FuriosaAI expands AI chip roadmap and software partnerships

South Korean AI semiconductor startup FuriosaAI is advancing its hardware and software ecosystem through strategic partnerships and next-generation chip development. The company has signed a Memorandum of Understanding (MOU) with Orchestro Group to collaborate on NPU-based AI infrastructure and platform technologies. This partnership aims to integrate FuriosaAI’s second-generation NPU, ‘RNGD’, with Orchestro’s ‘CONCERTO AI’ inference management platform, targeting public, financial, and enterprise markets with integrated ‘K-AI infrastructure’ solutions.

In terms of hardware roadmap, FuriosaAI revealed specifications for its third-generation chip, which is being co-developed with Broadcom using TSMC’s 2nm process. Targeting a 2028 commercialization, the massive chip is designed to feature 12 High Bandwidth Memory (HBM) units, providing 576GB of capacity and a bandwidth of 48TB per second. This architecture aims to overcome single-chip scaling limits to compete in the global hyperscaler market. While industry observers have raised concerns regarding the procurement of such large volumes of HBM, FuriosaAI intends to leverage Broadcom’s supply chain strength to mitigate potential bottlenecks.