Structured Cabling Drives AI-Driven Data Centres

AI workloads are rapidly driving data centres to support ever-higher network data speeds, advancing from 400Gb/s to 800Gb/s and even 1.6Tb/s. To meet these demands, structured cabling has become the foundation for scalable, efficient, and high-performing infrastructures. The surge in data transmission requirements is pushing a shift from traditional low-density connections toward high-density, flexible cabling systems that can sustain the throughput and reliability required by AI operations. As the backbone of performance and scalability, structured cabling must now handle significantly increased workloads while ensuring maximum return on investment.

To achieve this, AI-driven networks are transitioning from conventional duplex LC connections to MPO-based connectivity. GPU-based servers typically use 8-fibre MPO-08 connectors supporting aggregated 800G data rates, as defined in IEEE 802.3df, over 8 duplex lanes (16 fibres) for both multimode (800GBASE-SR8) and single-mode (800GBASE-DR8) channels. This design effectively multiplies the number of fibres by at least four compared to traditional LC setups. When deployed in AI pods, which host vast numbers of servers, the overall fibre density can be up to eight times greater than in standard data centres - highlighting the crucial role of advanced cabling in enabling AI performance and scalability.

This transition also facilitates the already prevalent move to parallel optics infrastructures and together with the massive increase in fibre utilisation by as stated a factor of eight. This shift is necessary to accommodate the bandwidth growth required for AI, but it also introduces new challenges related to fibre density and management.

The rapid adoption of 400Gb/s and 800Gb/s network speeds requires significantly more fibre links. AI clusters rely on APC multi-fibre MPO connections for server to leaf links, and your more traditional single-mode MPO connections for leaf to spine links, which means fibre volumes have increased exponentially. Without a structured approach, data centres risk excessive cable congestion, increasing difficulty in maintenance and reducing airflow optimisation. 

AI workloads operate within clustered architectures, often requiring shorter cabling runs, with a large percentage of AI networks built in <50m SuperPods.  This means the concern over propagation latency due to the additional connection points that structured cabling introduces is not a concern, as light propagation delay over such distances (<50m per SuperPod) remains below 250 nanoseconds, which is negligible compared to switching and signal processing delays. 

The misconception that structured cabling introduces excessive latency compared to direct point-to-point cabling is being countered with strong evidence that latency from structured cabling is minimal. Moreover, most delays in AI networks arise from forward error correction (FEC) and buffering at the switch level, not from the additional fibre connectors which only introduce optical loss.  One of the concerns against structured cabling in AI networks is the added connector loss that may cause channel performance risk.  This argument should be discussed by pointing out that when working with transceivers fully compliant with Ethernet channel specs which allocate connectivity losses of 1.5 dB for MMF channels and about 2.5dB for SMF channels.  The concern can be addressed even when looking at proprietary designs, there is extensive testing that shows when keeping connector losses within the mentioned limits and observing good installation and cleaning practices with links will comply with IEEE 802.3df. Thus, a well-designed structured cabling infrastructure does not negatively impact latency-sensitive AI workloads.

Implementing Structured Cabling in AI Workloads

To maximise return on investment, optimise longevity, and ensure seamless operation, structured cabling systems must be designed with the following criteria in mind:

• Scalability and Modularity - Data centres must implement modular patch panels and high-density MPO cabling to allow for seamless upgrades as network speeds increase. A structured approach allows for better management of fibre expansion without requiring frequent overhauls.
• Optimised Cable Pathways and Management - High-density cabling can lead to congestion in pathways, negatively impacting airflow and serviceability. Structured cabling mitigates these risks by consolidating multiple fibre runs into high-count trunks. This approach significantly reduces the physical footprint of fibre pathways, with estimates showing a reduction of up to 70% in pathway utilisation when structured cabling is deployed.
• Reliability and Reduced Network Downtime - Structured cabling improves maintainability and minimises risks associated with excessive cable slack, improper bend radii's, and disorganised cable management. Implementing structured pathways ensures that connections are well-documented, labelled, and accessible, which simplifies troubleshooting and reduces mean time to repair (MTTR).
• Futureproofing with High-Density Connectivity - The transition to high-speed networking requires infrastructure that supports evolving standards. The shift towards 16-fibre MPO connectors for 800Gb/s deployments allows for network scalability while maintaining compatibility with existing 400Gb/s systems. Investing in structured cabling that accommodates future higher-density connectors ensures a seamless migration path for increasing bandwidth needs.
• Energy Efficiency and Sustainability - As AI workloads consume significant amounts of power, structured cabling can contribute to power efficiency through the adoption of multi-mode fibre. Multi-mode transceivers consume up to 15% less power than their single-mode counterparts, making them an attractive option for AI workloads operating within shorter reach distances.

Structured cabling vs point-to-point cabling in AI networks

While some organisations still rely on direct point-to-point cabling for their high-speed AI networks, this approach introduces several challenges:

• Increased Fibre Management Complexity: Point-to-point cabling can create a chaotic infrastructure with excessive slack, making moves, adds, and changes difficult.
• Limited Scalability: Expanding point-to-point networks requires additional fibre runs, which leads to congestion and inefficient space utilisation.
• Higher Operational Costs: The difficulty of maintaining and troubleshooting point-to-point systems increases operational expenses over time.

By contrast, structured cabling provides a well-organised, scalable, and maintainable network infrastructure that is better suited for AI-driven environments.

Good installation practices and complying to industry standards, such as IEEE802.3df for 800Gb/s over multimode fibre, provide guidelines that help maintain network integrity and facilitate future upgrades, while also ensure compatibility and performance consistency as well as longevity and reliability of the cabling infrastructure. Factors such as mapping and documenting the physical infrastructure are critical especially as network speeds get progressively faster. Complete and accurate records must be kept from day one to create a data base that can be used to audit current installations and are invaluable when changes need to be made.

The roadmap to 1.6Tb/s and beyond

The future of structured cabling is evolving in line with industry trends. As networking standards advance towards 1.6Tb/s and beyond, structured cabling solutions must adapt. Emerging technologies such as co-packaged optics (CPO) and next-generation small form-factor connectors, such as MDC, are expected to further optimise fibre management in AI workloads. Additionally, enhanced modulation techniques such as PAM-4 will enable higher transmission rates over existing fibre infrastructure, reducing the need for frequent cabling replacements.

AI-driven workloads demand high-speed, high-density network infrastructures, making structured cabling essential for modern data centres. The transition from LC to MPO, increasing fibre density, and the need for scalable, efficient network architectures highlight the importance of a structured approach. By implementing best practices in structured cabling, organisations can optimise network performance, minimise latency concerns, reduce pathway congestion, and ensure a future-proof infrastructure.

With AI workloads continuing to expand their effect on our daily lives, a structured cabling approach is not just beneficial, it is a necessity for ensuring seamless operation, scalability, and long-term reliability in high-performance data centres.