The stack map
Every startup on the roster, placed on the layer of AI it works on. Read it top to bottom: the product a person uses sits on agents, agents sit on models, models run on inference, and all of it runs on rented GPUs, chips and power. A dashed box is a part of the stack with no roster company in it yet.
01 · 8 companies
Applications
What a person or a team actually uses — the finished product, in software or in the world.
- AI at work4
- Smart home1
- Robotics0
None on the roster yet.
- Autonomous driving0
None on the roster yet.
- AI-run labs3
02 · 22 companies
Agents
The building blocks every agent product needs, and the open problems of keeping many agents in step and cheap to run.
03 · 13 companies
Trust & control
Checking what an agent does before it ships, defending it from text written to trick it, what it is allowed to do once live, where its output came from and why it answered as it did.
04 · 12 companies
Models
The models themselves, and the tools that train and tune them.
05 · 14 companies
Inference
Turning a model into answers — in the cloud, or on a machine you own.
06 · 4 companies
Compute
The GPUs, rented by the hour or scheduled across a cluster, and the markets that price them.
07 · 2 companies
Silicon
The chips the work runs on.
08 · 0 companies
Power
The electricity under all of it.
- Energy0
None on the roster yet.
- Energy0