You're Watching the App. The Money Is in the Concrete.

That's the actual argument in Anish Moonka's AI stack explainer, and it's worth your time — with one exception you should know before you get there.

What makes it worth your time: the piece doesn't just assert that infrastructure companies are winning. It shows you the margins. NVIDIA at 75% gross margins. TSMC with 70% of the global foundry market while its nearest competitor sits at 7.2%. The four major cloud providers committing $450 billion in a single year — not to products, to buildings and cables and cooling systems. Most people making this argument wave at it. Moonka prices it out.

Here's where it breaks down.

Section VI contains one paragraph on DeepSeek — the Chinese lab that released a model approaching frontier performance at a fraction of the training cost. The piece acknowledges it, calls it "a variable," and moves on.

That's a problem. The entire infrastructure thesis rests on the assumption that more compute equals better AI, which justifies ever-larger data centre spend, which justifies the picks-and-shovels trade. DeepSeek challenged that assumption directly. If efficient models keep closing the gap, the capital concentration in Layers 1–3 looks less like a structural advantage and more like a bet that's still open.

Dismissing it in a paragraph and then spending three pages on a video game metaphor isn't analysis. It's avoidance.

Read it for the numbers. They're the best part. Just don't let the confidence of the conclusion outrun the work the piece actually did to earn it.