Rethinking AI Infrastructure for the Age of Reasoning | Jeehoon Kang, FuriosaAI | RAISE 2026

Furiosa AI is initiating mass production of its second-generation Renegade chip, a 180-watt tensor contraction processor fabricated on TSMC 5nm technology to deliver 30% to 100% higher token-per-watt performance in traditional air-cooled data centers. The device utilizes a hybrid HBM-SRAM architecture and a deterministic compiler stack to eliminate power efficiency bottlenecks while offering immediate deployment via OpenAPI and Kubernetes plugins. Following this launch, the company has partnered with Broadcom to develop the third-generation Torque chip, which targets hyperscale direct liquid cooling systems with expected performance scaling of up to 30 times the current generation.