Everyone knows about Nvidia (NVDA) and Microsoft (MSFT), but here’s the part most people miss: AI runs on giant data centers, and building those is a whole economy in itself.

‍

‍

Think of a data center as a factory for computing. Each one costs about $39 million per megawatt to build, and most of that money doesn’t go to chips — it goes to the “boring” but essential parts that keep the machines running:

‍

• Servers (the brains): Dell (DELL), HP Enterprise (HPE), Super Micro (SMCI).
• Networking (the nerves): Arista (ANET), Cisco (CSCO).
• Power backup (the generators): Caterpillar (CAT), Cummins (CMI).
• Cooling: Vertiv (VRT), Johnson Controls (JCI).
• Electrical systems (the wiring): Eaton (ETN), Schneider Electric.
• Builders (the construction crews): Jacobs (J), Holder.

‍

Here’s the “wow factor”: the air conditioning alone can cost hundreds of thousands per megawatt, because if those servers overheat, the whole system fails. Generators and wiring eat up millions more. In other words, the AI trade isn’t just about chips — it’s about keeping the lights on and the servers cool.
‍

Takeaway: With NVDA and ORCL at all-time highs, the smarter angle may be the “picks and shovels” — the companies building and maintaining the data centers themselves. This is the less obvious, but just as critical, backbone of the AI boom.

‍