The stack
The AI vendor stack, drawn from the capability taxonomy: each layer is a capability category, and each layer lists only the companies tracked in our sources. Layers are classification buckets, not a market map — coverage is never complete, and no vendor is placed by hand.
Rentable GPU-backed virtual machines or clusters sold as cloud infrastructure.
Companies selling monitoring, logging, tracing, and error tracking.
Hosted model APIs returning completions, embeddings, or generations.
Global-scale multi-service cloud platforms (AWS, Azure, Google Cloud class).
Registries, fine-tuning hosting, or deployment platforms for customer-supplied model weights.
Per-second or per-request billed GPU execution without server management.
Isolated execution environments for AI agents (browsers, code runners, VMs).
OLAP databases and cloud data warehouses.
Content delivery, edge networks, DDoS protection, DNS, private connectivity.
12 lighter layers are listed on the categories page. Coverage is never complete.