Edgecore, an Accton Company, announced the Edgecore Open Fabric: Built for IOWN® at Computex 2026. The solution combines optical switching, open networking, composable compute, and liquid cooling into a single AI‑focused infrastructure platform, positioning it for enterprises confronting rising power density, latency, and vendor lock‑in challenges.
Edgecore Open Fabric: Built for IOWN® Revealed
The Edgecore Open Fabric is described as a “next‑generation, all‑photonics network architecture purpose‑built for the AI era.” Developed with NTT’s IOWN initiative and partners AMD, Astera Labs, Broadcom, InLC Technology, Intel, LIQID, Marvell, NTT, Penguin Solutions, and 1Finity, the platform integrates:
- Optical switching – Edgecore’s IRX3032 OWS routes 32 wavelength channels with 160 ns end‑to‑end latency and no O‑E‑O conversion.
- High‑bandwidth fabric – 32 × 800 G wavelengths per fiber deliver 102.4 Tbps intra‑datacenter bandwidth.
- Composable compute – CXL 3.1 memory pooling up to 20 TB, enabled by Astera Labs PCIe retimers, Broadcom fabric silicon, and LIQID software.
- Liquid cooling – An integrated Cooling Distribution Unit provides 100–200 kW per rack with N+1 pump redundancy and reported PUE gains of up to 20%.
The architecture adheres to NTT’s IOWN All‑Photonics Network (APN) specification, using Broadcom silicon‑based spine switches that are SONiC‑compatible and programmable.
Positioning Within the AI Infrastructure Stack
Edgecore frames the Open Fabric as a response to three pressing data‑center constraints:
- Heat – Modern AI clusters exceed 100 kW per rack, outstripping air‑cooling capacity.
- Latency – Electrical interconnects add latency that hampers AI model performance.
- Lock‑in – Proprietary solutions limit flexibility and cost control.
By moving the data path to light, the platform aims to keep pace with GPU acceleration while reducing power consumption. The optical switch’s 5 µs per kilometer propagation speed and zero electromagnetic interference are highlighted as inherent advantages over copper.
The ecosystem partners contribute specific layers: AMD supplies GPUs, Intel provides accelerator and inference technology, Broadcom delivers high‑radix switching silicon, and InLC Technology supplies the wavelength‑selective switch (WSS) core of the IRX3032 OWS. 1Finity extends the solution to metro and carrier networks via DWDM transport.
Operational Relevance for Enterprise Deployments
Edgecore’s announcement includes a live demonstration at Computex 2026 (Booth K1309, Taipei Nangang Exhibition Center Hall 1) featuring the full rack architecture, optical fabric, and liquid‑cooling CDU in operation. The company states that the platform is “production‑ready” and already supports real‑world AI workloads such as:
- Smart remote surgery – Ultra‑low‑latency robotic collaboration (NTT).
- Distributed media production – Sub‑millisecond live broadcast workflows (NTT).
- Industrial safety monitoring – Low‑latency factory‑wide sensing (IOWN Forum).
These use cases are presented as deployments rather than prototypes, indicating that the All‑Photonics Network is already in service.
Key Takeaways
- Edgecore announced the Edgecore Open Fabric: Built for IOWN®, an all‑photonics AI infrastructure platform integrating optical switching, open networking, composable compute, and liquid cooling.
- The IRX3032 optical switch routes 32 wavelength channels with 160 ns latency and supports 102.4 Tbps intra‑datacenter bandwidth.
- The integrated Cooling Distribution Unit delivers 100–200 kW per rack with documented PUE improvements of up to 20%.
TechInsyte's Take
Edgecore’s all‑photonics stack offers a concrete alternative to traditional copper‑based data‑center designs, directly addressing power‑density and latency limits that many AI workloads now encounter. Buyers should monitor early field data from the Computex demo and verify that the claimed cooling and latency metrics hold under their specific workload mixes before committing to a full‑scale rollout.
Source: Businesswire
