The AI boom has sparked the largest wave of data center construction in history, but according to veteran datacenter analyst Rich Miller, the industry’s biggest challenges have shifted well beyond adding more compute.
After more than 26 years covering the evolution of data centers, Miller says today’s conversation is increasingly about power, cooling, construction, and the practical realities of deploying infrastructure at unprecedented scale. While AI has fueled enormous demand, he thinks the ability to actually deliver new capacity has become the defining challenge. We had an in-depth chat on this and a wide range of other topics around the future of datacenters on the most recent Shared Everything podcast episode (take a listen below and subscribe wherever you get your podcasts).
The first is that we are in a constrained environment. There are constraints on datacenter space, memory and compute. There are constraints on development from community resistance. In the meantime, there’s all this demand, and so we have this environment where there’s this big disconnect between the hype and aspirations of all the folks who’d like to build datacenters and the reality of having the experience and capital and the construction staff to be able to build all of these things.
I think they’ve gotten the message, but there’s a lot of communities where that’s probably too late. So there’s this interesting path to navigate between the constraints that are facing the industry, the demand that’s out there, and just the general environment for trying to build.
Those realities are also changing how investors think about the market. Miller says that while hundreds of gigawatts of new capacity have been proposed, the capital is increasingly flowing toward experienced operators with proven execution rather than newcomers with ambitious announcements.
As projects compete for limited power, construction resources, and suitable sites, he expects consolidation to play an increasingly important role as established developers acquire promising projects that lack the operational experience to reach completion.
The engineering challenges inside the datacenter are evolving just as quickly. Networking, power delivery, and cooling have become central design considerations as AI systems continue pushing rack densities higher. Rather than simply scaling existing architectures, operators are rethinking everything from facility power distribution to how efficiently data moves through increasingly complex AI infrastructure.
There’s so much focus on the network right now because everybody’s trying to move these enormous amounts of data as efficiently as possible. The place where people are looking at efficiency the most is in cooling because that’s where some of the largest challenges are.
We’re rearchitecting the power from the chip to the grid. Folks are rethinking everything to simultaneously address both scale and efficiency.
Looking ahead, Miller sees innovation happening across multiple fronts, from advanced liquid cooling and smarter use of the electric grid to virtual power plants and eventually small modular nuclear reactors. He also believes more experimental concepts, including space-based datacenters, deserve exploration, but cautions against expecting them to solve today’s infrastructure bottlenecks.
His broader message is that AI’s future won’t be determined by compute alone but will depend on the industry’s ability to solve the practical engineering and infrastructure challenges that come with building datacenters at a scale the world has never attempted before.
