"Homegrown NPU Joins Intel Ecosystem"... DeepX, Intel and KTNF Form AI Infrastructure Alliance

Pursuing OpenVINO-NPU integration
Exploring collaboration on physical and edge AI and next-generation chips
Jointly developing AI inference servers based on heterogeneous computing

DeepX is joining forces with Intel Korea to expand AI inference technology and its ecosystem.

DeepX announced on November 13 that it had signed a memorandum of understanding (MOU) with Intel Korea the previous day at COEX in Seoul to cooperate on AI semiconductor technology. The two companies plan to advance DeepX's neural processing unit (NPU) technology by leveraging Intel's computing platforms and software ecosystem. Their four main areas of cooperation are integrating Intel's OpenVINO with DeepX's NPU, collaborating on physical and edge AI platforms, validating system architectures, and exploring next-generation generative AI solutions.

(From left) DeepX CEO Kim Nokwon, Intel Korea CEO Bae Taewon, and KTNF CEO Lee Jungyeon pose for a commemorative photo at a signing ceremony for a three-party business agreement at COEX in Seoul on the 12th. DeepX

The core of the collaboration is combining Intel's AI inference software, OpenVINO, with DeepX's NPU. The companies will explore ways to optimize model conversion and execution compatibility in the OpenVINO environment. Their aim is to create an environment in which developers can easily use DeepX's NPU with existing AI models and development tools. In physical AI, they will integrate the host and control functions of Intel x86 processors with the perception and inference capabilities of DeepX's NPU. Building on this, they will pursue technology demonstrations and broader applications in robotics.

The companies will also explore developing solutions using DX-M2, a next-generation generative AI semiconductor. DX-M2 is a next-generation chip under development based on Samsung Electronics' 2-nanometer (nm; 1 nm = one-billionth of a meter) process. It is designed to run large language models (LLMs) and vision-language models (VLMs) in a low-power environment of approximately 5 watts. The companies will assess the feasibility of developing enterprise AI appliances and on-device AI services using the chip.

DeepX has also expanded its cooperation into server infrastructure. On the same day, it signed a separate three-party MOU with Intel Korea and KTNF, a domestic server specialist. The three companies agreed to collaborate on on-premises AI inference servers using heterogeneous computing technology. On-premises refers to building and operating infrastructure directly in a company's own data center or on its own servers rather than using an external cloud.

DeepX will provide inference NPUs and AI model optimization technology. Intel Korea will support CPU and xPU computing platforms and related software. KTNF will contribute its capabilities in server design and manufacturing and system integration.

The three companies will jointly explore developing heterogeneous computing platforms and designing reference systems for on-premises AI inference servers. They also plan to validate the integration of AI software stacks and support joint proofs of concept (PoCs) for domestic and international customers. Through these efforts, they aim to demonstrate AI inference performance and compatibility in real-world server environments, respond to demand from businesses and public institutions for their own AI infrastructure, and explore joint commercialization.

This content was produced with the assistance of AI translation services.