What Moreh does

Moreh is an AI infrastructure software company founded in September 2020 in Seoul, South Korea, that builds a full-stack inference platform — the MoAI framework — optimized for heterogeneous GPU clusters spanning AMD Instinct, NVIDIA, and Tenstorrent accelerators. The core product line includes the MoAI Inference Framework (cluster-scale routing, scheduling, auto-scaling, KV cache management), Moreh vLLM drop-in replacements for AMD and Tenstorrent GPUs, MoAI Performance Gateway for cross-vendor workload distribution, MoAI Fabric for RDMA-based KV cache transfer across vendors, and proprietary primitives including HetCCL (cross-vendor collective communication), SLOPE (long-context prefill engine), and TIDE (runtime draft model trainer). The company reports differentiated capabilities such as cross-vendor prefill-decode disaggregation between NVIDIA H100 and AMD MI300X, with published benchmarks claiming up to 1.84x lower latency and 67% higher throughput versus single-vendor baselines.

The company monetizes via enterprise software licensing (subscription, quote-based, multi-year) combined with professional and managed services that bundle hardware procurement, cluster design and deployment, and ongoing optimization for AMD and Tenstorrent clusters; pricing is not publicly disclosed. Distribution runs through direct enterprise sales to AI data centers, cloud service providers, and telecom operators, augmented by strategic channel partnerships with AMD and Tenstorrent for joint go-to-market and open-source contributions to ROCm, SGLang, SkyPilot, and Tenstorrent Metalium. Key customers include KT Cloud (Hyperscale AI Computing service running 1,200+ AMD MI250 GPUs) and the Korea Ministry of Science and ICT (lead consortium for the national sovereign AI foundation model project); subsidiary Motif Technologies develops proprietary Korean-language LLMs. The company is backed by AMD, KT, and Forest Ventures, with cumulative funding exceeding $50 million USD and a post-money valuation of approximately 350 billion KRW (~$250M USD) as of September 2025.

Name

Moreh

Legal name

Moreh, Inc.

Website

https://moreh.io

Company type

Private

Founded year

2020

Operating status

Operating

Headcount range

51–100 employees

Short description

Moreh is a Seoul-based AI infrastructure software company building the MoAI full-stack inference platform that orchestrates heterogeneous GPU clusters (AMD, NVIDIA, Tenstorrent), serving AI data centers, cloud providers, telecom operators, and government AI initiatives with cross-vendor disaggregation and custom inference engines.

Ownership category

akta.pro rank

Product category

AI Inference Software

NAICS

Computer Systems Design and Related Services (5415), Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (518), Custom Computer Programming Services (541511)

SIC

Services-Prepackaged Software (7372), Services-Computer Integrated Systems Design (7373), Services-Computer Programming, Data Processing, Etc. (7370)

akta.pro primary industry

Model Serving, Inference & Deployment Platforms (APIs, Edge/On-Prem) (HDAEANAC)

akta.pro secondary industries

AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers) (HDAAAAAI), AI Compute Virtualization & Scheduling (GPU virtualization, cluster schedulers) (HDAAAAAG), Model Deployment, Serving & Inference Platforms (HDAAABAF), Model Hosting, Serving & Inference Platforms (HDAAACAB)

Keywords

• AI inference software
• GPU cluster management
• LLM serving infrastructure
• Heterogeneous computing platforms
• GPU virtualization software

Headquarters

HQ city

Seoul

HQ country

South Korea

HQ region

Asia

Offices3 records

Markets served

GTM type

B2B

Offering type

Software

Cost components

Personnel, Technology or R&D, Operations, Marketing or Sales, Infrastructure

Revenue model

1. AI Infrastructure Software Licensing: Software licensing for AI inference infrastructure, including MoAI Inference Framework and Moreh vLLM. Provided as container images with regular updates for AMD GPU clusters and Tenstorrent solutions. Pricing appears to be enterprise quote-based with multi-year contracts.
2. Turnkey GPU Cluster Solutions: End-to-end AMD GPU cluster solutions combining hardware procurement, cluster design/building, software deployment, and ongoing technical support. Covers rack layout, power planning, network topology design, and deployment of Moreh inference software.
3. Tenstorrent AI Cluster Solutions: Scalable AI cluster solutions built around Tenstorrent Wormhole processors, including hardware supply, cluster construction, software deployment, and continuous technical support for Tenstorrent-specific issues and performance tuning.
4. Custom Benchmarking Services: Custom benchmarking engagements where Moreh runs performance tests on customer workloads to demonstrate cost reduction effects and token-per-dollar improvements, likely as part of enterprise sales process.

Pricing tiers

Go-to-market motion3 records

Distribution channels3 records

Marketing channels6 records

Core offering

Moreh develops and licenses full-stack AI inference software (the MoAI platform) that enables organizations to run and serve large language models on heterogeneous GPU clusters combining AMD Instinct, NVIDIA, and Tenstorrent accelerators. The offering spans chip-level custom kernels, Moreh vLLM inference engines optimized for AMD and Tenstorrent chips, cluster-level orchestration with cross-vendor prefill-decode disaggregation, and turnkey GPU cluster solutions for AI data centers, cloud service providers, telecom operators, and government AI projects.

Product overview

Moreh is an AI infrastructure software company founded in 2020 by HPC experts, offering a modular platform-plus-modules architecture. Its core products include the MoAI Inference Framework (cluster-scale orchestration), Moreh vLLM for AMD and Tenstorrent (chip-level inference engines), MoAI Performance Gateway (intelligent workload routing), MoAI Fabric (cross-vendor KV cache transfer), and MoAI Training Framework (fine-tuning and training). Additional capabilities include SLOPE (long-context prefill engine), TIDE (runtime draft model training), and HetCCL (cross-vendor collective communication). Moreh delivers turnkey GPU cluster solutions for AMD Instinct and Tenstorrent hardware, enabling organizations to unify heterogeneous GPUs across vendors and generations into a single inference cluster. Its subsidiary Motif Technologies develops Korean-language generative AI models. The company targets AI data centers seeking to maximize tokens per dollar through chip-level, cluster-level, and infrastructure optimization.

Differentiator

Problem solved

Functional benefit

Products and services

• MoAI Inference Framework Production-grade end-to-end inference stack optimized for heterogeneous accelerators (AMD, NVIDIA, Tenstorrent) that provides routing and scheduling, auto scaling, SLO-driven optimization, and KV cache management across cluster-scale deployments. Targets enterprise AI data centers running large language models.
• MoAI Performance Gateway Intelligent workload distribution component across heterogeneous accelerators, enabling cross-vendor GPU orchestration and automated optimization decisions within the MoAI Inference Framework. Provides dynamic routing and load balancing across GPUs from different vendors and generations.
• MoAI Fabric Software-defined, cross-vendor GPU memory fabric for KV cache transfer, enabling RDMA-based communication between GPUs from different vendors (e.g., NVIDIA and AMD) across cluster nodes for inference workloads.
• Moreh vLLM for AMD Drop-in vLLM replacement optimized for AMD GPUs that delivers up to 2x higher throughput on AMD Instinct GPUs. Includes custom kernels for GEMM, Attention, and MoE operations, and is distributed as container images with regular updates for AMD GPU clusters.
• Moreh vLLM for Tenstorrent High-performance vLLM serving engine purpose-built for Tenstorrent Wormhole accelerators. Supports state-of-the-art MoE models including DeepSeek, GPT-OSS, and Qwen, with 450+ optimized operators for general heterogeneous GPU inference.
• MoAI Training Framework Training and fine-tuning framework for AMD and Tenstorrent GPU clusters. PyTorch-compatible with 450+ operators, enabling fine-tuning and training on the same cluster used for inference.
• Turnkey AMD GPU Clusters Full-stack turnkey AMD Instinct GPU cluster solution that includes hardware procurement, rack layout and power planning, network topology design (RoCE), Kubernetes platform deployment, and Moreh's inference software stack. Delivered to AI data centers, with 1,200+ AMD MI250 GPUs deployed at KT Cloud for training operations.
• Tenstorrent AI Clusters Cost-efficient AI cluster solution built around Tenstorrent's network-integrated Wormhole processors and Galaxy servers. Moreh provides end-to-end support from hardware supply and cluster building to software deployment and ongoing technical support for Tenstorrent-specific issues and performance tuning.
• Heterogeneous GPU Inference Solution Solution unifying GPUs across vendors, architectures, and generations into a single inference cluster. Supports three scenarios: legacy + new generation GPUs (e.g., H100 + B200), NVIDIA + AMD mixed clusters (e.g., H200 + MI355X), and GPU + AI accelerator mixed (e.g., GPU + Tenstorrent).
• Inference Cost Optimization Solution Cost-per-token optimization solution operating across three levers: chip-level optimization (Moreh vLLM), cluster-level optimization (PD disaggregation, prefix caching, auto scaling), and infrastructure cost reduction (heterogeneous GPU utilization). Achieves up to 2.2x throughput on 40% fewer servers.
• Motif Technologies AI Models Subsidiary that develops Korean-language generative AI models including Motif 12.7B with proprietary GDA (group differential attention) architecture, plus cloud services. Selected as lead team for Korea's government 'Independent AI Foundation Model' project developing multimodal foundation models across text, images, video, and audio.

Quantifiable outcome

• 1.68x throughput improvement for DeepSeek R1 671B on AMD MI300X vs ROCm vLLM baseline
• +6 more outcomes

Named customers4 records

Segments4 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration4 records

AI capability6 records

Feature10 records

Partnerships

Five partnerships are on record, tiered core and secondary.

• Korea Ministry of Science and ICTcoreStrategic or Co-development Partner · 20 February 2026Selected as lead team for Korea's 'Independent AI Foundation Model' project, forming consortium with 17 participating institutions and 12 demand-side institutions. Consortium includes Moreh for GPU optimization, academic institutions for multimodal design, and companies for data creation and synthetic VLA data generation.
• SGLangsecondaryStrategic or Co-development Partner · 11 September 2025Collaboration to expand into deep learning inference market. Joint development of AMD-based distributed inference system. Showcase of distributed inference system on AMD hardware at AI Infra Summit 2025, demonstrating DeepSeek model optimization more efficient than NVIDIA systems.
• Motif TechnologiescoreStrategic or Co-development Partner · 1 February 2025Subsidiary established in February 2025 for model development and cloud services. Led by Motif CEO Lim Jung-hwan, developing proprietary AI architecture including GDA (group differential attention) structure. Motif 12.7B model ranked first among domestic models in Artificial Analysis intelligence index. Selected as lead team for Korea's sovereign AI foundation model project.
• TenstorrentcoreStrategic or Co-development Partner · 18 November 2024Strategic partnership to challenge NVIDIA in AI data center market. Combines Tenstorrent's AI semiconductors (NPUs/Wormhole processors) with Moreh's parallel processing software to support inference and LLM training applications. Planned for commercialization in H1 2025. Joint AI data center solution unveiled at SuperComputing 2025 with cost-efficient alternative to NVIDIA infrastructure.
• KT CloudcoreChannel Partner/ Reseller/ Distributor · 11 November 2022Launched Hyperscale AI Computing service in December 2021 based on Moreh's MoAI platform and AMD Instinct MI250 accelerators. Service became paid in August 2022. Selected as official cloud service for Korean government HPC program. KT Cloud resells GPU access through Moreh's virtualization platform.

Scale indicators13 records

Recent moves10 records

Expansion highlights6 records

Direct peers

• Anyscale: Anyscale operates a Ray-based AI compute platform for production AI workloads including LLM serving and batch inference. Comparable to Moreh as a platform layer that abstracts heterogeneous compute (Ray clusters vs Moreh's MoAI cluster abstraction) for enterprise AI deployment, with similar developer-led go-to-market and OSS roots.
• Together AI: Together AI provides an AI cloud and inference platform optimized for open-source LLMs with custom kernels and inference engines. Directly comparable to Moreh in offering inference-as-a-service with proprietary optimizations on top of open-source serving frameworks (vLLM, FlashAttention), targeting cost-sensitive enterprise and developer customers.
• Fireworks AI: Fireworks AI is a model-serving platform optimized for low-latency LLM inference with proprietary kernel and routing optimizations. Comparable to Moreh's MoAI Inference Framework as a production-grade inference orchestration layer, with similar focus on throughput-per-dollar and serving reliability for frontier models.
• Modular: Modular builds the Mojo language and a unified AI inference platform (MAX) that targets heterogeneous accelerators with high-performance compilation. Comparable to Moreh in pursuing compiler/runtime-level optimization (Moreh IR vs Modular's Mojo/MAX) and in abstracting GPUs from the application developer, with similar AI infrastructure software positioning.
• OctoAI (acquired by NVIDIA): OctoAI built an enterprise AI inference platform optimized across hardware backends before being acquired by NVIDIA. Comparable as a prior independent inference-software specialist competing on cost-per-token across heterogeneous accelerators — a direct competitive template Moreh continues to operate within.
• Replicate: Replicate operates a cloud platform for running and deploying ML models, with proprietary inference optimization on heterogeneous hardware (AMD, NVIDIA, Apple Silicon). Comparable to Moreh's MoAI as a software layer that makes running open-source models efficient across accelerators, targeting developers and enterprises.

Emerging players

• RunPod: RunPod provides GPU cloud infrastructure with inference endpoints and custom kernel optimizations for AMD and NVIDIA hardware. Comparable as a smaller, AMD-friendly alternative GPU cloud whose software stack overlaps with Moreh's turnkey AMD cluster offerings for cost-sensitive AI customers.

Broad incumbents

• CoreWeave: CoreWeave is a large-scale GPU cloud provider with proprietary orchestration and inference serving software (TensorWave). Comparable as an AI infrastructure operator competing for the same cost-optimized inference workloads, though NVIDIA-centric and significantly larger and more capitalized than Moreh.
• Lambda Labs: Lambda provides GPU cloud, on-prem clusters, and inference serving software for AI workloads. Comparable to Moreh in offering turnkey GPU clusters with proprietary software optimizations, though primarily NVIDIA-focused and operating at a larger scale.

Others

• Tenstorrent: Tenstorrent is an AI accelerator (NPU) company and Moreh's strategic partner, with joint AI data center solutions unveiled at SuperComputing 2025. Listed as a peer rather than partner because Tenstorrent's success directly determines Moreh's addressable market on the Tenstorrent hardware ecosystem; the strategic alignment is also a mutual competitive dependency.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights7 records

Customer concentration

• Xx.com/moreh_inc
• Facebookfacebook.com/moreh.inc
• LinkedInlinkedin.com/company/moreh-corp

Compliance1 record

Revenue estimate

Valuation estimate

Number of profiles

Profiles6 records

Subsidiaries2 records

Funding overview

Funding rounds6 records

Investors4 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Moreh