I came across this chart from Benedict Evans on LinkedIn. (I must have been doing my quarterly check-in to see what the AI models are saying.) It shows US construction value broken down by offices, retail, warehouses (industrial), and data centers.
The office cyclicality is interesting to see. But of course, the point of this chart is the dramatic rise in data center construction. As of August 2025, it's about 2x retail construction and it's on track to overtake both offices and warehouses.
Will this happen and is this pace of construction sustainable? I frankly don't know. But AI and the need for more compute are obviously here to stay.
I'm reminded of something that venture capitalist Fred Wilson once wrote on his blog. His argument was that when it comes to speculative frenzies, or just new stuff in general, it's common to be directionally correct, but be off in terms of the order of magnitude.
I wouldn't be surprised if this proves to be true for AI-driven investments.