AEWIN shows DC-MHS AI servers with Arivor two-phase liquid cooling at OCP 2026
AEWIN Technologies, a Qisda Group company, will present modular AI servers based on the DC-MHS framework at OCP 2026, paired with two-phase direct liquid cooling (2P DLC) from its sister company Arivor. The systems target rack-scale deployments and AI factories, according to the company.
- AEWIN, a Qisda Group company, shows DC-MHS AI servers with Arivor's 2P DLC at OCP 2026.
- The lineup includes the 1U BG18-A10710, 2U BG28-A10710 and flagship 4U BG48-A20813 with up to eight double-width FHFL GPUs.
- According to AEWIN, the cooling approach targets infrastructure of more than 200 kW per rack with reduced PUE.
- Systems use AMD EPYC 9006 CPUs, PCIe Gen6, an ASPEED AST2700 BMC and 80 PLUS Titanium power supplies.
- AEWIN exhibits at booth D9 at the OCP Global Summit, 12 to 15 October.
Servers built on AMD EPYC 9006 and PCIe Gen6
AEWIN says its new portfolio uses server CPUs from the AMD EPYC 9006 series, which it describes as offering superior compute performance and PCIe Gen6 connectivity for the AI era. The company argues that higher compute power creates critical thermal management challenges, and it is addressing them by integrating rack-scale 2P DLC technology from its subsidiary.
According to AEWIN, this approach is meant to handle high-density infrastructure of more than 200 kW per rack, with reduced PUE and long-term sustainability. The company also says it draws on the Qisda Group's combined expertise in computing, networking, storage and thermal management to deliver flexible, energy-optimized infrastructure for AI factories.
The server lineup
AEWIN names several models in the announcement:
- BG18-A10710: a 1U system that integrates 2P DLC to support CPUs with 600 W TDP and two single-width 600 W GPUs.
- BG28-A10710: a 2U AI server with four double-width FHFL PCIe 6.0 x16 slots, supporting 600 W GPUs.
- BG48-A20813: the flagship and newest addition, a 4U AI server holding up to eight double-width FHFL GPUs. AEWIN says it is designed for complex agentic AI and training of next-generation models.
- BS28-A10710: a data-intensive model with 16 E3.S drive bays.
- BX28-A10710: supports up to 12 U.2 NVMe SSDs or 24 SATA drives for different storage configurations.
The storage-oriented systems are positioned for AI data pipelines and intelligent data management.
Management, power and target workloads
The platforms use system management based on OCP DC-SCM2, controlled by an ASPEED AST2700 BMC, and redundant 80 PLUS Titanium power supplies. AEWIN says this gives a reliable foundation for AI inference, HPC, cloud-native and neocloud deployments.
The company says the modular DC-MHS architecture and rack-scale approach show its ability to build powerful, scalable and reliable servers for AI infrastructure. By combining its server expertise with Arivor's 2P DLC technology and the wider Qisda ecosystem, AEWIN claims the solution meets growing demand for AI compute, high-speed connectivity, storage scalability and thermal efficiency.
Where to see it
AEWIN will exhibit at the OCP Global Summit 2026 from 12 to 15 October at booth D9. Further information is available at www.aewin.com. For buyers building AI infrastructure, the announcement points to a combined server and cooling offering aimed at dense, power-hungry GPU deployments, though the source gives no pricing or availability details.
