Macro forces shaping the future of the data center—from extreme densification to gigawatt‑scale capacity—are giving rise to distinct technology trends. Energy autonomy is accelerating to support that scale, while densification is driving the adoption of liquid cooling and new power architectures.
While these forces and trends, explored in the latest Vertiv Frontiers report, will shape future data center design, build, and operations, near term market dynamics are even more tangible.
Martin Olsen, VP for segment strategy and deployment, data centers at Vertiv and TIRIAS Research founder and principal analyst Jim McGregor recently discussed the trends they see shaping data centers over the next 12 months with Data Center Dynamics (DCD). 2026 Trends and Outlooks: AI and high-density computing - DCD
DCD: I’d like to begin by discussing the high-level themes and trends you are seeing right now.
Jim McGregor, TIRIAS (JM): Well, you know, the big thing is we tend to think of AI as mature, and it's not. We're still learning. We're still growing. And when we think of AI demand, it still is insatiable. I should say we have a phenomenal amount of opportunity as we continue to advance AI.
Martin Olsen, Vertiv (MO): I've seen a tremendous shift just over the past three or four years. I think we all still remember about maybe five years ago now, we were at 20, 30, maybe 35 kilowatts per rack. Which we all thought of as being highly dense, and that was something typically reserved for the hyperscalers, perhaps some of the larger wholesale colocation providers. But we fast forward, and it has just really accumulated in just size and scope. We've gone from that to very quickly hit 80, and then we hit 130 kilowatts. We're now operating at 140 kilowatts and NVIDIA just announced their Vera Rubin generation that will take us straight into 240, 250 kilowatts
DCD: How mainstream are these high rack densities and what are some of the most common next steps that you're seeing today, as far as density goes?
JM Well, the densification of computing is coming from all aspects and impacting the entire industry, from mobile devices all the way up to the cloud. We really are in this new era where Moore's law, goes from millimeters squared to millimeters cubed. How do we pack all this stuff tighter and tighter? You know, it's kind of funny, because two years ago, we were talking about maybe a five-kilowatt module that had the GPU, CPU, the memory and the network. It quickly, within a year, jumps to 35 kilowatts per module, and now we're probably talking 100 kilowatts per module. It is a huge challenge, and the entire industry is looking at ways to really support this. You know, immersion cooling is one of the solutions. Quite honestly, I don't think that's going to be it. I still think direct to chip is better. I think we just go from using water-based solutions to using chemical-based solutions, like liquid nitrogen or something like that. But the industry is looking at all options, and all options are on the table, even switching networking from using copper to using fiber. I mean, that has a huge impact on thermals and speed and density. We're trying to fit all these things in at a pace we've never seen before.
DCD: There’s a lot of conversations currently happening about how the industry is moving from processor focused design to full stack thinking. From your perspective, what does this shift really mean?
MO: This shift from individual processes to full stack is not new. You buy a car, buy a phone, and it’s full stack because you get everything, and that's really the best way to think about it. When you get a phone, you don’t have to go and buy the board, the case and the glass and start putting it together. It is a full stack, right? And that's how we were thinking about it in the data center side as well. And that's how it's being thought of on both sides, meaning GPUs, network, storage as well as on the power and cooling infrastructure. In general terms, a GPU processor, in and of itself, is interesting, but without the memory, without the connectivity, without the software and applications it's just a piece of electronics. And that's where someone like NVIDIA, that we work closely with, is a big, important partner of ours in this space. They've obviously done a very good job. They have really built that full stack, or are at least getting very, very close to it. We've focused very intently on taking a lot of cues from NVIDIA. We have really built our stack out to where we have designed both the power train and the thermal chain together. And when we say train and chain, we mean the entire thing, from grid to chip and chip to heat reuse.
DCD: What are some of the most pressing constraints that are slowing AI scaling today? But also, when we talk about cooling or power innovations, are there any that are closest to keeping up with the scaling for AI over the next few years?
MO If we look at it more broadly, in terms of all these multiple gigawatts of AI factories, it's getting the power that is in the grid somewhere to the right place which is a challenge. So that's the distribution, the transmission side of it which is being worked on by the sort of establishment. But we also have some new players that are coming in, companies such as Oklo, for instance, or other kinds of small modular reactors (SMRs). Nuclear powers is a way of scaling that. These are not multi gigawatt nuclear reactors. These are smaller ones that deliver hundreds of megawatts, which is very significant. There's a little, there's a runway in maturing, however. We look at the commercial readiness and the technical readiness of the technology. Technical Readiness Levels (TRL) is a NASA process for looking at maturing technology. Nuclear is one of those interesting ones where, from a technological standpoint, the technological readiness is far ahead. We use them in submarines. We use them in a lot of different applications, primarily government or military types of applications. But from commercial readiness, there is a potential lag to be overcome due to regulatory issues.
This is an excerpt from a DCD webcast. The full session is available via DCD. 2026 Trends and Outlooks: AI and high-density computing - DCD
Vertiv Frontiers provides more insights into the macro forces and trends shaping the data center.
