelligence companies are actually delivering on their promises. The research suggests that despite all the hype and massive investments, very few AI startups are seeing meaningful returns from the technology.
The study, called “The GenAI Divide: State of AI in Business 2025,” came from MIT’s NANDA initiative and looked at 300 real AI implementations across various companies. The researchers also surveyed 350 employees and spoke with 150 business leaders to get a complete picture of how AI is performing in the real world.
What they found wasn’t encouraging for investors. Only about 5% of companies testing AI technology saw rapid revenue growth. The vast majority hit roadblocks and couldn’t show measurable improvements to their bottom line.
AI Stock Sell-Off and Market Trends
The news sent shockwaves through tech markets. Major AI-related companies saw their stock prices drop on Tuesday, with Nvidia, Palantir, Arm, Oracle, AMD, and AppLovin all losing single-digit percentages. The sell-off wasn’t limited to individual companies either.
The Nasdaq fell 1.4%, the S&P 500 dropped 0.7%, and international markets followed suit. Europe’s Stoxx index declined 0.6%, Japan’s Nikkei 225 slipped 1.5%, and South Korea’s Kospi fell 0.6%.
The MIT report wasn’t all doom and gloom. Lead researcher Aditya Challapally told Fortune that the AI companies that are succeeding share some common traits. They typically focus on solving one specific problem really well, rather than trying to be everything to everyone. They also work closely with their clients to understand exactly what’s needed.
Some of these focused startups have impressive growth stories. The most successful ones managed to grow their revenue from zero to $20 million within just one year. But these companies represent a tiny fraction of the AI landscape.
The research also revealed something interesting about where AI works best. While most investment dollars have flowed into flashy marketing and sales tools, the real money is being made in less glamorous back-end automation. Companies are finding success with AI that streamlines outsourcing and other operational processes that customers never see.
Why building your own AI is a bad idea
Another key finding challenges how most companies approach AI. The study showed that businesses trying to build their own AI tools from scratch are far less likely to succeed than those who buy specialized solutions from expert providers.
The most effective AI tools are highly specialized and adapt to specific business parameters. This is quite different from general-purpose systems like ChatGPT, which are designed to handle a wide range of tasks but may not excel at any particular business function.
The MIT findings align with warnings that have been coming from some unexpected sources. Even OpenAI CEO Sam Altman, whose company helped spark the current AI boom, recently cautioned that many new AI startups are setting unrealistic expectations. He warned that this could create a bubble that leads to “historic losses” when it eventually bursts.
AI: A Modern Bubble?
Economist Torsten Slok has gone even further, arguing that the AI hype is worse than the dot-com bubble of the 1990s. Financial Times columnist Robert Armstrong has pointed out that the S&P 500’s strength has become dangerously dependent on Nvidia and other AI chipmakers.
Tech analyst Edward Zitron highlighted another concerning trend: Nvidia’s GPU sales remain the primary source of actual AI-related profits, accounting for nearly 8% of the entire US stock market’s value. Since a handful of tech giants now make up more than one-third of the market, any disruption to this narrow profit base could have widespread consequences.
The MIT study suggests we may be reaching a reality check moment for the AI industry. While the technology clearly has potential, the gap between promise and performance appears wider than many investors realized.