Surging AI workloads are forcing a fundamental rethink of rack-level power. As traditional designs give way to open compute power shelves and high-voltage DC architectures the challenge is no longer just delivering more power but doing so safely, flexibly and at scale. Calvin Nicholson, Senior Director of Product Management at Legrand, explains how infrastructure strategies are evolving to meet the demands of next-generation high-performance environments.
What fundamental shifts are you seeing in rack-level power architecture and how are operators adapting their distribution monitoring and resilience strategies?
There has been a lot of changes recently with densities rising sharply and we’re seeing this effect multiple parts of our business. Typical power distribution within the cabinet is 60 A / 400 V but now we’re seeing a fundamental shift to 125 A internationally.
To combat this radical change we’re targeting multiple PDUs in the cabinet to meet the higher workloads as well as increased adoption of open compute solutions where multiple power shelves can be stacked to achieve higher densities. Although there is a higher cost with such changes many operators are integrating open compute type solutions.
As AI workloads push traditional rack designs to their limits how is the shift towards open compute standards redefining how data centres balance extreme power density with operational flexibility?
Open compute has been around for a long time and the advantages speak for themselves. The ORv3 design eliminates the need for lots of cables and benefits include flexibility and higher efficiency on power supplies. With the GB200/300 processors driving huge demand adding multiple power shelves to reach 150 kW densities is how traditional rack designs are transforming.
When you reach high densities an increased number of power shelves are required which means total rack power consumption rises significantly. This is driving a shift to high-voltage DC which is still yet to be fully understood and perfected. In North America Meta, Google, and Microsoft all operate at plus or minus 400 V DC whereas NVIDIA operates at 0-800 V DC. While there has been a shift to open compute the high-voltage DC is part of the whole evolution.
How are teams evaluating AC versus DC distribution and how are they positioning their infrastructure to support future high-voltage deployments?
Some customers have fully embraced open compute early on whereas others are testing their shelf environment carefully – and both approaches are acceptable. Generally what matters is investing in that architecture which requires having the right cabinet, busbar, and, power shelves needed while being comfortable with safety and easing into it.
This is where Legrand can assist – our trusted advisors work with clients to understand their needs and deliver the right solutions required for their business. Our on-hand product suite allows customers to visualise test and evaluate what is best for them. For customers who aren’t yet comfortable or fully on board our mission is to help them understand the technology and review it together.
As racks incorporate more PDUs higher-current busbars and denser power layouts what new challenges emerge in physically integrating that much power into a single rack?
The busbars in a typical high-density application are around 70-80 lb. They are also located at the back of the cabinet where it’s very crowded. So it becomes a form of architecture – planning the design engineering it and then producing the solution. It’s all planned, tested, and integrated together to make sure it works from the cabinet all the way through to the power. Every setup is different whether it’s PDUs busbars power shelves or anything else; it’s about making sure your solution works in your specific data centre.
Looking ahead what should operators prioritise now to ensure their rack power infrastructure can scale with next-generation AI hardware and increasing density?
Thanks to AI, densities are going to continue to increase. Everybody loves to talk about when there’s going to be a one-megawatt rack but we’ll see. If you aren’t utilising AI you will most definitely fall behind because its value is already paying off. I believe people were a little unsure when they looked at some of the expenditures and the investments hyperscalers were making but that should now be a thing of the past and business leaders should be in full implementation stage.
I also believe high-voltage DC is going to continue to evolve with power sidecars and cooling sidecars. I mentioned 0-800 earlier but I think high-voltage DC will likely get up to 1500. So I think it’s about embracing what’s to come and being prepared for even greater power shifts.
What practical steps or best practices can operators adopt today to scale high-density rack power safely and efficiently while maintaining flexibility for the future?
I’ve been in the data centre space for many years and I’ve never seen this much change. It’s hard to keep up with let alone implement them all. The truth is; many of the regular applications are perfectly suitable for a typical operator. The demand for typical PDUs is still doing very well and the 100 A PDU is still providing a considerable amount of good value. Not everyone needs 100-150 kW. If you are monitoring and evaluating regularly and operations are smooth and successful there’s no need for concern.
On the flip side if you need to change or upgrade innovation is constant. We’re working on power shelves that will have an 18 kW supply. These will be 3U – much larger and bulkier. The busbars will continue to evolve as will these high-voltage DC applications. As aforementioned change and innovation are so rapid that evaluating your needs is crucial – this includes assessing your philosophy and what is and isn’t required.
How does Legrand’s ORv3-ready infrastructure help operators and what specific features or benefits make it well-suited for high-density AI-driven rack deployments?
Our approach is to be a trusted advisor. In some of the open compute presentations I’ve seen over the last few years I’ve noticed everyone seems to be concentrating on the hyperscalers financials insurance safety compliance and more. But the hardest part is getting somebody who will work with them share information and help in the design because it’s still very new – it’s a completely different structure and operators are wary. At Legrand we offer a trusted advisor; someone who’ll work with you inspect your applications consider the kW requirements and understand what you’re trying to achieve in terms of design. Once this is established the cabinet and busbar will be acquired along with power shelves and anything else whether that’s a typical cabinet or the AI open compute infrastructure.
