It’s becoming something of an annual occurrence for the DCD broadcasting team to run a special end-of-year series, which brings together industry leaders from multiple disciplines and geographies to hypothesize the trends and outlooks shaping the next twelve months of digital infrastructure.

Across almost a dozen panels covering the full breadth of the data center landscape, all 38 speakers kept returning to the same conclusion: the year ahead will test how quickly the industry can adapt to pressures that have moved from theory into operational realities.

What connected these conversations was their pragmatism rather than distant forecasting or discussion of sensationalist headlines. The speakers consistently returned to the challenges already shaping decisions today, from power to scale and from design to delivery, and the ways the industry is responding to these pressures.

Of course, AI ran through almost every conversation, not necessarily as an end goal, but as a force intensifying the constraints and challenges the industry is already facing.

Power remains the primary constraint

It will be no surprise that power sat at the center of almost every discussion. Across every region, the conversations pointed towards the rapid escalation in scale of recent projects and what this means for the year ahead.

Deployments that only a few years ago would have been considered large at 10 or 20 megawatts are giving way to projects planned in the hundreds of megawatts, with AI factories and gigawatt-campuses becoming part of mainstream planning. As a result, power availability, grid timelines, and delivery risk have become even more critical factors in decisions about where projects can be deployed and how quickly they can be brought online.

There was little indication across the episodes that power constraints will ease anytime soon. The speakers consistently pointed to grid delivery timelines stretching into 2027 and 2028, and in some cases beyond; a reality that is already reshaping how projects are planned, phased, and deployed, and in some instances determining whether they can proceed at all.

As Thor Johnson, Digital Gravity Infrastructure Partners, noted, “fundamentally, we need more power resources across Europe,” a sentiment that was echoed across all geographies.

Consequently, this will lead more and more operators to reassess their location strategies. Energy access has always been key to site selection, but in increasingly demand-heavy, power-constrained markets, it has become the deciding factor in where and how new deployments take place, forcing operators to deploy outside traditional regions. Michael Geanu, of Honeywell, summarized the sentiment well as he observed, “everyone is starting to go where they can actually get power.”

The impact of scale on delivery

Building capacity at the scale and pace that is now being demanded is becoming an increasingly nebulous problem, with new complexities being compounded at every stage of delivery. Deployment sizes are snowballing, while at the same time supply chains are under mounting pressure, resulting in longer lead times and project delays, threatening the reliability of delivery.

To overcome the execution challenges, many of the speakers described how planning is becoming more grounded in what can be procured, built, and deployed now, rather than aiming for the ‘ideal’ design, which could lead to insecure supply chains. During the North America discussion, Scott McBride, Northampton Capital Partners, commented, “the risk now isn’t demand, it’s delivery,” highlighting the shift in priorities as project sizes have grown.

However, for organizations with more available capital, proactive planning in the form of earlier procurement and the warehousing of long lead-time equipment is becoming a favored strategy. This might have once seemed to be costly, but as projects continue to scale, having greater control over access to components and their timelines is becoming essential for derisking delivery.

Workloads increasingly drive design

Whilst factors like power and supply chains are set to have a constraining effect on future data center deployments, workloads are becoming increasingly pivotal in shaping how infrastructure is designed. As AI and advanced computing continue to evolve and move into the mainstream, operators will have to ensure that their infrastructure is not only scalable, but adaptable and flexible too.

As DCD’s Stephen Worn summed it up, “the next phase of data center evolution will be driven by agility, not just capacity,” highlighting how long term success will depend on flexible and adaptive infrastructure design as operators navigate uncertainty around workload evolution and performance requirements.

Of course, not all operators will be going down the high-density route, but in a time of such rapid change and accelerating tech refresh cycles, flexibility will be essential to ensure businesses can seamlessly adopt new technologies.

Training and inference conversations diverge

As workloads become more central to infrastructure planning, the divergence between training and inference will become even more pronounced. Training workloads, with their large power demands will continue to be sited in locations with long-term energy availability, relatively unaffected by latency. On the other hand, whilst all workloads require reliable power access, inference workloads will increasingly be pushed closer to the user, driven by factors such as latency sensitivity, regulatory requirements and data sovereignty. However, this shift will likely lead to a growing concentration of inference workloads in more constrained, Edge locations, presenting new pressures around site selection and energy access.

In the Workloads and Workloads Placement episode, Owen Rogers, Uptime Institute, described how “the next phase of AI will be about intelligent placement, balancing compute with proximity to data,” highlighting how workload placement is becoming an increasingly strategic consideration, shaped by both workload evolution and market pressures.

Looking ahead, these decisions will become even more critical, with organizations needing to think tactically about how different stages of the AI lifecycle are distributed, while maintaining the ability to adapt as conditions change.

As Mark Boost from Civo said during the panel, sovereignty is increasingly being treated “as a strategic consideration, not just a compliance requirement.”

Cooling is catching up with reality

Cooling has always sat at the heart of infrastructure discussions, but over the last year, the focus has shifted from theoretical preparedness towards the practical realities of implementing liquid at scale. Whilst these technologies have been around since the 1950s, the industry is only now reaching the tipping point for mainstream integration, a challenge that comes with not just technical considerations, but also significant operational and design implications, as businesses look to deploy liquid cooling at pace.

As more and more organizations adopt their own AI engines and models, cooling will become an increasingly defining factor in supporting these high-density workloads. As Dustin Demetriou of ASHRAE remarked, “we’re now at the point where cooling decisions have to be made alongside compute,” reiterating the necessity of prioritizing next-generation thermal management technologies in line with AI adoption.

What this means is further investment into liquid cooling for both new builds and for retrofits. However, whilst adoption will continue to accelerate, we can expect an ongoing period of transition, particularly as a single dominant approach to thermal management is yet to emerge. As several speakers made clear during the series, the greater risk is not choosing the wrong cooling solution, but delaying cooling infrastructure decisions as densities rise.

So where are we going?

If there was one consistent theme woven throughout the Trends and Outlooks Series, it is that the industry has moved from theory to execution. AI is here, yet the headwinds of power availability will set the pace of how quickly, and how much, capacity the digital infrastructure sector can bring online. It’s a very real constraint, but so too is the response of the industry.

Of course, there are no silver bullets or shortcuts around the challenges of scaling deployments, negotiating supply chains, or finding access to clean, reliable power, but the sense from the series was that these issues can be navigated. From integrating liquid cooling and rethinking workload placement, to adopting more flexible infrastructure, it is clear the industry is readying for the complex demands to come.

This summary can only capture part of what was explored across the eleven episodes of the Trends and Outlooks Series. Most of the nuance and regional variation sits in the full discussions (and they are well worth watching in full!).