People often think of tech bubbles in apocalyptic terms, but they don’t have to be that serious. In economic terms, a bubble is when the stakes are so high that supply exceeds demand.
The result: It’s not all or nothing. Even a good bet can go against you if you are not careful with your bets.
What makes the question about the AI bubble so difficult to answer is the schedule mismatch between the breakneck pace of AI software development and the slow pace of building and powering data centers.
These data centers take years to build, so a lot will inevitably change between now and when they go live. The supply chains supporting AI services are so complex and fluid that it is difficult to know how much supply will be needed in the coming years. It’s not simply a question of how much AI people will use in 2028, but rather how they use it and whether there will be any breakthroughs in energy, semiconductor design, or power transmission by then.
When the stakes are this high, there are many possibilities for failure. And the stakes with AI are indeed very high.
Last week, Reuters reported that an Oracle-affiliated data center campus in New Mexico had drawn up to $18 billion in loans from a consortium of 20 banks. Oracle already has a $300 billion contract with OpenAI for cloud services, and the two companies are working with SoftBank to build a total of $500 billion in AI infrastructure as part of the Stargate project. Not to be outdone, Meta has pledged to spend $600 billion on infrastructure over the next three years. We’ve been tracking all the major initiatives here, but the sheer volume is making it hard to keep up.
At the same time, there is real uncertainty about how fast demand for AI services will grow.
tech crunch event
san francisco
|
October 13-15, 2026
A McKinsey study released last week looked at how top companies are adopting AI tools. Results were mixed. While almost every company we contact uses some form of AI, very few are using it at any real scale. AI has enabled companies to reduce costs in specific use cases, but it has not impacted the business as a whole. The bottom line is that most companies are still in “wait-and-see” mode. If you’re counting on these companies to buy your data center space, you may be waiting a long time.
But even if the demand for AI were infinite, these projects could run into simpler infrastructure problems. Last week, Satya Nadella surprised his podcast listeners by saying he was more worried about a lack of data center space than a chip shortage. (As he says, “It’s not a chip supply issue; it’s the fact that there’s no warm shell to connect to.”) At the same time, entire data centers are sitting idle because they can’t keep up with the power demands of the latest generation of chips.
Nvidia and OpenAI are moving forward as fast as they can, but the power grid and the built environment are still moving at the same pace. This leaves many opportunities for expensive bottlenecks to occur, even if everything else goes well.
This week’s stocks podcast dives deeper into this idea. You can listen below.
