Many people view technology bubbles as catastrophic, but the reality is often less dramatic. Economically, a bubble simply means an investment that was too ambitious, resulting in an oversupply compared to actual demand.
The takeaway: Outcomes aren’t always black and white, and even well-considered investments can go awry if not managed carefully.
What makes the debate around an AI bubble so complex is the disconnect between the rapid evolution of AI software and the much slower process of building and powering data centers.
Since constructing these data centers takes several years, significant changes are bound to happen before they become operational. The intricate and ever-changing supply chain behind AI services makes it nearly impossible to predict future supply needs with certainty. It’s not just about how many people will use AI in 2028, but also about the ways they’ll use it and whether there will be advancements in energy, chip technology, or power infrastructure by then.
With investments of this magnitude, there are countless ways things could go off track — and the stakes in AI are only getting higher.
Just last week, Reuters revealed that a data center campus in New Mexico associated with Oracle has secured up to $18 billion in financing from a group of 20 banks. Oracle has already committed $300 billion in cloud services to OpenAI, and together with SoftBank, they’re working on the “Stargate” initiative, aiming for $500 billion in AI infrastructure. Not to be left behind, Meta has promised to invest $600 billion in infrastructure over the next three years. We’ve been monitoring all these major investments here — and the sheer scale makes it difficult to keep pace.
Meanwhile, there’s genuine uncertainty about how quickly demand for AI services will actually rise.
A McKinsey survey published last week examined how leading companies are adopting AI tools. The findings were varied. Nearly every business surveyed is using AI in some capacity, but only a few have implemented it on a large scale. While AI has helped companies reduce costs in certain areas, it hasn’t significantly impacted their overall operations. In essence, most organizations are still taking a cautious approach. If you’re relying on these companies to fill your data center, you might be waiting quite a while.
Even if demand for AI never slows, these massive projects could still face basic infrastructure challenges. Last week, Satya Nadella surprised podcast listeners by admitting he worries more about running out of data center capacity than a shortage of chips. (In his words, “It’s not a supply issue of chips; it’s the fact that I don’t have warm shells to plug into.”) Simultaneously, some data centers remain unused because they can’t meet the power requirements of the newest chips.
While Nvidia and OpenAI are pushing forward at breakneck speed, the power grid and physical infrastructure are progressing at their usual, slower pace. This mismatch creates plenty of opportunities for costly delays, even if everything else goes smoothly.
We explore this topic further in this week’s Equity podcast, which you can listen to below.
