Three South Korean startups are taking part in the business of ai&, a newly launched Japanese neocloud company.
According to information technology industry sources on the 30th, MOREH, Motif Technologies and Rebellions are participating in the operations of the Japanese neocloud provider ai&. A neocloud refers to a cloud service operator that specializes in supplying high-performance computing resources, such as graphics processing units (GPUs), needed to develop and run artificial intelligence services. Launched in March this year, ai& delivers AI model inference services and AI app development platforms through the cloud.
ai& is building an AI data center in Tokyo to provide AI services and is expanding it to 40 megawatts. It has adopted MOREH's software to operate the Tokyo facility. ai& uses Nvidia's B300 GPU and AMD's MI355X GPU simultaneously as AI server resources, seeking to cut the token costs incurred when AI services are used and to improve those services. Computing requests from ai&'s customers must be allocated to different GPUs as they arise, and that is where MOREH's AI infrastructure operating software, the Token Factory operating system (OS), comes in. MOREH has supplied the Token Factory OS to ai& since April, keeping a data center environment that mixes Nvidia and AMD GPUs running stably. The deal is also MOREH's first Token Factory OS contract with an overseas company. MOREH is in talks with ai& on further business cooperation.
Rebellions has joined Nvidia and AMD as an AI chip supplier for ai&'s Tokyo data center. Rebellions develops neural processing units (NPUs), chips specialized for AI inference. The company will supply ai& with 100 racks of servers equipped with REBEL100, its second-generation NPU. The export is also the first time Rebellions has generated revenue in the Japanese market by selling products by the rack. Until now, the company had to settle for limited revenue from exporting products as NPU cards or individual servers.
On the AI infrastructure built this way, ai& provides AI inference services to its clients. Customers can choose from a total of 11 AI models, including Google's Gemma and OpenAI's gpt-oss. Among Korean AI models, Motif Technologies' large language model (LLM) Motif 3 is the only one included. Motif 3 is an LLM created through the company's participation in the government's program to develop homegrown AI foundation models. Motif Technologies is also a subsidiary of MOREH, and MOREH's involvement in the ai& business led to the LLM supply deal.
The three startups expect to use the collaboration with ai& as a business reference in Japan and expand their operations there further. "In Japan, demand is growing for AI inference infrastructure to be applied to industries and services, and in that process the importance of various AI chips and software is increasing," MOREH CEO Cho Kang-won said. "We expect that Korean AI startups will broaden their points of entry into Japan's AI business through their participation in the ai& project."
