Interactive Editorial

The AI Shakeout

What 25,000 Companies Tell Us About Who Survives

Joel MascarenhasMarch 2026Interactive
Act I

The Pattern Nobody Talks About

There are roughly 25,000 AI companies in the world right now. In ten years, there will be maybe 50 that matter.

This isn't pessimism. It's the math. Every technology market in history follows the same consolidation curve — from thousands of hopeful entrants to a handful of survivors.

0+
AI companies today
~50
will matter in 10 years
0%
failure rate
OPENINGSCALEFOCUSBALANCEHIGHLOWMARKET PLAYERSTIME →
Vulnerable
Contested
Positioned to survive
Act II

The Graveyard

Out of every 100 AI companies founded today, history says 95 will fail. Not because they're bad — because they're building on rented land.

95 dead5 survive

The companies that die won't die because their models aren't good enough. They'll die because they don't own anything underneath.

Act III

Follow The Money

OpenAI raised $13B to build AI. NVIDIA made $60B+ in profit selling them the tools. The model builders get the headlines. The supply chain gets the money.

MONEY FLOWS DOWN — MARGINS INCREASE ↓
Application Layer
OpenAI / Anthropic
~15-20%

Build the model, charge subscriptions. But they rent everything underneath.

Cloud / Compute
AWS / Azure / GCP
~25-35%

Rent out GPU clusters. Take a margin on every training run.

Chip Design
NVIDIA
~60-75%

Design GPUs and CUDA ecosystem. Massive moat through developer lock-in.

Fabrication
TSMC
~50-55%

Only company that can fabricate at cutting edge. Single point of failure.

Energy
Utilities / Grid
Regulated

Power the data centers. Increasingly the bottleneck as AI scales.

Act IV

The Moat Test

Five things determine whether an AI company survives: data ownership, compute control, distribution, feedback loops, and irreplaceable talent. Compare the major players.

Data3/3
Compute3/3
Distribution3/3
Feedback Loop3/3
Talent & IP3/3
Total: 15/15
Act V

The Intimacy Thesis

There's a moat nobody talks about: intimacy. You've already split your AI brain into compartments — and the company that holds your deepest layer has the strongest lock-in.

DEEPEST

The 2 AM confessional. Unfiltered thoughts. The real moat.

“The AI that knows your work can be replaced. The AI that knows your secrets can't.”

Act VI

What Companies Get Wrong vs. What Works

WHAT THEY THINK
Best model wins
WHAT WORKS
Best infrastructure wins
WHAT THEY THINK
Raise more capital
WHAT WORKS
Own more of the stack
WHAT THEY THINK
Move fast, ship features
WHAT WORKS
Build feedback loops
WHAT THEY THINK
Partner with cloud providers
WHAT WORKS
Become the cloud provider
Act VII

Go Deeper

This editorial is the overview. Each thesis has a full analysis behind it.

DEEP DIVE
The AI Company That Wins Won't Be The Smartest
How consolidation has played out in every previous tech era — and what it means for AI.
DEEP DIVE
Stop Comparing AI Models. Start Asking Who Owns The Stack.
First-principles analysis of the AI value chain — where the money actually flows.
DEEP DIVE
Why Your AI Transformation Will Fail
What enterprises get wrong about AI adoption — and the operational playbook that works.
DEEP DIVE
How to Build an AI Department Inside a Traditional Company
The practical guide to standing up AI capability from scratch.

Positions reflect one analyst's framework.
Not investment advice.

← Back to Writing