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

TAIPEI, Taiwan. As artificial intelligence reshapes every sector of the global economy, the networks and data centers that power it face a reckoning. Heat is the enemy. Latency is the bottleneck. Proprietary lock-in is the tax. At Computex 2026, Edgecore (an Accton Company) is answering all three challenges at once, with light.

The two companies today announced the Edgecore Open Fabric: Built for IOWN®, a next-generation, all-photonics network architecture purpose-built for the AI era. Designed in close collaboration with НТТ’s groundbreaking Innovative Optical and Wireless Network (IOWN) initiative and an industry-spanning ecosystem of technology leaders, the platform combines optical switching, open networking, composable compute, and direct liquid cooling into a single, deployable solution that sets a new standard for what AI infrastructure can, and must, look like.

The timing could not be more urgent. Modern AI training clusters are pushing rack power densities beyond 100 kW, levels that traditional air cooling cannot safely sustain, and that conventional electrical interconnects cannot serve without introducing latency that undermines the very intelligence they are meant to support. The industry has known for years that a fundamental shift was coming. Computex 2026 is where Accton Group draws the line and says: it has arrived.

Why Photonics? Why Now?

The physics are unforgiving and, for once, entirely on our side. Light travels through fiber at roughly 5 microseconds per kilometer, with zero signal degradation and zero electromagnetic interference. An optical switch like Edgecore’s IRX3032 OWS routes 32 wavelength channels simultaneously at 160 nanoseconds of end-to-end latency, with no optical-to-electrical-to-optical (O-E-O) conversion penalty. Scale that across a modern AI fabric (32 × 800G wavelengths per fiber, delivering 102.4 Tbps of intra-datacenter bandwidth), the result is infrastructure that finally keeps pace with the GPUs it connects.

This is the beating heart of the IOWN All-Photonics Network (APN) architecture: a vision where light, not electrons, carries the data from edge to core, where the constraints of electrical signaling no longer define the ceiling of AI capability. The Edgecore Open Fabric is built to realize that vision at production scale, today.

The Open Fabric: Built Layer by Layer

01. Open Networking

At the core of the platform is Edgecore’s IOWN DCI Gateway, anchored by the IRX3032 OWS and the AMX3200 Trans/Muxponder. At the heart of the IRX3032 OWS is InLC Technology’s Wavelength Selective Switch (WSS), the in-house photonics engine that enables precise, low-loss optical channel routing without any electrical conversion. Together they deliver DWDM wavelength switching with zero O-E-O conversion, ROADM programmability, and APN-native protocol support. The spine layer is powered by the Edgecore AIS1600-64O, a 102.4 Tbps AI/ML switch powered by AMD processors and Бродком switching silicon, open, programmable, and SONiC-compatible. From leaf to DCI, every hop is designed to be composable, standards-based, and vendor-transparent. The entire architecture is built to НТТ’s IOWN APN specification, ensuring every component meets the rigorous requirements of the IOWN All-Photonics Network standard.

02. Composable Compute

The compute layer takes advantage of CXL 3.1, the memory interconnect standard that allows GPU clusters to share up to 20 TB of pooled memory across nodes as if it were local. Paired with Astera Labs’ PCIe retimers, Бродком’s high-speed fabric interconnect silicon, and Marvell’s CXL controller technology, the platform enables flexible, software-defined disaggregation of memory and compute resources. LIQID’s software-defined composable infrastructure layer makes this flexibility operationally accessible, while Penguin Solutions supplies the CXL cards inside the CXL expansion chassis, the physical memory expansion hardware that makes large-scale pooling a deployable reality. This means AI workloads can be right-sized, reconfigured, and scaled without forklift upgrades, a critical capability as model architectures evolve at a pace that outstrips any fixed hardware assumption. Intel contributes both accelerator technology and edge-accelerated AI inference capabilities to this layer, enabling the composable compute stack to serve demanding real-time AI workloads at the point of need. Preferred Networks, whose AI research spans custom chip design through generative AI deployment, represents exactly the kind of workload this composable layer is built to serve.

03. Liquid Cooling: The Unsung Hero of the AI Era

Every watt of AI compute generates heat, and the industry’s most honest open secret is that air cooling is already failing at the rack densities the latest GPU generations demand. Edgecore’s integrated Cooling Distribution Unit (CDU), delivering 100–200 kW of in-rack heat exchange capacity with N+1 pump redundancy, real-time flow and leak telemetry, and full BMS/SDN integration, is not an accessory. It is infrastructure. With documented PUE gains of up to 20%, liquid cooling is simultaneously the most critical sustainability lever and the most direct enabler of higher-density, higher-performance AI deployment. The Edgecore Open Fabric treats thermal infrastructure as a first-class citizen, not an afterthought. Beyond the rack, 1FINITY’s optical transport and DWDM expertise extends the platform’s reach across metro and carrier networks, ensuring the all-photonics vision scales from the datacenter floor to the wide area.

Ecosystem: Stronger Together

No single company builds the future of AI infrastructure alone. The Edgecore Open Fabric is the product of a carefully assembled ecosystem whose members contribute decades of domain expertise across silicon, optics, memory, software, and systems integration. Together, they close every gap in the stack:

AMD brings GPU and adaptive compute solutions that populate the composable server layer, delivering the raw AI processing power that the Open Fabric is designed to unleash at scale.

Astera Labs delivers the PCIe and CXL connectivity silicon that makes composable, disaggregated compute architectures practical, a foundational technology for the memory-pooling capabilities of the platform’s compute layer.

Бродком supplies the high-radix switching silicon that powers Edgecore’s spine and leaf switches, the same silicon that has defined hyperscale networking for a decade, now optimized for AI fabric workloads.

InLC Technology (an Accton Group subsidiary with an industry-leading IP portfolio in intelligent optical switching) contributes in-house photonics innovation that delivers networks with 10× lower power consumption, directly supporting the platform’s energy efficiency story.

Intel contributes both accelerator technology and edge-accelerated AI inference capabilities, including demonstrated expertise in AI medical imaging applications already running on IOWN APN infrastructure today.

LIQID provides software-defined composable infrastructure technology that enables dynamic resource allocation, the software intelligence that makes the hardware’s flexibility operationally accessible.

Marvell provides the CXL controller technology at the heart of the composable compute layer, enabling high-speed, low-latency memory pooling and disaggregation across GPU and CPU nodes.

НТТ provides the IOWN APN architecture and certification framework, the backbone of the entire open photonics strategy, and a technology leadership position НТТ has nurtured for years in partnership with global standards bodies.

Penguin Solutions supplies the CXL cards inside the CXL expansion chassis, providing the physical memory expansion hardware that enables large-scale, low-latency memory pooling across compute nodes.

Preferred Networks (PFN) contributes deep AI research and vertical integration expertise spanning AI chips, computing infrastructure, and generative AI, bridging the gap between cutting-edge AI development and the high-performance network fabric it demands.

1FINITY extends the optical transport and DWDM expertise critical to multi-site, carrier-grade APN deployments, enabling the Edgecore Open Fabric to reach beyond the datacenter wall to service provider and metro network environments.

From Vision to Reality: AI Applications Enabled by IOWN Today

The most compelling proof of any infrastructure platform is not its specification sheet; it is the real-world applications it enables. The IOWN community, led by НТТ and powered by ecosystem partners, has already demonstrated a compelling portfolio of production-grade AI use cases that would simply not be possible on conventional electrical networking infrastructure:

Application	Возможности	Enabled By	Partner
Smart Remote Surgery	Ultra-low-latency surgical robotics collaboration between remote surgeon and operating room	IOWN APN: Remote Support Base	НТТ
Remote Media Production	Real-time distributed content workflows for live broadcast, deterministic sub-millisecond delivery	IOWN APN: Remote Operation Center	НТТ
Construction Operations	Live remote control of heavy excavation machinery from an off-site command center	IOWN APN: Remote Operation of Excavator	НТТ
AI Medical Diagnosis	Edge-accelerated medical imaging and heatmap visualization for radiological analysis at the point of care	IOWN APN: Chest X-Ray AI	Intel
Industrial Safety Management	Low-latency industrial safety monitoring across online factory environments	IOWN APN: Remote Operation Center	IOWN Forum
Distributed AI Infrastructure	All-photonics connectivity for distributed AI training and inference across geographically dispersed data centers	IOWN APN: Full Fabric	Edgecore / Preferred Networks

These are not demonstrations. They are deployments, real-world proof that the All-Photonics Network era is not a roadmap item for 2030. It is operating today, and it is only waiting for the infrastructure ecosystem to catch up with the architecture НТТ pioneered. At Computex 2026, Accton Group and its partners are closing that gap.

Taiwan’s APN Island: Accton Group’s Role in Building It

Коммутатор Edgecore Open Fabric carries a significance that extends beyond any single product launch. Taiwan’s national AI Island vision, a strategic initiative to position the island as a globally integrated hub of AI infrastructure, healthcare, energy, and connectivity, requires precisely the kind of next-generation network architecture that IOWN provides. High-bandwidth, ultra-low-latency, energy-efficient optical networking is not merely desirable for an APN Island. It is definitional. Intel’s edge-accelerated AI inference capabilities are already proving this in production, with medical imaging applications running live on IOWN APN infrastructure across Taiwan. Preferred Networks brings the AI research depth to push these applications further, vertically integrating AI chip design, computing infrastructure, and generative AI on top of the optical fabric.

Accton Group, with over four decades of experience designing and manufacturing network infrastructure from silicon to system and facilities serving hyperscalers, telecoms, and enterprises across the globe, is uniquely positioned to be the manufacturing and technology backbone of that vision. From design to manufacturing, from switching ASICs to optical modules, from software to CDU, Accton Group provides the cutting-edge infrastructure technology that turns national vision into operational reality.

InLC Technology, a fellow Accton Group subsidiary and in-house photonics innovator, adds a strategic dimension to this story. Its IP portfolio delivers up to 10× lower network power consumption, making Accton Group uniquely positioned to address AI infrastructure’s growing energy challenge from both the networking and optics layers simultaneously.

What to Expect at Computex 2026

Visitors to Edgecore (an Accton Company) at Computex 2026 (Booth K1309, Taipei Nangang Exhibition Center Hall 1) will have the opportunity to explore the Edgecore Open Fabric on full display, the complete rack architecture, optical switching fabric, and composable compute stack. The centrepiece of the booth is a live liquid cooling demonstration, with the CDU running in real time to show what high-density AI thermal management looks like in practice. Technical briefings with product engineers and ecosystem partners will be available throughout the show.

The All-Photonics Era is not a metaphor. It is a migration, and it is underway. Edgecore (an Accton Company) invites the global technology community to come see what infrastructure built for the speed of light and the demands of the AI age looks like.