Rack power requirements have jumped from under 10 kW to as much as 150 kW in just a few years, driven by AI and HPC workloads. Chris Osian, Product Manager at Starline, a brand of Legrand, explains how operators can build flexibility into their power distribution, why plug-in space is now at a premium and why 800-volt DC could be the next big shift.
As rack densities continue to increase, particularly with AI and HPC workloads, how are power distribution requirements changing for modern data centers?
Around five years ago, we weren’t seeing rack power requirements exceeding 10 kW. Over the years, they’ve evolved to around 22 kW and then 50 kW. Today, we’re deploying racks requiring 100 to 150 kW.
That represents an order-of-magnitude increase over roughly five years. Before that, the pace of growth in power requirements and rack density was much slower.
This change has been driven by the evolution of GPUs and the way they are transforming modern data centers.
We now have IT equipment at the end of the power chain that requires increased power compared to five years ago. As a result, all the infrastructure upstream, from the busway and power distribution units (PDUs) to the connection with the utility, must accommodate these requirements.
We’re seeing a dramatic change in the power distribution chain, which must now deliver significantly more power per rack, per row and across the entire white space.
As operators plan for higher power demands, what are some of the key factors or trade-offs they need to consider when selecting a power distribution system?
Before this rapid evolution of GPUs, power requirements were growing relatively slowly, which meant operators could choose fairly static systems.
You could develop your design and expect it to remain valid for the next 18–36 months, which is a typical data center project cycle. What we’re seeing today is that GPU iterations are happening much faster, creating considerable design uncertainty.
One of the key considerations when designing a data center and selecting equipment is moving away from static designs that could become outdated within a year or two.
Operators need to think about keeping pace with increasing power requirements while future-proofing their infrastructure against the unknown.
The biggest takeaway is to focus less on static equipment and more on flexible solutions. If your rack power requirements change 18 months from now, potentially while your project is still under development, you need to be able to adapt.
That could mean adopting an open-channel busway, which allows rapid installation and reconfiguration, or planning additional space to accommodate future developments.
Ultimately, it comes down to building flexibility into your design and ensuring that the technology you install doesn’t become outdated almost as soon as the data center becomes operational.
What safety considerations come with these higher power levels?
It all stems from increasing power requirements. A 10-kW rack is very different from a 150-kW rack.
As power requirements increase, incident energy can also increase, potentially raising the risks to personnel working with electrical equipment. Incident energy is an important measure when assessing electrical safety risks.
Operators should revisit their safety programs and ensure their personnel are properly trained to work with these higher power levels. They need to have appropriate safety procedures in place, including proper personal protective equipment (PPE) and training programs.
The key is to ensure that the rapid evolution of power requirements does not outpace the organization’s safety program. My advice would be for operators to review their existing safety programs and ensure they remain aligned with the equipment being deployed today.
The current environment is very different from what it was just two years ago.
How is the industry responding to these changes? What are manufacturers doing differently to help operators cope with higher power requirements?
Manufacturers are expanding their product roadmaps and we’re no different. Starline’s new T5 DoubleBar Busway delivers up to 2000A with the most pluggable space available, and our High-Capacity Plug-In Unit (HCP) distributes up to 250A directly to the highest-density racks. Again, it comes back to the pace of change.
As a vendor, we’re expanding our roadmap to ensure we’re well positioned to support the 100- and 150-kW racks being deployed today, while also preparing for future requirements of 200 or 250 kW.
We’re already hearing discussions in the market about rack power requirements potentially reaching 500 kW within the next couple of years. Our priority is to keep pace with these developments without falling behind, while continuing to support current deployments.
I believe it’s increasingly important for operators to establish strong relationships with their vendors, both upstream and downstream.
With technology evolving so quickly, businesses need to understand what solutions are available and what is coming next so they can plan accordingly. Vendors are responding by developing new products and strengthening their supply chains.
We’re also seeing businesses secure equipment and components well in advance to ensure they have sufficient supply for future projects. Above all, operators should maintain close communication with their vendors and treat them as strategic partners.
Sharing information and understanding the latest technologies available will be critical to planning infrastructure requirements over the coming years.
Beyond simply delivering more amperage, why are available plug-in space and flexibility becoming increasingly important in high-density data center environments?
Plug-in space is extremely important. To illustrate why, let’s go back to the increase in rack power requirements, from around 10 kW to more than 100 kW today.
The physical footprint of the rack hasn’t changed significantly. Previously, racks were typically around 600 millimeters, or 24 inches, wide. Today, many racks still occupy roughly the same amount of space.
However, the amount of power that needs to be delivered within that footprint has increased dramatically. That makes plug-in space a premium.
To deliver more power, you need more copper, larger circuits and potentially multiple circuits, particularly when redundancy requirements are considered.
We’re also seeing changes in rack designs, including the introduction of sidecars. Sidecars expand the overall footprint available for power distribution, which can provide some relief when accommodating higher power requirements.
Another important consideration is the increasing competition for overhead space. Previously, lower-density racks relied primarily on air cooling. Today, we’re seeing much greater adoption of liquid cooling in data centers.
That introduces additional infrastructure into the white space and makes space planning even more challenging. Everything is becoming more densely packed, so operators need to be very careful about where they position equipment.
This is why plug-in space has become so valuable and why specifying the right plug-in units for each application is essential to making the best use of the available space.
As rack-level power requirements increase, how is the way data centers deliver power from the busway to individual racks evolving?
It’s changing dramatically, and this ties back to the space issue.
We essentially have the same amount of physical space, but the power requirements have increased significantly. What we’re seeing between the busway and the rack is a growing number of circuits being routed into a single rack space, which in many cases is still around 24 inches wide.
The whips, or flexible power connections, being routed into these spaces are also getting larger because they need to carry higher currents and deliver more power.
In some cases, we’re seeing as many as eight whips, representing eight circuits, going into a single rack space. I expect that number to increase further over the coming years.
Sidecars, as I mentioned, also provide additional space for routing power connections, though they take up more room in the aisle.
Ultimately, we’re trying to deliver significantly more power within essentially the same footprint.
The number and size of circuits and power connections are only going to increase. That means operators, designers and manufacturers need to be much more strategic about how power is routed, what equipment is specified and how redundancy is designed.
All these factors need to be considered to ensure that the available space is used as efficiently as possible.
How do you see power distribution architectures continuing to evolve as AI and HPC workloads push data centers toward even higher rack densities? Are we heading toward a fundamentally different architecture?
I believe so. I expect sidecars to become more prevalent because they provide additional flexibility when routing power to the rack.
Although they take up more physical space, they also give operators a little more breathing room when accommodating higher power requirements.
I think we’ll see greater adoption of these solutions over the coming years. However, one of the biggest changes will continue to be the exponential increase in rack power requirements.
That means we’re going to see larger circuits, a greater number of circuits or a combination of both. This will create additional challenges for vendors and designers, and addressing those challenges will require considerable coordination across the industry.
I think one of the most significant developments over the next year or two could be the early adoption of 800-volt DC busway distribution directly to the rack. Instead of converting power at the rack, we could begin to see architectures that distribute 800-volt DC power directly to IT equipment.
I don’t know exactly what rack power threshold will drive that transition, but I believe it’s something we’re going to start seeing as rack densities continue to increase. Getting there will require considerable work because vendors across the entire power infrastructure ecosystem will need to adapt to this architectural change.
