This Piece Buries Its Most Important Paragraph

Anish Moonka's AI stack explainer is doing the rounds this week. Worth reading, with one significant caveat.

What it actually does that most AI explainers don't: it attaches real numbers to the infrastructure layer and refuses to let you skip past them. TSMC at 70% foundry market share. NVIDIA gross margins hovering around 75%. The four major cloud providers spending $450 billion on infrastructure in a single year. Most pieces gesture at "the picks and shovels companies are winning" — this one names the picks and shows you the margins. That's the difference between a framework and an argument.

But here's the problem.

The DeepSeek section is one paragraph long. A Chinese lab releases a model that approaches frontier performance at a fraction of the training cost, and the piece responds with "I don't think DeepSeek killed the thesis" before moving on to the video game metaphor.

That's not engagement — that's absorption. If efficient models keep closing the gap with frontier models, the entire infrastructure-spending logic that the piece is built on softens considerably. That's not a footnote. That's the load-bearing wall, and it gets less space than the Jensen Huang quotes.

The piece is investment commentary wearing the costume of an explainer. That's not a dealbreaker — but it means the conclusions ("farm Level 2 and Level 3") are presented as obvious when they depend entirely on the variable the author just glossed over.

Read it for the numbers on the infrastructure layers. Stop before you treat the strategy section as settled.