Understanding AI Providers
Not every AI provider plays the same role — some are model labs, some are cloud platforms, and some are infrastructure vendors.
Overview
It’s tempting to lump OpenAI, AWS, and NVIDIA together as "AI companies," but they play genuinely different roles in an architecture — one publishes models, one hosts infrastructure and offers managed access to several providers, and one supplies the hardware underneath almost all of it.
Where It Fits
Model Labs
OpenAI, Anthropic, Mistral AI, …
Cloud Platforms
AWS, Azure, Google Cloud.
Infrastructure Vendors
NVIDIA.
Key Points
- Model labs
- Companies whose primary AI business is training and publishing models — OpenAI, Anthropic, Mistral AI, Cohere, AI21 Labs.
- Cloud platforms
- Companies that host infrastructure and often provide managed access to multiple model providers — AWS, Microsoft (Azure), Google.
- Blended roles
- Some companies span categories — Google is both a model lab (Gemini) and a cloud platform (Google Cloud); Amazon is a cloud platform that also publishes its own Nova models.
Interview Question
Why does it matter whether a company is a "model provider" versus a "cloud platform" in an AI architecture discussion?
It changes what you’re actually evaluating them on — a model lab is judged on model quality and capability, while a cloud platform is judged on infrastructure, compliance, and how many model providers it gives you managed access to. Conflating the two makes it harder to reason clearly about vendor selection and lock-in.
Explain It in 30 Seconds
AI providers split into a few distinct roles — model labs that train and publish models, cloud platforms that host infrastructure and often provide managed access to several providers, and infrastructure vendors supplying the hardware underneath — and some companies span more than one role.
Real-World Stack
Technologies commonly used to implement this in production.