AI semiconductor design company Moblint is accelerating the expansion of its neural processing unit (NPU) business.
The company is scaling up mass production of industrial and on-device NPUs while strengthening its push into the on-premise AI server market. It has recently taken its first domestic NPU space demonstration, broadening the technology's application areas into space.
The launch of the fifth Nuri rocket on the 7th marked the beginning, with two types of NPU boards based on Moblint's 'REGULUS' carried aboard. This is the first time a domestically designed NPU semiconductor has traveled to space.
Moblint CEO Shin Dong-ju cited the small size, low power consumption, and high compatibility with the latest AI models as the reasons REGULUS was selected for the space demonstration.
“Currently, there are almost no AI semiconductors specialized for the space environment, so there is a need for semiconductors that consume little power, are small in size, and can broadly support the latest AI models among existing products,” said Shin. “REGULUS was chosen for these technical characteristics.”
REGULUS is the second-generation AI semiconductor developed by Moblint. While the first-generation product 'ARIES' is an on-premise AI accelerator installed in servers and industrial sites, REGULUS is an on-device system-on-chip (SoC) that runs AI inside devices such as robots, drones, and cameras.
Designed for environments with limited power and space, such as robots and drones, its small size and low power characteristics aligned with the AI computing requirements of satellites.
Moblint plans to connect this demonstration to follow-up space AI business. It is currently proposing to supply AI computing modules to the space microgravity-based pharmaceutical research and manufacturing platform being pursued by SpaceLynk, and is conducting proof of concept (PoC) on the ground. Depending on the progress of the project, it is also considering expanding the scope to the International Space Station (ISS), free-flying vehicles, and private space stations.
In its existing NPU business, the company is also accelerating mass production. Moblint is increasing supply of products based on the first-generation chip ARIES this year while pushing for mass production of REGULUS and securing new customers.
“On-device is not the only important area in the edge AI market; the importance of on-premise will also grow rapidly,” said Shin. “In the on-premise market, price competitiveness of AI servers that can lower initial adoption costs and operating costs is key.”
He added, “Since we have completed product verification with global server manufacturers and are actually proceeding with deliveries, we will quickly expand our customer base and partnership partners in AI agents, control centers, and smart factories.”
In fact, the first-generation AI accelerator ARIES is supplied in various forms, including PCIe cards, MXM modules, and AI boxes. It has secured application cases in industrial sites such as video analytics, industrial safety, and security monitoring. Notably, it supplied 87 MLA400 cards based on ARIES to the Korea Expressway Corporation's AI CCTV project, marking a case where a domestic NPU was actually introduced into public infrastructure.
