FuriosaAI to Mass Produce Second-Gen AI Chip, Supply to Samsung SDS Cloud From July

Why It Matters

The partnership accelerates NPU adoption in cloud services, giving enterprises AI‑optimized compute without capital expense and challenging GPU‑centric data‑center models.

Key Takeaways

• 20,000 Renegade chips slated for 2024 production
• Samsung SDS to launch NPUaaS using Renegade in July
• HBM3E memory upgrade boosts capacity to 72 GB
• TSMC continues as sole manufacturer for mass production
• Rack‑scale NXT RNGD delivers up to 20 petaflops

Pulse Analysis

FuriosaAI’s second‑generation Renegade chip marks a decisive step in the race to supply high‑performance neural processing units for data‑center AI workloads. Built on a dual‑HBM stack, the RNGD delivers 1.5 TB/s bandwidth and a 180‑watt PCIe variant, while the rack‑scale NXT RNGD pushes 20 petaflops per enclosure. The upgrade to fifth‑generation HBM3E raises memory capacity from 48 GB to 72 GB, a 1.5‑fold increase that enables larger model inference. Production, handled by TSMC, moves from an initial 4,000 units to a target of 20,000 chips in 2024.

Samsung SDS’s decision to integrate Renegade chips into its cloud platform underscores the growing demand for NPU‑as‑a‑service offerings in Asia. Launching in July, the company will provide Korea’s first subscription‑based NPU infrastructure, allowing enterprises to rent AI‑optimized servers without upfront hardware costs. By pairing FuriosaAI’s high‑bandwidth memory with Samsung’s managed cloud services, the partnership promises lower latency and higher throughput for generative‑AI and real‑time analytics workloads. This collaboration also diversifies Samsung’s AI hardware portfolio beyond traditional GPUs, positioning it as a key player in the emerging NPU market.

Looking ahead, FuriosaAI has already outlined a roadmap that targets a third‑generation chip by 2028, aiming to extend the Renegade architecture with further performance scaling. Securing HBM3E supplies from SK Hynix and maintaining TSMC as its foundry gives the company a stable supply chain amid global semiconductor shortages. As data‑center operators seek to reduce energy per inference, the 180‑watt PCIe version and the 3‑megawatt‑class rack solution provide flexible power envelopes. Success of the Renegade line could pressure established GPU vendors to accelerate their own NPU development programs.