• Global Research
    • General Research Insights
    • Data science
    • Disruptive technology

Beyond the bubble

  • Article

A recent study challenges the “AI bubble” narrative, reporting that 74% of companies see measurable benefit from AI investments, which is a major contrast to the gloomy 95% failure rate reported by MIT NANDA earlier this year.

Reassessing the “AI bubble” narrative

Fears of a trillion-dollar “AI bubble” are growing among investors. Critics claim that artificial intelligence is overhyped, with unclear return on investment (ROI) and limited commercial applications. Mentions of the phrase “AI bubble” in news stories have surged in recent months (Chart 1). This has contributed significantly to any mention of the word “bubble” in the financial media (Chart 2). As a result, some investors are concerned that markets have already stepped into a bubble territory. We disagree.

95% – the dubious figure that started it all

A report released by MIT NANDA (a project overseen by MIT Media lab) in July 2025 contained a startling statistic: “95% of organizations are getting zero return” from their GenAI investments. This single result has had a disproportional impact on the AI bubble debate, despite the rather weak nature of the evidence backing up this startling-sounding claim.

More recent evidence disagrees

A recently published study from Wharton-GBK (a collaboration between a US business school and an analytics consultancy) reported that “many enterprises are already seeing tangible benefits in productivity and performance” from their GenAI investments, with 74% already reporting positive returns on investment (ROI).

If current trends continue, these gaps could magnify, creating a sharper divide between empowered, AI-enabled employees and companies and those struggling to keep pace.

Wharton Human-AI Research and GBK Collective

How to reconcile this

It is implausible that the success rate from AI implementation could rise from 5% to nearly 75% over just three months. More likely, measuring the success of AI implementation is difficult and the results are very sensitive to precise methodological choices. As such, the 95% failure statistic from the MIT NANDA study can probably be taken with a pinch of salt. More interesting is seeing how the numbers from a repeated methodology change over time. This is a major benefit of the Wharton-GBK study, which is now in its third iteration.

What’s in the full note?

In the full note, we highlight the key results from a newer study disputing the 95% failure figure, which has become a critical part of the AI bubble debate.

Would you like to find out more? Click here* to read the full report. Please note, you must be a subscriber to HSBC Global Investment Research to access this link.

To find out more about HSBC Global Investment Research, including how to subscribe, please email us at AskResearch@hsbc.com

AI: Adopt & outperform

We used AI-powered sentiment analysis to analyse what company management is saying about key themes.

Need help?

For more information, please contact your HSBC representative.