Skip to content
DeployGraph

Reference Architecture

The AI-Native Stack

Every layer an AI application touches, in the order a user request flows through it. AI-native layers sit among traditional infrastructure. Vendors are ranked by how many of the 1636 companies we track use them.

329
AI-native vendors
2244
Traditional vendors
60%
Use a foundation model
3.8
Avg AI vendors per AI user
Developer Platform Communications CRM & Support Analytics & BI Billing & Payments Data Pipeline Databases Observability Evals & Testing AI Observability Image & Video Generation Voice & Speech AI Guardrails & Safety Browser Automation Code Execution AI Compute Foundation Models Model Serving Memory & State RAG & Retrieval Vector Databases Embeddings Agent Frameworks Auth & Identity Cloud & Networking

Layer Details

AIAI-native layerTraditional infrastructure

Request In

The request reaches the edge and the user is identified.

The Agent Loop

Orchestrate, assemble context, call a model, use tools, repeat.

Code Execution

AI-Native
2 vendors
Modal70E2B34

Output Processing

Check the response and turn it into speech, images or video.

Async & Cross-cutting

Trace, evaluate and monitor every step.

Supporting Infrastructure

Store data, bill customers, ship code and stay compliant.

Business Software

Tools the team uses to run the company. They do not sit in the request path.

See real stacks in production

Browse which AI companies use which combination of these layers.

Browse CompaniesCompare Stacks