At MWC 2026, FuriosaAI and LG U+ have officially launched the industry’s first “ready to run” Sovereign AI Appliance.
By integrating Furiosa’s RNGD inference accelerator with LG’s EXAONE 4.0 model and ixi-Enterprise platform, this solution provides complete, secure, and powerful AI infrastructure in a single server package.
Solving the AI infrastructure bottleneck
For AI infra decision-makers, high operating costs and complex deployments have been an unavoidable aspect of GPU-based infrastructure.
The Sovereign AI Appliance provides a 30% reduction in Total Cost of Ownership (TCO) compared to traditional GPU clusters. Because RNGD is built on Furiosa’s Tensor Contraction Processor (TCP) architecture, it is fully compatible with standard 15kW air-cooled data center racks. This eliminates the need for expensive and time-consuming liquid-cooling retrofits required by modern GPU-based systems.
“The Sovereign AI Appliance enables enterprises to move quickly and easily from standard cloud-based LLM services to a solution that maintains absolute data isolation and is engineered for 24/7 mission-critical operations,” said Sang-Yup Lee, Chief Technology Officer of LG U+. “With RNGD in mass production and LG’s advanced models and agentic platform available now, this is the solution that enterprises in highly regulated industries have been waiting for.”
Sovereign by design
The Sovereign AI Appliance operates entirely on-premise. It is designed to be plugged in and immediately ready to run a complete Retrieval-Augmented Generation (RAG) pipeline.
Since no external internet connection or cloud API calls are required, the Sovereign AI Appliance is well-suited for enterprises where data security is paramount:
- Defense: Turning large numbers of physical documents into a local knowledge base, so an AI agent can answer tactical or maintenance queries on a closed, secure network.
- Healthcare: Implementing Clinical Decision Support (CDS). The appliance can process patient records and lab results locally to assist doctors with diagnosis without exposing patient data to the cloud.
- Public Sector: Automating administrative workflows and document-based data retrieval within secure facilities.
- Manufacturing/R&D: Protecting intellectual property. Engineers can query internal blueprints and logs without uploading sensitive IP to a public cloud.
AI as an operational utility
The Sovereign AI Appliance reference configuration utilizes a two-server cluster housing 14 RNGD cards, delivering:
- Performance: 7,168 TOPS (FP8) aggregate throughput.
- Context: Native support for 32K context lengths, optimized for document-heavy workloads (averaging 4-6K tokens).
- Architecture: Simultaneous execution of multi-modal parsers (Qwen3-VL), high-fidelity indexing (Qwen3 Embeddings), and the EXAONE 4.0 sovereign backbone.
LG U+ and Furiosa are developing future iterations that will include NPUaaS (NPU-as-a-Service) for data centers and Physical AI Appliances designed to act as the "on-site brain" for industrial robotics and factory control systems.
See it live at MWC
At MWC? Stop by Hall 3, Stand 3M11 to meet the team, schedule a briefing or technical deep dive, and learn more about Sovereign AI Appliance.
Written by
