🖥️ Everyone's arguing about whether AI data centers are the next dot-com bust. The history people cite to make the case is mostly wrong.

Take the mainframe comparison. The story goes that PCs made mainframes obsolete, so on-device AI chips will do the same thing to data centers. Except PCs never killed the mainframe. 71% of Fortune 500 companies still run them, they process 90% of credit card transactions, and IBM shipped a new AI-focused model in 2025. The market is still growing. Old compute and new compute have coexisted for 50 years.

The fiber comparison is half right. Telecoms really did overbuild in the late 90s: only 2-5% of buried fiber was lit in 2001-02. But that dark fiber wasn't wasted. It became the backbone of the modern internet within a decade. Investors who overpaid got wiped out. The infrastructure didn't.

And the "data centers will sit empty" prediction has a problem: vacancy is at record lows right now (0.3% in Northern Virginia), 80% of space under construction is already preleased, and 2026 hyperscaler capex is around $725B, up 77% in a year.

None of this means there's no bubble. Spending $725B a year concentrated in a handful of companies is genuinely unprecedented. But the crash case deserves better evidence than a history that never happened.

Read more:
ibm.com/think/news/world-runs-…
cbre.com/press-releases/global…
usatoday30.usatoday.com/money/…

#AI #datacenters #investing