Revelion Achieves Global GPU-Level AI Infrastructure with Domestic NPU

Revelion has successfully implemented infrastructure performance comparable to global GPU servers using domestic AI semiconductors and a large language model. The company demonstrated its AI semiconductor server 'Revelion Server' running SK Telecom's proprietary AI foundation model 'A.X K1,' handling concurrent user requests via an agent service demo. The test validated infrastructure efficiency by executing a 500+ billion parameter domestic large language model on a single server, reducing data center construction and operational costs. Revelion claims this configuration provides infrastructure capabilities equivalent to global high-performance GPU servers. A.X K1 employs MoE (Mixture of Experts) architecture for efficient large model operations, with Revelion developing optimized software for distributed computation. The demo also proved practicality of 'map-based agent services,' showing real-time processing of concurrent user queries without latency. Revelion, participating in SK Telecom's consortium for the Ministry of Science and ICT's AI foundation model project, plans to contribute to the government's 'Everyone's AI' initiative. The CEO emphasized this demonstration as a sovereign AI case proving both infrastructure efficiency and practical applicability through domestic AI technology.