Analysis
Infrastructure
The AI Infrastructure Landscape: A Critical Map of Who Actually Matters
Forget the model leaderboards. The companies building the picks and shovels of AI — silicon, networking, power, hyperscale, and the inference layer on top — are where the durable economics live. Here is the unsentimental map.
By The Memo · Friday, June 19, 2026 · 13 min read
Every guide to "AI infrastructure companies" you've read in the last two years has been a list. Nvidia, Microsoft, Amazon, Google, a few names from the networking layer, a token mention of CoreWeave, and a closing paragraph about how interesting it all is. That is not a map. That is a memorial wall.
A real map of the AI infrastructure landscape has to do three things. It has to name the layer each company actually operates in. It has to be honest about which layers have durable economics and which are commodities pretending otherwise. And it has to say plainly which incumbents are about to be disintermediated by the layer above them. This is that map.
The AI infrastructure stack has six layers worth caring about: silicon, systems and networking, power and real estate, hyperscale capacity, the neocloud / GPU-as-a-service tier, and the inference and model-serving layer on top. Each one has a different competitive structure, a different margin profile, and a different set of names that actually matter. Treating it as one undifferentiated "infrastructure" market is how investors lose money and how operators pick the wrong vendor.
Subscribers only
Keep reading — it's free.
The Model Memo is a free daily newsletter. Drop your email to unlock the rest of this essay and get tomorrow's in your inbox. Always free, unsubscribe anytime.
Free daily newsletter. Unsubscribe anytime. No spam — ever.
Keep Reading
