Enterprises around the world have spent a jaw-dropping $30–40 billion on generative AI. Yet according to new research from MIT, 95% of those projects deliver zero measurable results.
The study calls this gap the GenAI Divide, a chasm between the hype of AI adoption and the reality of business transformation.
On one side of the divide are employees and consumers. Tools like ChatGPT and Copilot have become everyday companions for drafting emails, summarizing documents, and brainstorming ideas. They’re easy to use, flexible, and crucially work straight out of the box.
On the other side are the big corporate AI projects. Custom-built tools and vendor solutions rarely get past the pilot phase. Out of 100 companies that evaluate such systems, only five actually make it to full deployment. Why? Because most AI systems don’t learn, don’t remember, and don’t fit into day-to-day workflows.
Even the sectors that have spent the most, finance, healthcare, retail, show little disruption. Only Tech and Media have seen real change, with AI-native startups and new content models shaking up incumbents.
Meanwhile, a “shadow AI economy” is thriving: 9 in 10 employees quietly use personal AI accounts at work, often getting more value than from their company’s expensive pilots.
The lesson is simple: AI isn’t failing because of bad models or regulators. It’s failing because companies buy demos, not solutions. Until AI can integrate, adapt, and improve over time, the billions will keep going down the drain.