vLLM
Open-source high-throughput LLM inference engine
vLLM (originating at UC Berkeley) became the dominant open-source inference serving engine via its PagedAttention memory technique, widely used by clouds and enterprises to cut inference cost. It is a key reason inference economics improved even as models grew. It anchors the open inference stack against proprietary serving.
How it fits the stack
vLLM with what it depends on (above) and what it feeds (below). The figure renders as a crawlable diagram and upgrades to an interactive 3D graph as it scrolls into view.
vLLM in the AI stack. vLLM with its immediate upstream dependencies (top) and downstream dependents (bottom) in the AI value chain. Hover a node in 3D, or read the full relationships below.
Graph data (text) — 9 entities, 9 relationships
- vLLM —depends on→ CUDA / software moat
- vLLM —uses→ Nvidia Data-Center GPU (Blackwell/Rubin)
- vLLM —depends on→ Nvidia Data-Center GPU (Blackwell/Rubin)
- AMD —used by→ vLLM
- Fireworks AI —uses→ vLLM
- Inference economics / token deflation —used by→ vLLM
- KV-cache & inference memory tiering —uses→ vLLM
- Together AI —uses→ vLLM
- vLLM —partners with→ Nvidia
Depends on ↑ · 3
Feeds ↓ · 5
Context — capital, rivals, policy · · 1