Fact or Fiction – AI is a Bubble About to Burst

They say the four most dangerous words in investing are “It’s different this time”. And I’ve read some other opinions that make that case. Is it ‘different this time’? It feels different but given the CapEx delivered and proposed it could well be a bubble that will pop spectacularly at some point.

But that’s the big question – when? I think there’s enough in the pipeline to keep things going for 12 – 18 months. What I worry about is another DeepSeek moment as from January of this year – some breakthrough that shifts everyone’s thinking about how to make progress through the scale of data centres and city sized power plants; some thing that will come out of the blue and seem obvious in hindsight.

Richard Sutton spoke about LLM’s not being the path and he is working actively to convey the importance of The Era of Experience and his Alberta Plan. A breakthrough by him and Carmack could be the bubble popper. So could myriad other initiatives out there (Sakana AI Darwin Godel/Evolutionary AI, Embodied World Models, Mixture of Experts (DeepSeek again prominent), and many other problems that we don’t even know about yet.

So yes, we should be concerned about the AI Bubble but if we zoom out and look at where things are heading, the trajectory is up and to the right. With disruption and turmoil in between.  “In the case of the AI boom, the process underlying the trend is a lot fuzzier and it’s hard to attribute the improvements to any single phenomenon. It seems to be closer to something like Moore’s law, where the whole industry focuses on finding new innovations, and new ways of scaling and stacking the S-curves. But this means that the improvements here are not inevitable. Had we not had the reasoning model breakthroughs, the AI capabilities would have definitely hit a wall.” Quoting Atharva Raykar from the article https://www.julian.ac/blog/2025/09/27/failing-to-understand-the-exponential-again/

Size matters HERE

Demand excessive HERE

Electricity demand to fuel AI is outstripping the capacity to supply with broad social, economic, and environmental consequences. The hype around artificial intelligence often celebrates its promise to transform economies and professions. But behind the headlines lies an inconvenient truth: AI’s “compute warehouses” are set to consume electricity on a scale rivaling entire nations. Could the race to build ever-larger data centers destabilize global energy grids, raise costs, and strain water and environmental systems? Without sober attention to energy limits, the AI boom may become a threat to infrastructure, governance, and sustainability itself.