MediaTek has expanded NVLink Fusion work for custom accelerator programmes. The collaboration adds advanced packaging, HBM integration, rack-scale interconnect, and a $3.5 billion investment.
MetaX has moved its C600 accelerator into commercial mass production. Rising GPU shipments lifted first-half revenue 44.67%, while the 1,000W OAM 2.0 module adds high-bandwidth memory, ECC, and MetaXLink scaling for systems of up to 128 GPUs.
AMD will power Finland’s €387.8m LUMI-AI supercomputer with next-generation silicon. Instinct MI430X GPUs and 256-core sixth-generation EPYC processors underpin a BullSequana XH3500 system scheduled for deployment during 2027.
AWS and NVIDIA will deploy two million additional GPUs globally. The programme extends through Blackwell Ultra, Rubin and Rubin Ultra hardware, Vera CPUs, NVLink Fusion, custom high-bandwidth memory, government infrastructure, and robotics.
IBM is designing a processor supporting native Arm and Z. The 2 nm chip contains 11 cores above 5.7 GHz while executing both instruction architectures within each core.
Achronix has released JESD204C IP for high-bandwidth Speedster7t FPGA systems. The core supports 32 Gbps lanes, deterministic latency, and aggregate serial throughput reaching 1 Tbps.
Primemas has detailed rack-scale CXL memory for demanding PNNL workloads. The Abaco project combines high-density controller silicon, Micron DDR5, and memory pooling to exceed 100TB per rack.