Broadcom and FuriosaAI partner to build rack-scale inference platform for AI infrastructure
Type: Partnership · Technology: Data Center · Country: United States · Announced: 2026-05-27
Broadcom and South Korean AI infrastructure startup FuriosaAI announced a strategic partnership to develop a rack-scale inference platform designed to shift AI infrastructure beyond GPU-centric architectures. The collaboration pairs FuriosaAI’s third-generation Tensor Contraction Processor with Broadcom’s networking, packaging, and interconnect technologies to create a multi-die chiplet inference system for hyperscale AI deployments. The platform integrates Furiosa’s accelerator architecture with Broadcom’s Ethernet fabrics, PCIe technologies, and advanced packaging to scale inference clusters across thousands of nodes. Furiosa’s third-generation accelerator will use a 2nm compute die paired with dual-layer HBM4/4E memory and Broadcom’s packaging technology, while Broadcom’s Ethernet and PCIe technologies provide high-bandwidth rack-scale connectivity. The partnership reflects growing industry momentum toward Ethernet-based AI infrastructure as vendors seek alternatives to proprietary GPU fabrics.