Layer L5
Training & Inference
Where compute gets operated: clouds, serving platforms, and the systems they run on.
GPU cloud & neocloud, inference serving, training-data services, cluster storage, server & rack OEMs
GPU cloud & neocloudinference servingtraining-data servicescluster storage systemsserver & rack OEMshyperscaler compute roles
Topics in this layer
GPU Cloud & NeoCloud
The compute tap: who rents GPUs by the hour — CoreWeave, Lambda, Crusoe, Together, Fireworks.
AI Data Stack
Labeling, synthetic data, vector DBs, and the RAG plumbing that feeds every model.
AI Servers & Rack-Scale Systems
The integrators turning GPUs, power & cooling into a deployable NVL72 rack: Supermicro, Dell, HPE.
Inference-as-a-Service
Token-economy platforms that serve models by the token — inference is now ~2/3 of accelerator demand.
Companies (27)
Ranked leaderboard →Ranked suppliers for Training & Inference
Companies grouped by what they build, with the latest verified signal per company.
View the L5 leaderboard →Engineering reference, not investment advice.