What this atlas argues

Fable 5 · data curated through July 2026

A directory has no opinion. This one has 6, and each names the companies, layers and people it rests on so the claim can be checked rather than believed. Quotable under CC BY 4.0.

Layers are value capture, not operations

Companies sit where they capture value, not where they operate. That single rule is what makes the cake a map instead of a directory.

What it rests on.

The capital pyramid — picks and shovels win

Cloud pays Silicon roughly thirty times what Tooling pays Models ($85B vs $3B a year). Whatever the apps story is, the money story is chips. The bottom of the cake eats first.

What it rests on.

The OpenAI diaspora

The same few dozen humans keep steering the industry — and an outsized share of them passed through one door. OpenAI alumni founded or now run Anthropic, SSI, Thinking Machines, Perplexity, and xAI. Track the people, not the org charts.

What it rests on.

Vertical integrators are the ones to watch

A company spanning two or more layers by value capture is playing a different game — margin at every floor it occupies. The integrator count grows every era; that growth is the industry consolidating.

What it rests on.

Physical AI is a dimension, not a category

Humanoids, autonomous driving, embodied foundation models, and world models are grouped by form factor because the physics actually matters. Embodiment got its GPT-3 moment with pi-zero; the cake grows hands.

What it rests on.

China's parallel stack

Under export controls, China is not catching up inside the Western stack — it is building a parallel one, layer by layer: SMIC and Huawei below, DeepSeek, Qwen, Zhipu, and Moonshot above. DeepSeek R1 was the day the parallel stack stopped being theoretical.

What it rests on.

The Five Layer AI Cake is published by ConeApex, the trading name of Nirukta Digital XR Technologies Private Limited (CIN U62090TS2025PTC198275), Hyderabad, Telangana, India. The dataset is licensed CC BY 4.0 — cite it freely. See llms.txt for how to attribute it.