LG U+ and FuriosaAI Unveil Sovereign AI Appliance
Key point
LG U+ and FuriosaAI unveiled an on-premise inference server combining a domestically developed model, platform, and AI silicon.
Details
At MWC 2026, FuriosaAI and LG U+ officially unveiled the Sovereign AI Appliance. It is a 'ready to run' AI inference solution that packages a foundational model, platform, and AI silicon developed in Korea into a single server package.
The core is the combination of the Furiosa RNGD inference accelerator, LG EXAONE 4.0, and the ixi-Enterprise platform. This provides secure AI infrastructure that requires almost no separate configuration, and is designed to run RAG pipelines immediately in an on-premise environment.
Operational efficiency was also emphasized. The companies highlighted a 30% reduction in TCO compared to existing GPU clusters, explaining that RNGD, based on TCP architecture, is compatible with standard 15kW air-cooled data center racks, eliminating the need for costly liquid-cooling retrofits.
The target applications were presented as industries where data security is critical.
- Defense: Converting documents into a knowledge base within closed networks to answer tactical and maintenance queries
- Healthcare: CDS support that processes patient records and test results locally
- Public Sector: Automating administrative tasks and document search within secure facilities
- Manufacturing/R&D: Utilizing design drawings and logs without uploading them to external cloud
The reference configuration features a 2-server cluster equipped with 14 RNGD cards, delivering 7,168 TOPS(FP8) aggregate throughput. It supports a 32K context length, and is configured to simultaneously run Qwen3-VL, Qwen3 Embeddings, and the EXAONE 4.0 sovereign backbone, tailored for document-centric workloads.
The two companies plan to further develop NPUaaS and a Physical AI Appliance for industrial robots and factory control.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.