en.Wedoany.com Reported - Recently, Edgecore, an open networking equipment company from Taiwan, China, launched the Edgecore Open Fabric: Built for IOWN all-photonic AI infrastructure platform at COMPUTEX 2026. Designed to meet the high bandwidth, low latency, thermal management, and open networking requirements of AI data centers, this platform integrates optical switching, open networking, composable computing, and liquid cooling into a single deployable architecture.
The platform launched by Edgecore directly targets the infrastructure bottlenecks facing AI data centers. As large model training and inference clusters continue to scale, data exchange between GPUs increases rapidly, amplifying the pressure on traditional electrical interconnects in terms of power consumption, distance, latency, and heat dissipation. Based on NTT's IOWN all-photonic network concept, the Edgecore Open Fabric aims to reduce interconnection losses within data centers and minimize delays caused by optical-electrical conversions using optical transmission and switching. The platform incorporates components such as the Edgecore IRX3032 optical wavelength switching device, the AMX3200 transmission/multiplexing device, and the AIS1600-64O AI/ML switch. The IRX3032 can simultaneously route 32 wavelength channels with an end-to-end switching latency of 160 nanoseconds; based on 32 800G wavelengths, the bandwidth within a single fiber for a data center can reach 102.4Tbps. For AI clusters, the significance of this architecture lies not only in increasing point-to-point bandwidth but also in enabling lower latency and higher energy efficiency for resource scheduling across racks, nodes, and data centers.
Ecosystem participants for this platform include AMD, Astera Labs, Broadcom, InLC Technology, Intel, LIQID, Marvell, NTT, Penguin Solutions, Preferred Networks, and 1Finity.
From a system architecture perspective, Edgecore positions open networking, composable computing, and liquid cooling as three layers of capability within the same platform. The network layer employs open switching equipment compliant with the SONiC ecosystem, designed according to IOWN APN specifications, emphasizing programmability, composability, and multi-vendor transparency; the computing layer introduces CXL 3.1 memory interconnect capabilities, supporting GPU clusters in sharing up to 20TB of pooled memory between nodes, allowing computing and memory resources to be reconfigured per task; the thermal layer integrates rack-level liquid cooling distribution units, providing 100kW to 200kW rack heat exchange capacity, equipped with N+1 pump redundancy, real-time flow monitoring, leak detection, and integration with building management systems and software-defined networking. Competition in AI data centers is shifting from simply stacking GPUs to the overall engineering capability encompassing "compute, network, memory, optical interconnect, cooling, and software orchestration." By integrating these modules into a single all-photonic AI infrastructure platform, Edgecore demonstrates that open networking vendors are extending their reach toward more comprehensive data center infrastructure providers.
Future deployment outcomes will depend on the maturity of the IOWN ecosystem, the mass production capability of optical switching equipment, the engineering stability of CXL pooled memory, the delivery cost of liquid cooling systems, and the testing progress by cloud service providers and AI infrastructure operators. As AI cluster power density continues to rise, all-photonic networks, open switching, and composable infrastructure will become key directions for upgrading high-density data centers, potentially reshaping the procurement boundaries between traditional data center networking equipment, optical modules, cooling systems, and server architectures.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
