When your mother starts asking about AI stocks and SPACs are relaunching perhaps you should get a bit worried about an AI bubble. Yes, I have a few gray hairs and have lived through a few bubbles—dotcom circa 2000 and real estate circa 2008—and the aftermath isn’t great.

As a general rule, when you hear people standing in line at Wawa talking about AI, crypto and meme stocks you know you’re near a top. They don’t ring a bell and insanity can last longer than you think.

That said here are some mileposts to consider as the AI bubble inflates.

We hit the watercooler at Constellation Research to get opinions on the AI bubble. Here are a few thoughts from our analysts.

📌 Overall Context

The watercooler debate revolved around whether today’s AI boom—driven by generative AI, LLMs, and infrastructure spending—resembles a bubble. The contributors weigh in with different levels of skepticism and optimism, comparing current AI hype to past tech cycles like dot-com, crypto, and GPUs.


💡 Key Perspectives

Esteban Kolsky

  • Bubble signs:
    • Reports of high failure rates for GenAI projects and LLM oversaturation.
    • Many users are scrapping projects due to ROI as low as 1–5%.
    • Complaints about deteriorating performance in GPT-5, Claude, and Gemini—slower, more error-prone, and costly to run.
  • Hopeful note:
    • Unlike past cycles, VCs are cutting back earlier, which may limit damage.
    • Draws parallels to crypto and GPU cycles where optimization of cheaper tech separated winners from losers.

Larry Dignan

  • Moderate stance:
    • AI could “pause” for a few years and enterprises would still benefit if they focus on data readiness.
    • Warns that current valuations and constant Nvidia hardware spending are unsustainable, calling it a looming “shit show.”

Michael Ni

  • Not a blind bubble:
    • Acknowledges froth at the edges (overfunded, pre-revenue startups with 100x valuations).
    • But the core is strong: platforms like ServiceNow and Databricks are building durable AI-native systems for decision intelligence, governance, and execution.
    • Sees these as foundational enterprise value creators, not speculation.

Holger Mueller

  • Critical of execution:
    • Notes OpenAI’s failed launches (Project Ghibli, GPT-5 redirector issues) as “strike two” for enterprise trust.
    • Warns that if critical infrastructure can’t deliver reliably, enterprise adoption is at risk.
    • Enterprises are just starting with AI and the end of experimentation from them may pop the vendor bubble.
    • It's likely there will be a murky AI vendor field.

Martin Schneider

  • Bubble of expectations:
    • Says this isn’t a classic bubble yet, but rather a “bubble of inflated expectations” driven by the pace of innovation without sufficient end-user value.
    • Infrastructure providers will thrive (chips, compute), but application providers must deliver pragmatic, functional value or risk irrelevance.
    • Highlights Salesforce’s smaller acquisitions (Regrello, Bluebirds, Moonhub) as examples of moving toward practical AI applications instead of hype.

🧾 Takeaways on the “AI Bubble”

  1. Yes, bubble symptoms exist – low ROI, failed launches, hype-driven startups, inflated valuations.
  2. But not fully a bubble – enterprise platforms and infrastructure are building lasting value.
  3. Current state = bubble of expectations – more hype than results at the application layer, though infrastructure is solid.
  4. Future hinges on execution – pragmatic value delivery, cost control, and reliability will determine whether AI sustains or collapses in speculative segments.

👉 In short: The consensus is that AI has frothy edges but a grounded core. We may be in a contained bubble of expectations, not a full-blown collapse scenario.