ZML · The clearest live test of whether inference is really portable — one server, one model, five accelerators, and published numbers for each, which is the only way to price a task instead of a token.

One inference server that runs the same open-weight LLM on Nvidia, AMD, Google TPU, Apple Metal or Intel, with no Python anywhere in the path. The stack is written in Zig on MLIR/OpenXLA and compiles the model ahead of time into a single self-contained binary, so the thing you ship is the thing that runs. The free product is LLMD — a self-contained OpenAI-compatible LLM server shipped as a Docker image per platform (and a Homebrew formula on Apple Silicon).

Sector

Inference

Stage

Series A

On the roster since

15 Sept 2026