Skip to content
The Nexus
DossierENTITY

HBM4

Coverage of HBM4 in the Nexus archive.

Earliest in view: May 27 · 13:00 UTCMost recent: Jul 29 · 05:18 UTC
Co-mentioned in this coverage
Recent coverage
  • BUSINESSJul 29 · 05:18 UTCTHE REGISTER
    Big Tech demanding deals that smooth out memory prices, says SK Hynix

    SK Hynix has secured long-term supply deals with major AI industry customers to stabilize volatile memory prices, reporting a 257% revenue increase and 557% profit surge in Q2. The company attributes growth to higher DRAM and NAND memory prices and anticipates slower Q3 growth, emphasizing strategic partnerships and mechanisms like deposits to ensure demand visibility and business stability.

  • TECHNOLOGYJul 7 · 17:59 UTCTHE REGISTER
    South Korean chip startup FuriosaAI invades European datacenters

    South Korean AI chip startup FuriosaAI has deployed its RNGD AI accelerators at Equinix's Lisbon datacenter, targeting European demand for sovereign AI compute. The RNGD cards offer 48 GB HBM3, 1.5 TB/s bandwidth, and 512 teraFLOPS of FP8 performance with an 180W TDP, while Furiosa collaborates with Broadcom on a next-gen chip using HBM4 memory.

  • TECHNOLOGYMay 27 · 13:00 UTCTHE REGISTER
    Broadcom's custom ASIC biz adds South Korea's FuriosaAI to its empire

    Broadcom partners with South Korean AI startup FuriosaAI to develop AI accelerators using Broadcom's advanced chip packaging technology. The collaboration aims to create a 2nm Tensor Contraction Processor with dual-layer HBM4 memory, leveraging Broadcom's 3.5D XDSiP tech to streamline multi-die chip design. FuriosaAI's RNGD chips, while less powerful than Nvidia's B200, offer significantly lower power consumption.