← The AI Systems Map
L0 · LabsStandard / protocolUSA

PyTorch

The default training/inference framework

PyTorch (governed by the PyTorch Foundation under the Linux Foundation, originated at Meta) is the substrate nearly every frontier model is built on, sitting above CUDA and below the labs. It is a genuine standards-layer dependency that binds labs to Nvidia via the CUDA backend, and its device-abstraction is the battleground for AMD/TPU portability.

How it fits the stack

PyTorch 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.

depends onhostsusesdepends onused byusesdepends onusescompetes withcompetes withcompetes withdesignsPyTorchLabsCUDA / software moatchokepointPyTorch Foundation(Linux Foundation)Triton (OpenAI)AnthropicHugging FaceOpenAIBaidu PaddlePaddleHuawei MindSporeJAX / XLAMeta AI (FAIR)
PyTorchDepends on ↑Feeds ↓Related

PyTorch in the AI stack. PyTorch 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) — 11 entities, 12 relationships
  • PyTorchdepends onCUDA / software moat
  • PyTorchhostsPyTorch Foundation (Linux Foundation)
  • PyTorchusesTriton (OpenAI)
  • PyTorchdepends onTriton (OpenAI)
  • PyTorchused byTriton (OpenAI)
  • AnthropicusesPyTorch
  • Hugging Facedepends onPyTorch
  • OpenAIusesPyTorch
  • PyTorchcompetes withBaidu PaddlePaddle
  • PyTorchcompetes withHuawei MindSpore
  • PyTorchcompetes withJAX / XLA
  • PyTorchdesignsMeta AI (FAIR)