September hardware announcements focused on AI acceleration and memory efficiency. Micron introduced 512 GB RDIMM DDR5-9200 modules, enabling up to 12 TB of memory in two-socket servers, with mass production slated for late 2027. NVIDIA expanded its Blackwell lineup with the RTX PRO 5500 card featuring 84 GB of GDDR7 memory.
Huawei and Alibaba detailed their AI accelerator roadmaps. Huawei plans to release the Ascend 960DT in Q1 2027, followed by the 960PR, 970, and 980 models through 2029. Alibaba is testing its Zhenwu 810E PPU, while HyperAccel introduced Bertha accelerators using LPDDR5X memory to reduce costs, targeting 90% utilization during inference.
Axelera AI began shipping its Europa chips for corporate inference, available as standalone chips or in Edge 232p and Server 250p configurations. The company reports 600 clients and a sales pipeline valued at $1.5 billion. Meanwhile, Arm unveiled the Neoverse CSS N4 Ranger platform, supporting up to 128 cores per die.
Fujitsu announced global sales for its MONAKA processors, which utilize a 3D-stack design with 2nm compute cores and 5nm I/O. Unigen launched two PCIe 5.0 SSD series: the Cheetah SCM for high-performance tasks with capacities up to 3.2 TB, and the Cheetah HC for high-capacity storage reaching 122.88 TB.
ASUS introduced five new servers at the AI Tech 2026 event in Seoul. These include models supporting Intel Xeon 6 processors with up to 8 TB of memory, as well as platforms for AMD EPYC 9006 and 9005 chips featuring PCIe 6.0 support. The lineup includes configurations for both high-density storage and multi-accelerator setups.
