A better way to think about the AI ​​bubble

People often think of tech bubbles in apocalyptic terms, but they don’t have to be all that serious. In economic terms, a bubble is a bet that turns out to be too large, leaving you with more supply than demand.

Conclusion: It’s not all or nothing, and even good bets can turn bad if you’re not careful about how you make them.

What makes answering the AI ​​bubble question so difficult is the mismatch of timelines between the rapid pace of AI software development and the slow crawl of getting a data center up and running.

Since these data centers take years to build, a lot is bound to change between now and when they are up and running. The supply chain that supports AI services is so complex and fluid that it is difficult to get any clarity on how much supply we will need a few years from now. It’s not simply a question of how much people will use AI in 2028, but how they will use it, and whether we’ll have any breakthroughs in energy, semiconductor design, or power transmission in the meantime.

When the bet is this big, there are only so many ways things can go wrong – and the AI ​​bets are already very large.

Last week, Reuters reported that Oracle’s associated data center campus in New Mexico had attracted as much as $18 billion in credit from a consortium of 20 banks. Oracle has already partnered with Open AI for $300 billion in cloud services, and the two companies have joined Softbank to build a $500 billion overall AI infrastructure as part of the “Stargate” project. Meta, not to be outdone, has He pledged to spend $600 billion on infrastructure during the next three years. We’ve tracked all major commits here -And the sheer size made it difficult to keep up with.

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

The McKinsey poll was released last week I looked at how big companies are using AI tools. The results were mixed. Almost all of the companies contacted use AI in some way, but few do it on a real scale. AI has allowed companies to reduce costs in specific use cases, but it does not impact the business overall. In short, most companies are still in a “wait and see” mode. If you’re relying on those companies to buy space in your data center, you may have to wait a long time.

But even if the demand for AI is endless, these projects could face more obvious infrastructure problems. Last week, Satya Nadella surprised podcast listeners by saying he was more interested in the topic Data center running out of space Of running out of potato chips. (As he put it: “It’s not a chip supply issue; it’s the fact that I don’t have warm cases to plug them into.”) Meanwhile, entire data centers sit idle because they can’t handle the power demands of the latest generation of chips.

While Nvidia and OpenAI are moving forward as fast as they can, the electrical grid and built environment are still moving at the same pace as ever. This leaves plenty of opportunities for expensive bottlenecks to occur, even if everything else goes well.

We’ll dive deeper into the idea in this week’s Stocks Podcast, which you can listen to below.