The Cabling Solution for Today’s Datacenters

Much like pipes carry gas or oil, cables carry data, the fuel for computing.

The kinds of cables that data and communication equipment need depend on various factors, including the type of traffic they transport and the distances they travel. For instance, cables for data storage route traffic from computers to storage switches to storage units, while cables for network traffic route data from computers to network switches to routers. Telecommunications traffic moves from cell towers or cable boxes to central offices. Each will require different kinds of cable.

Distances matter, too—cabling can run over a very short distance, within the same server rack, or between different racks or rooms in a building. Still, other cabling needs to be campus-wide and run for miles.

Over the years, cabling decisions have become increasingly important due to one overarching development: the rapid growth of datacenters.

The rise of datacenters

The AI revolution runs on high-speed datacenters, which serve as the computing backbone for applications. With the explosive growth of AI, an attendant growth in the demand for datacenters is being seen. The U.S. alone will register a 10% increase in demand for datacenters, at least until 2030, according to McKinsey. Additionally, mostly due to the growing demand from AI servers, datacenter capital expenditures grew by nearly 50% in the second quarter of 2024 alone, according to a report from the Dell’Oro Group.

To meet the insatiable appetite for high-speed computing, a growth in the number of datacenters alone is no longer enough. The pressure is also on for better, faster, and higher throughput from high-performance servers, both to carry information within datacenters and between them over interconnects. While 100 G networks used to be the gold standard, 400 G deployments are now becoming more routine, with the IIoT, cloud computing, and AI spurring its adoption. Another datacenter development to monitor is the growing clamor for decreased energy consumption. This means the data transfer speed needs to increase and be more energy efficient.

What does this need for bigger, better, faster, and more energy-efficient computing mean for datacenter cabling? At its most basic, cables need to transport data quickly, have low latency, and do so without the loss of data packets or guzzling too much power. The cable also needs to do this job without generating too much heat, as cooling also requires energy.

While a datacenter has dozens upon dozens of types of equipment, including networking, cooling, storage, and power systems, for the purposes of this article, the focus will be on cabling to the hardware components of a typical datacenter rack. These might include switches, which act like a traffic controller, and transceivers, which convert the data from one system to another.

Cables for today’s datacenters

Three types of cables are commonly used for high-volume communication, such as 10 Gbps capacity, or the most modern, 400 Gbps. A typical home internet connection is less than 1 Gbps.

The CAT6 cable: Common in computer networking to carry Ethernet frames, the CAT6 cable uses RJ45 connectors. To connect to switch equipment, it uses an RJ45 transceiver to convert from the switch to RJ45-compatible signals and back again at the other end. Its latency is about 2.6 ns, and it can run for about 100 m. The transceiver adds power consumption of about 4 W.

Optical fiber: Common in video and audio communication, optical fiber also finds applications in networking and data. It uses optical connectors and needs a transceiver to convert electricity to light and then back to electricity. Once converted to light, the optical fiber’s latency is about 0.1 ns, and it can run for hundreds of meters. However, it’s very picky; the optical fiber contains glass or plastic that doesn’t like to bend, and if the end picks up a speck of dust, the capacity drops. Also, it’s expensive, especially when adding the optical transceiver, which increases power consumption by about 4 W.

Direct-attached copper (DAC): The DAC is the simplest and most forgiving cabling option. Made with copper leads, it’s best suited for short-distance applications—like components within the same rack. The DAC is inexpensive and flexible and can be used without transceivers when connecting compatible equipment, but it’s only good for a few meters. Also, the DAC should not be run too close to power supplies, large batteries, or magnets because it can suffer from interference.

DACs come in passive and active varieties. The passive DAC has no transceivers, and since transmission is passive, it transfers the original signal as is. The lack of transceivers helps keep power consumption to a minimum.

An active DAC has woven-in transceivers that also compensate for potential signal loss, making it a safer bet for long-distance applications in a datacenter. The addition of electronic elements like transceivers increases the active DAC’s power usage by a bit, typically about 1 W.

The advantages of DAC for datacenters

In a datacenter, latency—the time it takes for data to move from one source to another—needs to be as short as possible. Many time-critical applications, such as autonomous mobile robots (AMRs) in warehouses or day trading in finance, all run on split-second decisions. The DAC’s most significant advantage is that it has such low latency. This critical feature of the DAC is a direct result of its simplicity. It doesn’t have any complex intermediary components that the data must pass through, making designs less complex and easier to maintain.

The DAC is also an affordable cabling option, and the passive DAC, in particular, consumes very little power. The biggest limitation is the length over which these cables can operate without too much signal degradation, usually around a few meters. Not the most efficient for long-distance data transmission, the DAC is best suited for short-range connections within the same rack or between racks. Its ability to bend makes it a particularly good fit for dense interconnections that need to pass between each other and around tight corners.

The 3M 9V4 series 400G QSFP-DD DAC cable assemblies (Figure 1), utilize 3M twin axial cable technology to create a flexible, foldable, high-performance solution. Particularly noteworthy is the QSFP-DD (Quad Small Form-Factor Pluggable Double Density) form factor, a hardware standard that facilitates faster connections. SFP means the cable is of a standard shape and size that plugs into networking equipment; the “quad” indicates the four channels of data the cable can support; and double density enables twice the amount of data to flow through a connector of the same physical size.

Figure 1: The 3M 9V4 series 400G QSFP-DD DAC cable assemblies are especially valuable for low-latency short-range connections within the same rack or between racks in datacenters. (Image source: 3M)

The net result is that DAC cables like 3M 9V4 series 400 G QSFP-DD are the best in class to accommodate bandwidths up to 400 Gbps to connect servers, switches, storage, and other high-speed equipment.

Cabling design considerations for the DAC in datacenters

Given that the passive DAC is the most economical and low-latency fit for datacenters, it’s worth considering how it integrates into datacenter infrastructure racks.

A few key factors to consider include:

• Compatibility with hardware: Given that cables need to connect to transceivers, switches, routers, and more, it’s important to ensure that selections are compatible with existing systems and will be adaptable to future iterations. The 3M 9V4 series 400G QSFP-DD is compatible with most modern-day equipment. In case datacenters need to split a high-capacity port into multiple lower-capacity connections (like four 100 Gbps or eight 50 Gbps connections from a 400 Gbps one), the series also comes with breakout cable assemblies.
• Preserving data signals: The design for DAC must consider that the cables are particularly susceptible to electromagnetic interference (EMI), especially from power cords and cables. Therefore, the DAC data cables must be clearly separated from the power equivalents.
• Easy servicing access: The placement of the cables should facilitate easy access for maintenance technicians. Overhead cabling, where the DAC would cascade in from the room’s roof, is usually considered a better option for access because the cable doesn’t need to be too long or overly contorted for the interconnects.
• Efficient ventilation and cooling: Tech stacks emit a lot of heat, and ventilation plans must be factored into the management of DAC cabling. This might affect equipment density and related cabling requirements.
• Scalability: Tech stacks change, and DAC cabling must be able to adapt to such changes. Grouping cables and efficiently labeling and bundling them helps technicians manage entire components together instead of having to sort out each one individually.

Conclusion

As computing evolves to make room for edge AI, more virtualization, and hyper-converged environments, expect the needs for related hardware equipment to change as well.

In the future, it is likely that there will be more use of machine learning hardware, edge datacenters, and distributed infrastructure. Hardware with advanced security and sustainability features is also not too far away. Through it all, the DAC will likely continue to be the cable of choice, especially in short interconnects in tech racks. Its laser-fast latency and overall cost economics can’t be beat. As a result, the DAC will find continued utility in the datacenter and beyond.