Moreh is an AI infrastructure software company headquartered in Seoul with a US presence in Santa Clara, founded in 2020 to challenge the dominance of NVIDIA's CUDA in AI computing. Its core idea is a hardware-abstraction layer that lets standard machine learning frameworks run efficiently on a wide range of accelerators, so organizations are not locked into a single chip vendor for training and inference.
The company's flagship product, MoAI, functions as a software platform comparable in spirit to CUDA but compatible with existing tools like Meta's PyTorch and Google's TensorFlow. Crucially, MoAI is designed to run AI models that previously could execute only on NVIDIA hardware on alternative silicon, most notably AMD GPUs. This positions Moreh as critical middleware for data centers, cloud providers, and national AI projects that want to diversify their compute supply or take advantage of more available, cost-effective accelerators.
To prove the capability of its stack, Moreh trained MoMo-70B, a large language model that reached the top of Hugging Face's Open LLM Leaderboard with a score above 77 at the time, demonstrating that competitive frontier-class models can be built without relying exclusively on NVIDIA infrastructure. The company has since broadened its message to 'inference software for every chip,' reflecting the surge in demand for efficient model serving across heterogeneous hardware.
Moreh raised a $22 million Series B in October 2023 from strategic and financial investors including AMD and Korean telecom giant KT, alongside Smilegate Investment and Forest Partners, bringing total funding to around $30 million. The participation of AMD is especially notable given Moreh's role in making AMD accelerators viable for mainstream AI workloads. With this backing, Moreh is positioning itself as a key enabler of a more open, multi-vendor AI hardware ecosystem.
