Moreh

Founded in 2020, Moreh is a South Korean AI infrastructure company building a full-stack software platform for training and inference of large AI models across heterogeneous hardware.

Its core platform includes AI compilers, GPU/NPU libraries, distributed runtime systems, PyTorch- and TensorFlow-compatible frameworks, and the MoAI Inference Framework, enabling organizations to run frontier LLMs efficiently on AMD GPUs, NVIDIA GPUs, and Tenstorrent accelerators without being locked into a single hardware vendor. The company focuses on maximizing performance, reducing inference costs, and improving hardware utilization for enterprise AI deployments.

Moreh has established itself as one of South Korea’s leading AI infrastructure startups through close collaborations with major semiconductor and AI ecosystem partners, including AMD, Tenstorrent, SGLang, SkyPilot, and the ROCm open-source ecosystem.

Its inference software delivers measurable performance gains, including up to 1.68× faster inference than ROCm vLLM, over 20,000 tokens per second per node on AMD MI300X clusters, support for cross-vendor GPU clusters, and up to 2.2× higher throughput while using 40% fewer servers through intelligent routing and scheduling.