Skip to content

Stack map

The whole AI chain,
as one map you can use.

All 80 parts, from power to apps, with every link between them. Hover a part to see what it leans on. Push one and watch the knock-on effects travel.

See pricing

Included in Pro

Map80 parts · 5 layers
Find a nodeRelax layoutHeat BottleneckScenario modeTrace path
LayersEnergyChipsInfrastructureModelsAppsHorizonNowBuildoutStrategicDomainAll domains

Chips

HBM89

Memory stacked beside the GPU. Every AI accelerator needs it, and few companies can make it.

Open node →

Relies on · upstream

  • depends on DRAMMedium lag · Threshold · strength 78
  • depends on Test & burn-in capacityShort lag · Threshold · strength 62
  • depends on Foundry capacityLong lag · Linear · strength 60

Drives · downstream

  • bottlenecks GPUMedium lag · Threshold · strength 86
  • constrains DRAMMedium lag · Linear · strength 56
Energy86Transformer availability · Grid capacity
Chips81HBM · CoWoS / advanced packaging
Infrastructure63Scale-up fabric
Models44Inference serving
Apps28Agent workflow
80 parts

of the AI chain on the map, each with a plain-English reason it matters

5 layers

Energy, Chips, Infrastructure, Models and Applications, read from power to product

170 links

between parts, each with a strength, a lag and how the effect spreads

How it works

A map that knows how the parts lean on each other.

The map is more than a picture. Four things make it a working model of the chain.

  1. Step 1

    Eighty parts, in five layers

    Every part of the AI build-out sits in its layer, from grid power at the base to the apps people use on top.

  2. Step 2

    Every link says how hard it leans

    Each link carries a type, a strength, how long the effect takes and whether it spreads evenly or past a threshold.

  3. Step 3

    The binding set lights up

    Switch the heat to Bottleneck and the few parts that cap the whole chain stand out.

  4. Step 4

    Push a part, watch it travel

    Scenario mode sends a shock from any part. It arrives hop by hop, biggest effects first.

What you can do

One map. Four ways to read it.

Hover to learn, push to test, trace to connect, filter to focus.

Run a shock

See what a squeeze in one part does

Turn on Scenario mode, pick a part, and the map shows every part downstream that feels it, hop by hop.

Shock any part

Transformers, HBM, grid power: pick where it starts.

See it spread

The shock arrives one hop at a time, not all at once.

Read who is hit

Every part reached, listed by hop.

MapShock from: Transformer availability
Find a nodeRelax layoutHeat BottleneckScenario mode: onTrace path

The shock, hop by hop

Hop 1Grid capacity, Substation capacity, Power delivery chain, Data-center construction, Deployment lead time, AI buildout risk
Hop 2AI factory, Power density per rack
Hop 3Cost per token
Hop 4Agent workflow, Copilot

Before and after

Same question. Far less legwork.

Understanding how a squeeze in one place reaches everything else usually means drawing it yourself.

Doing it by hand
Read industry reportsLayer by layer
Sketch the linksOn paper
Guess the weak spotsFrom the news
Trace knock-on effectsIn your head
Redraw itWhen things move

To keep in your head

170links

between 80 parts, each with its own strength

With THE ENTITY
Open the mapEvery part, by layer
Hover a partWhat it leans on
Push itSee it travel
Read who is hitHop by hop

To read

1map

with every link drawn and weighed for you

The map does the tracing. The judgement stays yours.

A worked example

A transformer shortage, followed up the stack.

The part names and the path are the model’s own. The impact values are illustrative.

  1. The shock

    Transformers get scarce

    You turn on Scenario mode and shock Transformer availability, a part at the base of the energy layer.

  2. Hop 1

    Power and construction feel it first

    Grid capacity, substations, the power delivery chain and data-center construction are hit straight away.

  3. Hops 2 and 3

    It climbs into compute

    AI factories and power density per rack follow, then the cost of each token the models serve.

  4. Hop 4

    It reaches the apps

    Agent workflows and copilots sit at the end of the line. A grid problem has become an app cost problem.

Real parts and links from the model. Impact values are illustrative.

Is it a fit?

Built for people who want to see how it fits, not memorise a list.

Not built for

  • Picking stocks on its own. It maps parts. Companies are one click away on each part.
  • Minute by minute moves. Scores refresh every night.
  • Supply chains outside AI. It covers the AI build-out only.

Built for

  • Anyone who wants to understand how the AI build-out actually works
  • Investors who want to know what a headline in one place does elsewhere
  • People who think better with a picture than a table
  • Readers who like each link explained in plain words

Questions

Good questions, plain answers.

The short version of how Stack map works. The full method is public.

Read the methodology
Do I need a finance or engineering background?

No. Hover any part and it explains itself in plain words, with what it relies on and what it drives.

What do the links mean?

Each link is a relation between two parts, such as depends on, enables or bottlenecks. It carries a strength, a lag and how the effect spreads.

What is the binding set?

The few parts that cap the whole chain right now. Switch the heat to Bottleneck to see them, or open Bottlenecks for the full ranking.

Is the map live?

The scores behind the heat refresh every night, the same scores used across the rest of the app.

Which plan includes the Stack map?

Pro. See pricing for what each plan includes.

Is this investment advice?

No. It is research: a model of how the chain fits together, so you can form your own view.

Get started

See how the whole chain moves, not just one part of it.

Open the map, hover a part, and push it to see what follows.

See pricing