AI Hosting Platforms

Introduction

Where you host LLM inference matters: managed APIs (OpenAI, Anthropic), unified gateways (Vercel AI Gateway, OpenRouter), GPU hosts (Together, Groq, Fireworks), or your own GPUs.

Beginner analogy: Like choosing between AWS Lambda, Vercel, your own EC2, or a GPU box in your closet.

Understanding the topic

Core concepts to understand:

• Hosted APIs — OpenAI, Anthropic, Google. Easiest, most expensive.
• Gateways — Vercel AI Gateway, OpenRouter. Multi-provider, fallback, billing.
• GPU clouds — Together, Fireworks, Groq, Modal, Replicate.
• Self-hosted — vLLM / Ollama on your own GPUs.

Syntax reference

Visual workflow / architecture:

Easiest ──────────────────────────────────► Cheapest at scale
OpenAI ─► Vercel AI Gateway ─► Together / Groq ─► Self-hosted vLLM

Real-world use

Lovable AI uses a unified gateway under the hood, exposing OpenAI & Gemini models without keys. Cursor mixes hosted + Together AI for speed.

Best practices

• Start with hosted APIs; move to gateways for multi-provider.
• Self-host only at very high volume or strict data-residency.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Compare hosting options for LLMs.
• Q2. When self-host vs use OpenAI?
• Q3. What does an AI gateway add?