Today’s landscape is overwhelmingly characterized by change. For those who have witnessed significant transformations come and go across the industry, the rise of AI has brought about an unprecedented level of change in just a few short years.

And it’s not just the huge number of developments – but the rapid pace of transformation that defines the AI era. Upgrades that once took place over multiple years are now being deployed at breakneck speed in an attempt to accelerate AI adoption and unlock its full potential.

Against that backdrop, in a recent DCD>Broadcast, Alastair Waite, head of data center market development at CommScope, explores power, density, and global considerations for physical layer infrastructure in the AI era.

Learning from experience

A key takeaway from the early AI adopters has revealed that it’s essential to avoid becoming boxed in by a limited infrastructure. Anticipating future developments is now mission-critical, and prioritizing flexibility for future applications is an essential requirement.

“For example, as we move toward higher GPU densities, the way we cool that technology is also changing,” explains Waite. “We're seeing the introduction of liquid into the data center, which has never traditionally been a good combination alongside electricity, but now, the industry is finding ways to make them work together.”

With increasing power densities comes larger cabinets that need to be ready to accept new cooling infrastructure, such as larger pipes required to carry fluid to and from the racks.

The knock-on effect is less space inside the cabinet for other supporting infrastructure, like cabling, and any patching performed within the cabinets using traditional methods from the rear becomes difficult. A new approach utilizing front access patching configurations must be adopted.

As well as developing solutions for front access, CommScope is addressing the need for external cable management and faster installation via solutions such as suspended cable trays.

“Because of new technologies being deployed in AI-ready facilities, CommScope is listening to our customers’ challenges to adapt the way we look at how best to build structured cabling for AI,” says Waite. “We’re paying close attention to how our customers, and the industry as a whole, are deploying AI, and we’re responding.”

“Cable management is being impacted in a big way,” adds Waite. “The trend is for more and more compute to be deployed in less and less space – and so, when it comes to connectivity, having higher connector densities in terms of numbers of fibers, is going to be highly desirable.”

As operators race against the clock to build AI-ready infrastructure, reducing complexity and speeding up time to deployment is another key piece of the puzzle. In response, CommScope is working to reduce the ‘number of clicks’ that take place at installation.

That means designing new platforms and solutions that either reduce the number of individual connectors plugged into a panel, or simplifies the installation process as a whole. The result is less time spent on site connecting structured cabling into patch panels, and a quicker turn up time to revenue for the customer.

Speed to market

As the technology develops in leaps and bounds, the infrastructure that supports it must scale up at a similar rate.

“What we're seeing is a number of customers adopting a rack-and-roll approach to installation” says Waite. “In other words, building a backbone infrastructure on site, ahead of time so that it’s ready and tested, then having servers, and network racks built off site, and rolled into the data center to be hooked up to that backbone.

“For some customers it’s a very attractive way to deploy new AI compute nodes, because it's a more efficient use of human resources, and can require less time spent at the data center’s location.”

Another advantage of that approach is consistent build quality. When a compute rack is built off site, it can be tested in advance, providing a certain level of guaranteed performance when it leaves the factory floor. However, it’s not all plain sailing:

“There’s a risk of infrastructure mismatch,” says Waite. “You have to ensure that the contractor building the equipment off site is adhering to the same fiber polarity scheme as the infrastructure that’s been built on site. So, even though rack-and-roll can be a quicker method to use, there still needs to be a lot of careful planning involved.”

Transportation is another potential hurdle for the rack-and-roll approach. Newer, powerful, and dense racks are larger and heavier, which can mean higher transportation costs and risk of damage to sensitive components during transit. AI is altering the very structure of the cabinet – and in turn defining how the industry must adapt at every stage of the supply chain.

Rules and regulations

Sovereign AI is an increasingly hot topic across the industry, driving conversations about where and how data centers are built and operated. Wider political and social concerns are directly impacting the future of operations, and key geographies are developing their own legislation in an attempt to protect data autonomy. Waite shares his thoughts:

“We’re seeing governments and enterprises reevaluating where they're going to put their data, especially when it comes to AI. And so we're witnessing a move towards more localized instances – especially in Europe and Asia. Moving builds away from the traditional data center markets can potentially place a strain on the whole supply chain, including skilled human resources.”

What were once considered Tier 2 or emerging markets, are quickly outgrowing those labels with rapid growth that’s rivaling the traditional strongholds.

“For example, Johor in Malaysia has seen growth,” says Waite. “Locations like Madrid, Milan, and new areas in the US are taking off too, because there's great power availability in those regions, there is an existing and stable WAN, plus the necessary skilled individuals already live there.”

Within that context, CommScope’s global footprint is a key advantage – enabling a broad reach across markets, while maintaining a local presence when it matters most. The ability to meet customers where they’re at with a large ecosystem of partners who understand the local market, drives speed and quality for data center operators.

What’s next?

The shift toward sovereign AI presents a huge opportunity for telecommunications operators. For government organizations looking to shore data within national borders, the natural go-to for building those parameters is local telco companies with the necessary long-haul connections to other networks, plus the design and installation expertise required.

“The other area where telcos will have a huge opportunity is AI inference – the activity of trying to solve problems and answer questions,” adds Waite. “Due to the sheer size of queries and the computational power required for inference, it's going to have to be built on a low latency and highly scalable network. Distributed to multiple Edge locations, those low-latency networks can both connect to the end user and return data to a central location for model fine-tuning.

“When interacting with an AI agent and seeking natural language, any form of latency is not tolerable to an end-user. So, the Edge is going to play a big part in supporting AI inference, and if you look at who has the most locations that could potentially host those low latency inference instances, it’s typically the service providers.”

Looking ahead, Waite shares some final advice for operators looking to keep pace in the AI era: “Design for the future, and don't get boxed into one technology. Try and design your backbone and all your equipment for the next generation of what's coming along. Consider how you shape your data center architecture for the complete life cycle of AI, not just the training aspect but also the inference.”

In the data center space, anything is possible. With quantum computing no longer a far-off dream, and as the technology develops faster and further, the demand for quality structured cabling systems to support AI is rising, too. As Waite iterated from the start, it takes solid foundations and all-round quality to fuel the AI machine.

To hear more from Alastair Waite on power, density, and global considerations for physical layer infrastructure in the AI era, watch the full DCD>Broadcast episode on demand, here.