People often think about tech bubbles in apocalyptic terms, but it doesn’t have to be as serious as all that. In economic terms, a bubble is a bet that turned out to be too big, leaving you with more supply than demand.  

The upshot: It’s not all or nothing, and even good bets can turn sour if you aren’t careful about how you make them. 

What makes the question of the AI bubble so tricky to answer, is mismatched timelines between the breakneck pace of AI software development and the slow crawl of constructing and powering a datacenter. 

Because these data centers take years to build, a lot will inevitably change between now and when they come online. The supply chain that powers AI services is so complex and fluid that it’s hard to have any clarity on how much supply we’ll need a few years from now. It isn’t simply a matter of how much people will be using AI in 2028, but how they’ll be using it, and whether we’ll have any breakthroughs in energy, semiconductor design or power transmission in the meantime. 

When a bet is this big, there are lots of ways it can go wrong – and AI bets are getting very big indeed.  

Last week, Reuters reported that an Oracle-linked data center campus in New Mexico has drawn as much as $18 billion in credit from a consortium of 20 banks. Oracle has already contracted $300 billion in cloud services to Open AI, and the companies have joined with Softbank to build $500 billion in total AI infrastructure as part of the “Stargate” project. Meta, not to be outdone, has pledged to spend $600 billion on infrastructure over the next three years. We’ve been tracking all the major commitments here — and the sheer volume has made it hard to keep up. 

At the same time, there is real uncertainty about how fast demand for AI services will grow.  

A McKinsey survey released last week looked at how top firms are employing AI tools. The results were mixed.  Almost all the businesses contacted are using AI in some way, yet few are using it on any real scale. AI has allowed companies to cost-cut in specific use cases, but it’s not making a dent on the overall business. In short, most companies are still in “wait and see” mode. If you’re counting on those companies to buy space in your data center, you may be waiting a long time. 

But even if AI demand is endless, these projects could run into more straightforward infrastructure problems. Last week, Satya Nadella surprised podcast listeners by saying he was more concerned with running out of data center space than running out of chips. (As he put it, “It’s not a supply issue of chips; it’s the fact that I don’t have warm shells to plug into.”) At the same time, whole data centers are sitting idle because they can’t handle the power demands of the latest generation of chips.  

While Nvidia and OpenAI have been moving forward as fast as they possibly can, the electrical grid and built environment are still moving at the same pace they always have. That leaves lots of opportunity for expensive bottlenecks, even if everything else goes right. 

We get deeper into the idea in this week’s Equity podcast, which you can listen to below.