RENEGADE 2026 Summit: Customer and Partner Announcements

At the RENEGADE 2026 Summit, FuriosaAI unveiled a wave of new commercial partnerships and productization built on RNGD. In particular, Samsung SDS announced that starting in July, it will launch RNGD-based cloud AI compute services on Samsung Cloud Platform, marking Korea's first NPU-as-a-Service (NPUaaS) offering from a CSP.

The event was held in Seoul, bringing together global tech leaders, ecosystem partners, and policy stakeholders to discuss the direction of sustainable yet high-performance AI infrastructure. FuriosaAI explained that RNGD is rapidly expanding beyond the pilot stage into large-scale commercial deployment.

Key partner use cases were also announced.

• Anexia: Integrating RNGD into global services to expand into the EMEA market
• Baro AI: Unveiled an AI server combining RNGD with its POSEIDON liquid-cooling architecture
• Creverse: Applying RNGD to power its education platform HUMMINGo
• flexgrid.cloud: Adopted RNGD as a single AI compute solution for micro data centers in power-constrained urban environments
• Lablup: Integrated RNGD into Backend.ai
• LG AI Research / LG U+: Launched a Sovereign AI Appliance combining a Korean-made foundation model, platform, and AI silicon
• MangoBoost: Developed the world's first NPU-DPU scale-out architecture utilizing 400G RDMA fabric
• MegazoneCloud: Announced a new cloud compute project to address regional demand
• Nota AI: Demonstrated a VLM for industrial safety alerts and natural language video search
• Seoul National University: Running a RAG platform combining a large curated DB with real-time web data
• Upstage: Deployed its Solar foundation model on RNGD
• WISEnut: Integrated its agent-specialized LLMs WISE LLOA and WISE iRAG with RNGD

On the performance side, testing the Qwen3-32B (FP8) model under a 30-50 tokens/second per user SLO showed that RNGD served 2.2x to 7.4x more users per kW compared to NVIDIA's RTX PRO 6000. Operating at a chip TDP of 180W versus the 600W of high-end GPUs, RNGD also delivered a claimed 40% TCO reduction.

This performance stems from FuriosaAI's Tensor Contraction Processor (TCP) architecture. Rather than forcing multi-dimensional AI operations into the fixed matrix units of conventional GPUs, it handles tensor contraction as a native hardware primitive, improving efficiency. Combined with an SDK supporting vLLM, Kubernetes, and torch.compile, the company emphasized that RNGD is ready for enterprise deployment right away.