Edgecore (an Accton Company) and Partners Unite at Computex 2026: Powering the All-Photonics AI Era Together

Why It Matters

By replacing electrical interconnects with integrated photonics, the Open Fabric cuts latency, slashes power consumption, and enables the ultra‑dense AI workloads driving next‑generation services. Its open, composable architecture lowers vendor lock‑in and accelerates adoption of high‑performance AI infrastructure worldwide.

Key Takeaways

• Edgecore Open Fabric integrates optics, liquid cooling, and open networking for AI
• Optical switch IRX3032 OWS offers 160 ns latency across 32 wavelengths
• Liquid cooling CDU handles 200 kW per rack, improving PUE by 20 %
• CXL 3.1 enables up to 20 TB pooled memory across GPU clusters
• Ecosystem includes AMD, Intel, Broadcom, and NTT to meet IOWN standards

Pulse Analysis

The data‑center industry is at a tipping point as AI models demand ever‑higher compute density. Traditional copper interconnects struggle with heat dissipation and latency, limiting the speed at which GPUs can exchange data. Photonic networking, which transmits data as light, offers near‑zero signal loss and sub‑microsecond latency, effectively removing the electrical bottleneck. This physics‑driven advantage is now practical thanks to advances in LCoS optical switches and DWDM technology, allowing terabit‑scale bandwidth within a single rack.

Edgecore’s Open Fabric translates those advances into a turnkey solution. At its core is the IRX3032 OWS optical switch, paired with InLC’s wavelength‑selective engine, delivering 102.4 Tbps of bandwidth with no O‑E‑O conversion penalty. The platform’s compute layer leverages CXL 3.1 and Astera Labs’ retimers to pool up to 20 TB of memory across GPU nodes, while LIQID’s software orchestrates dynamic resource allocation. Integrated liquid‑cooling units provide 200 kW per rack with N+1 redundancy, cutting PUE by roughly 20 % and making ultra‑dense AI clusters sustainable. The open‑source SONiC‑compatible networking stack ensures interoperability and reduces vendor lock‑in.

Strategically, the launch positions Edgecore and its partners to capture a growing market for AI‑optimized infrastructure. Taiwan’s AI Island initiative, backed by the Accton Group, aims to become a global hub for high‑performance computing, and the Open Fabric aligns perfectly with that vision. As hyperscalers and enterprise AI teams seek to scale models without prohibitive energy costs, a photonics‑first, open architecture offers a compelling value proposition. Early deployments in remote surgery, real‑time media production, and distributed training signal that the all‑photonics era is moving from concept to commercial reality, setting a new performance baseline for the next decade.