| Mobile Web

Search results for HBM3

Industry

FuriosaAI to co-develop third-generation AI accelerator with Broadcom

FuriosaAI said on May 28 it has signed a strategic partnership with Broadcom to jointly develop a next-generation AI accelerator and inference platform. The companies plan to advance FuriosaAI’s Tensor Contraction Processor into a multi-die chiplet system and build an integrated infrastructure platform combining computing, networking and software. The collaboration is based on FuriosaAI’s second-generation RNGD accelerator, with third-generation samples planned for the first half of 2028.

Industry

AI chip cost breakdown shows memory share reaches 63 percent

Memory accounted for as much as 63 percent of AI chip component costs in the fourth quarter of 2025, an estimate showed. Epoch AI examined AI chips designed by Nvidia, AMD, Google and Amazon and estimated the memory share rose from 52 percent in the first quarter of 2024. In dollar terms, memory-related spending climbed to about $32 billion in 2025 from about $12 billion in 2024. The figures are estimates with stated uncertainty ranges.

Industry

China-led low-spec HBM front expands, raising concerns over impact on Korean memory profits

China’s memory push is moving into high-bandwidth memory, with CXMT seeking mass production of HBM3 and reallocating 20% of capacity, or 60,000 wafers a month, to HBM. While China may struggle to catch up quickly, analysts warn the bigger risk is price disruption as legacy DRAM output shrinks and AI-focused memory expands. TrendForce expects narrowing price gaps between HBM3E and DDR5, potentially squeezing margins below HBM4.