The rack is built around Ayar Labs' TeraPHY optical engines powered by the SuperNova remote light source, integrated directly into Wiwynn’s server architecture . This is not a simple plug-in card; it is a full re-architecture of the system from the chip up. The design incorporates an AI ASIC with 2.5D advanced packaging, combining electronic compute with optical I/O in the same package — the defining characteristic of co-packaged optics .

Key technical specifications and design features from the demonstration include:

• Scale: The optically connected topology is designed to scale to 1,024 AI accelerators and beyond, with each accelerator capable of more than 100 Tbps of optical connectivity, effectively allowing thousands of accelerators to act as a single unified system across multiple racks .
• Thermal and Power Design: The rack features a 100% liquid-cooled architecture, purpose-built to handle the thermal load of External Laser Small Form Factor Pluggable (ELSFP) light sources. It also incorporates a High-Voltage Direct Current (HVDC) power architecture to support the demands of next-generation accelerators .
• Deployment Maturity: The reference design spans from the L10 chassis to the fully integrated L11 rack, representing a complete hyperscale deployment model rather than a component-level proof of concept .

CommScope’s FastSelfClean: Cutting Installation Time by 98.5%

As compute and optical engines evolve, physical fiber deployment remains a critical bottleneck. CommScope, now an Amphenol company, addressed this at Computex with the debut of its FastSelfClean high-density fiber-optic connector technology .

The company claims the new connector reduces installation time by up to 98.5% compared to traditional field-termination methods, while also delivering ultra-low loss performance . The self-cleaning mechanism is designed to maintain signal integrity in the dense, often contaminated environments of AI data centers, where a single dirty connector can bring down significant bandwidth.

CommScope positions FastSelfClean as an enabler for faster deployment of both scale-up and scale-out networks in AI data centers, streamlining the connection of entire rack-scale fabrics . While the company has not publicly released detailed technical benchmarks for the self-cleaning mechanism itself, the claimed 98.5% time reduction aligns with its broader corporate focus on slashing deployment complexity through modular and pre-engineered solutions .

The Broader Ecosystem: A Full-Stack Optical Supply Chain

Wiwynn’s demonstration assembled eight ecosystem partners, revealing the complete CPO supply chain taking shape in real-time .

ASIC Design and Silicon Photonics Integration

GUC (Global Unichip Corporation) collaborated on the CPO-based rack design by integrating AI ASICs with 2.5D advanced packaging, working at the chip design stage to fully account for optical I/O, power delivery, and thermal management, reducing integration complexity .

Fiber Routing and Physical Connectivity

The rack’s internal fiber plant and blind-mate connections relied on:

• Corning, Molex, SENKO, and TE Connectivity: Providing Fiber Array Units (FAU), high-density optical connectors, blind-mate technology, and fiber management systems for both server-internal and rack-level cabling .

Passive Optical Components

Browave and FOCI contributed passive optical components and fiber-optic interconnect solutions, completing the end-to-end optical path from silicon photonics to rack-level connector .

The result is a demonstration that covers the complete chain: from CPO-packaged ASICs, through TeraPHY optical engines and SuperNova light sources, across managed fiber channels, to rack-level blind-mate connectors ready for data center deployment .

Nvidia’s $6.5 Billion Context: Why AI Is Going Optical

The Computex announcements do not exist in a vacuum. Since March 2026, Nvidia has committed at least $6.5 billion to photonics and CPO companies, making it the largest single investor in this technology transition . The specific deals include:

• $2 billion each into Coherent, Lumentum, and Marvell .
• Up to $3.2 billion into Corning .
• Participation in Ayar Labs' $500 million Series E funding round .

These investments, announced across just three months, represent one of the fastest strategic pivots in semiconductor history . The objective is explicit: replace copper with light-based interconnects as AI training clusters exceed the practical limits of electrical signaling on bandwidth, distance, and power . Analysts note that photonics uses light instead of electricity to move data, reducing energy consumption while handling far greater throughput .

The Computex 2026 CPO demonstrations from Wiwynn, Ayar Labs, CommScope, and their ecosystem partners are the first tangible proof of infrastructure being built specifically to absorb this wave of photonics investment, marking a transition from research papers to rack-scale production designs.

The rack is built around Ayar Labs' TeraPHY optical engines powered by the SuperNova remote light source, integrated directly into Wiwynn’s server architecture . This is not a simple plug-in card; it is a full re-architecture of the system from the chip up. The design incorporates an AI ASIC with 2.5D advanced packaging, combining electronic compute with optical I/O in the same package — the defining characteristic of co-packaged optics .

Key technical specifications and design features from the demonstration include:

• Scale: The optically connected topology is designed to scale to 1,024 AI accelerators and beyond, with each accelerator capable of more than 100 Tbps of optical connectivity, effectively allowing thousands of accelerators to act as a single unified system across multiple racks .
• Thermal and Power Design: The rack features a 100% liquid-cooled architecture, purpose-built to handle the thermal load of External Laser Small Form Factor Pluggable (ELSFP) light sources. It also incorporates a High-Voltage Direct Current (HVDC) power architecture to support the demands of next-generation accelerators .
• Deployment Maturity: The reference design spans from the L10 chassis to the fully integrated L11 rack, representing a complete hyperscale deployment model rather than a component-level proof of concept .

CommScope’s FastSelfClean: Cutting Installation Time by 98.5%

As compute and optical engines evolve, physical fiber deployment remains a critical bottleneck. CommScope, now an Amphenol company, addressed this at Computex with the debut of its FastSelfClean high-density fiber-optic connector technology .

The company claims the new connector reduces installation time by up to 98.5% compared to traditional field-termination methods, while also delivering ultra-low loss performance . The self-cleaning mechanism is designed to maintain signal integrity in the dense, often contaminated environments of AI data centers, where a single dirty connector can bring down significant bandwidth.

CommScope positions FastSelfClean as an enabler for faster deployment of both scale-up and scale-out networks in AI data centers, streamlining the connection of entire rack-scale fabrics . While the company has not publicly released detailed technical benchmarks for the self-cleaning mechanism itself, the claimed 98.5% time reduction aligns with its broader corporate focus on slashing deployment complexity through modular and pre-engineered solutions .

The Broader Ecosystem: A Full-Stack Optical Supply Chain

Wiwynn’s demonstration assembled eight ecosystem partners, revealing the complete CPO supply chain taking shape in real-time .

ASIC Design and Silicon Photonics Integration

GUC (Global Unichip Corporation) collaborated on the CPO-based rack design by integrating AI ASICs with 2.5D advanced packaging, working at the chip design stage to fully account for optical I/O, power delivery, and thermal management, reducing integration complexity .

Fiber Routing and Physical Connectivity

The rack’s internal fiber plant and blind-mate connections relied on:

• Corning, Molex, SENKO, and TE Connectivity: Providing Fiber Array Units (FAU), high-density optical connectors, blind-mate technology, and fiber management systems for both server-internal and rack-level cabling .

Passive Optical Components

Browave and FOCI contributed passive optical components and fiber-optic interconnect solutions, completing the end-to-end optical path from silicon photonics to rack-level connector .

The result is a demonstration that covers the complete chain: from CPO-packaged ASICs, through TeraPHY optical engines and SuperNova light sources, across managed fiber channels, to rack-level blind-mate connectors ready for data center deployment .

Nvidia’s $6.5 Billion Context: Why AI Is Going Optical

The Computex announcements do not exist in a vacuum. Since March 2026, Nvidia has committed at least $6.5 billion to photonics and CPO companies, making it the largest single investor in this technology transition . The specific deals include:

• $2 billion each into Coherent, Lumentum, and Marvell .
• Up to $3.2 billion into Corning .
• Participation in Ayar Labs' $500 million Series E funding round .

These investments, announced across just three months, represent one of the fastest strategic pivots in semiconductor history . The objective is explicit: replace copper with light-based interconnects as AI training clusters exceed the practical limits of electrical signaling on bandwidth, distance, and power . Analysts note that photonics uses light instead of electricity to move data, reducing energy consumption while handling far greater throughput .

The Computex 2026 CPO demonstrations from Wiwynn, Ayar Labs, CommScope, and their ecosystem partners are the first tangible proof of infrastructure being built specifically to absorb this wave of photonics investment, marking a transition from research papers to rack-scale production designs.