Two respected researchers are calling it plainly: the AI bubble is real, and it probably won't deflate gently.
It's not a fringe opinion anymore. In their 2026 trends analysis for MIT Sloan Management Review, a team of longtime AI researchers names the AI bubble as the single most-discussed issue in the field this year , eclipsing even the rise of agentic AI. The report's framing is worth quoting almost in full because of how direct it is:
Is there one? If so, when will it burst? Will the money rush out quickly or slowly? And what are the implications for the broader economy and the ongoing use of AI?
The report's answer is not comforting for anyone hoping this is overblown. Having lived through the dot-com crash, the researchers see unmistakable parallels: sky-high startup valuations, a focus on user growth over profitability, intense media hype, and enormously expensive infrastructure buildouts.
The researchers' conclusion: 'It seems inevitable to us that it will [burst], and probably soon.' The report notes it wouldn't take much to trigger it , a bad quarter from an important AI vendor, a cheaper foreign model matching Western performance (echoing the DeepSeek shock of January 2025), or a handful of large corporate customers pulling back AI spending.
This Doesn't Mean AI Was a Mistake
Gartner's own research adds a sobering data point that fi ts this pattern: more than 40% of agentic AI projects are projected to be cancelled by 2027, and one analysis suggests that of the thousands of vendors now marketing themselves as 'agentic AI' companies, only a small fraction , roughly 130 by one estimate , actually off er legitimate, diff erentiated agentic technology rather than repackaged automation. That gap between marketing claims and real capability is exactly the kind of froth that tends to get squeezed out in a correction.
Here's the distinction that matters for business leaders, and it's an important one: a bubble bursting in AI markets is not the same as AI itself failing to deliver value. The dot-com crash didn't mean the internet was a bad idea , it meant the money chasing it had outrun the actual, provable value being created, and the correction wiped out companies that never had a real business model to begin with. The same dynamic appears to be building in AI. Enterprise spending on AI reportedly reached around $37 billion in 2025, more than triple the 2024 figure , a pace of capital deployment that is difficult to sustain without commensurate, provable returns showing up on the other side.
Who Survives a Correction
History offers a reasonably reliable pattern here: bubbles clear out companies built on hype and capital access, and leave standing the companies that built something people were already paying for, based on measurable results. Applied to enterprise AI, that suggests the survivors of any 2026–2027 correction will be organizations and vendors who can point to concrete, attributable ROI , reduced claims cycle times, lower underwriting turnaround, measurable reductions in equipment downtime , rather than those whose case rests primarily on the promise of future capability.
This is precisely why the current market emphasis on proof over pilots, on governed and auditable AI, and on smaller, cost-efficient models rather than maximal general-purpose ones, matters so much right now. Those aren't just best practices , they're the characteristics of AI investment that tends to be bubble-resistant, because it's grounded in provable, day-to-day operational value rather than speculative future capability.
What Leaders Should Actually Do About This
The researchers' advice, echoed across much of the 2026 research, is not to panic or pull back from AI altogether , that risks ceding ground to competitors who use a downturn to double down intelligently. Instead, it's to stress-test AI investments the way a prudent CFO would stress-test any capital allocation: does this initiative have a clear, measurable business case independent of the broader AI narrative? Would we still fund it if AI hype cooled by 50% tomorrow? Is our vendor's viability dependent on continued speculative funding, or on actual paying customers and real unit economics?
The Bottom Line
The AI bubble conversation isn't a reason to slow down , it's a reason to get more disciplined. The organizations that treat 2026 as a moment to separate durable AI value from speculative hype, and invest accordingly, are the ones most likely to come out the other side of any correction stronger, not weaker.
References & Sources
• MIT Sloan Management Review. Davenport, T. H. and Bean, R. "Five Trends in AI and Data Science for 2026." sloanreview.mit.edu/article/five-trends-in-ai-and-data-science-for-2026
• Enterprise AI Agents Adoption Statistics compilation (Gartner, IBM, IDC sources), 2026.
• Afelyon Blog. "Gartner AI Predictions 2026: What Enterprise Leaders Need to Know." afelyon.com/blog/gartner-ai-predictions-2026-analysis