Skip to main content
AI Market Watch
Loading...

Moreh

Category: AI Infrastructure

Korean AI infrastructure software company building full-stack inference software (MoAI) that unifies AMD, NVIDIA and Tenstorrent accelerators into high-performance LLM clusters. Moreh was founded in 2020. The company is led by Gangwon Jo (조강원). Based in Seoul, South Korea. Total funding raised: $68.4M. Latest round: Series C. Key investors include AMD, KT, Forest Partners, Forest Ventures, Smilegate Investment.

Founded
2020
Headquarters
Seoul, South Korea
Total funding
$68.4M

Value proposition

Run frontier LLMs at peak inference performance on non-NVIDIA and heterogeneous accelerators, maximizing tokens per dollar on existing hardware.

Products and solutions

MoAI Inference Framework, MoAI Performance Gateway, MoAI Fabric (cross-vendor KV cache transfer), Moreh vLLM for AMD and Tenstorrent, TIDE (Temporal Incremental Draft Engine, self-improving speculative decoding, unveiled at SC26 in Nov 2026, up to 1.66x throughput and 3.02x faster draft-model training), Token Factory OS (in development), Motif Technologies subsidiary for model development.

Unique value

Full-stack inference software from custom kernels to cluster-level scheduling across AMD, NVIDIA and Tenstorrent GPUs, including cross-vendor prefill/decode disaggregation.

Target customer

Cloud providers, telcos, data center operators and enterprises running large-scale LLM inference

Industries served

Cloud/telecom, AI data centers, enterprise AI, sovereign AI

Technology advantage

Custom kernels and libraries, distributed and disaggregated inference, cross-vendor KV cache fabric, and runtime draft-model training for speculative decoding. Reported benchmarks include 1.68x vs ROCm vLLM on DeepSeek R1 and 20,000+ tok/s per MI300X node.

How they differentiate

Vendor-agnostic software that unifies heterogeneous GPUs and optimizes non-NVIDIA hardware, instead of being locked to one vendor stack.

Main competitors

Lablup (Backend.AI), NVIDIA Run:ai, vLLM/SGLang open-source ecosystem

Key partnerships

AMD (strategic investor and partner), KT / KT Cloud (strategic investor and customer), Tenstorrent (strategic partnership since Nov 2024)

Notable customers

KT Cloud

Major milestones

2020.09 founded, 2021.12 AMD-GPU-based public AI cloud with KT Cloud, 2023.08 $22M Series B, 2024.01 MoMo-72B ranked #1 on Hugging Face Open LLM Leaderboard, 2024.11 Tenstorrent partnership, 2025.02 Motif Technologies subsidiary established, 2025.09-12 Series C raise (Forest Ventures), 2026.02 selected for Korea's sovereign AI foundation model project, 2026.11 TIDE unveiled at SC26.

Growth metrics

Cumulative revenue exceeded 40B KRW by Jan 2023; Vietnam entity established Nov 2023. Total funding of about $68.4M is reported by theailandscape.com as of Sept 2025; Dealroom shows a lower $30M.

Market positioning

Korean inference-software specialist focused on non-NVIDIA and heterogeneous accelerator clusters; one of four consortia in Korea's sovereign AI foundation model project (Feb 2026).

Geographic focus

Korea, US, Vietnam, global

Patents and IP

Korean patent KR 10-2625797, registered January 11, 2024, covers pipeline parallel processing compilation for machine-learning models. The patent owner is Moreh Co., Ltd.

About Gangwon Jo (조강원)

Ex-Manycoresoft CTO; Moreh co-founder, Chief Architect (2020-2021) then CEO since Oct 2021.

Latest news about Moreh

More AI Infrastructure companies

Official website: