Field Atlas · The AI Sector · Interactive edition

The Stack,
bottom to top.

Seven layers from sand to software. Intelligence is manufactured at the bottom and refined upward; capital flows the other way. Where a name sits determines what it can charge, what it can defend, and who can eat it.

Explore the atlas Open the operator's bench

Below the map, the Operator's Bench lets you run the numbers yourself: what your own AI usage costs, how to adapt a model, and a concepts deck for the machinery underneath.

Frozen · 27 July 2026

The map stops here.

The week it was frozen, the people building these systems could not agree on what was happening.

Thesis “We’re now, like, in the singularity.” Sam Altman · OpenAI · Relentless podcast, reported 27 Jul 2026
Thesis We are in the “foothills of the singularity.” Demis Hassabis · Google DeepMind · reported 27 Jul 2026
Thesis “Coding is largely solved.” Boris Cherny · Anthropic, creator of Claude Code · Lenny’s Podcast, 19 Feb 2026
Counter “We aren’t there yet.” Seán Ó hÉigeartaigh · Centre for the Study of Existential Risk, Cambridge · reported 27 Jul 2026

Built July 2026 as one of my first Claude Code projects, and frozen here on purpose. The live trackers linked below carry the current numbers; updates from here are occasional and manual.

Pick a path: scan the layers, filter the market through a lens, trace one real dollar, or test what stuck. Every company tile opens a plain-English profile.

Find a company

View through

Trace one dollar

Bench

01 / Atlas

Read the market as a system.

Move from applications down to silicon. Every layer has its own economics, choke point, and way of getting squeezed.

↑ Intelligence & margin compound upward ↓ Capital & compute pour downward Plain text = public fact · reported = unconfirmed / in talks · ○ = thesis, an argument not a measurement

Part II · The Operator's Bench

Now run the numbers yourself.

The map shows where the money sits; the bench shows why. Three instruments — drag, choose, flip.

II · 01

Your AI Bill — what your own usage actually costs

Pick a model tier and a task you actually do. See what it costs per query on the raw API — then check whether your monthly subscription is a good deal at your real volume, or whether you're paying for headroom you never touch.

Pick a task you actually do

Model tier (sets $/M tokens ○)

Run open models on your own hardware?

—

cost / query (API)

—

est. cost / mo (API)

—

vs your subscription

II · 02

Adapt or Build — which lever do you pull?

Four ways to make a rented model yours. Answer four questions; the bench recommends a stack and names the failure mode you just signed up for.

II · 03

The Concepts Deck — the machinery, thirty seconds a card

One term per card: the plain-English version, the one thing to remember, and a sharp question to ask. Say your version out loud before you flip.

Capital & M&A — recent flow

Reading the map · ○ these are theses — arguments, not measurements

Where the live numbers live · this page is a dated snapshot — these update in real time