← The AI Systems Map
L2 · ChipsStandard / protocol⬦ Chokepoint

MLIR / OpenXLA / StableHLO

Compiler IR layer decoupling models from accelerators

MLIR and the OpenXLA stack (XLA, StableHLO portability layer, IREE runtime) are the intermediate-representation substrate that lets JAX and TensorFlow target TPUs, GPUs and custom ASICs. StableHLO is the emerging portable op-set that most non-CUDA accelerators (TPU, Trainium, Groq, Tenstorrent) compile through, making it a quiet but critical neutrality layer in the stack.

Governance

OpenXLA (Google-led, multi-vendor)

Portable op-set

StableHLO

How it fits the stack

MLIR / OpenXLA / StableHLO 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 ondepends ondepends onMLIR / OpenXLA /StableHLOChipsAWS Trainium /InferentiaGoogle TPUJAX
MLIR / OpenXLA / StableHLOFeeds ↓

MLIR / OpenXLA / StableHLO in the AI stack. MLIR / OpenXLA / StableHLO 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) — 4 entities, 3 relationships
  • AWS Trainium / Inferentia —depends on→ MLIR / OpenXLA / StableHLO
  • Google TPU —depends on→ MLIR / OpenXLA / StableHLO
  • JAX —depends on→ MLIR / OpenXLA / StableHLO
← Back to the AI Systems Map