The global AI boom has exposed how unprepared we really are for such rapid data center expansion, and we’ve already reached the point where construction is struggling to keep pace with the continued rate of innovation.
Nowhere is this more evident than across the US, where hyperscalers and cloud providers are racing to build out new data center campuses capable of supporting the next wave of agentic AI workloads. This is, of course, as companies continue to push the boundaries with next-gen frontier models, with both electrical supply and cooling infrastructure in hot demand.
However the hype has left utilities under pressure to make grid connections faster than ever, and contractors are facing strict and often unrealistic timelines to get facilities built and connected.
Rapid scaling is driving misalignment between tech, construction and utilities
However, Steel Tube Institute’s Dale Crawford doesn’t believe that the ongoing skills shortage is necessarily a lack of capable people. Instead, the problem lies in how quickly the sector is scaling before a shared understanding has fully developed across the workforce. In other words, the sector is expanding before companies have had time to upskill their employees.
The challenge extends far beyond AI data centers alone, though, with similar high-density electrical systems increasingly appearing in hospital, industrial facilities and food processing plants, suggesting the industry may be entering a much bigger shift in how infrastructure demands are to be met.
It’s the speed of AI growth in particular that’s really highlighted this problem, though, leaving little time to develop standardized best practices, leaving suppliers to learn in real time instead,
To better understand the AI boom’s impacts on electrical infrastructure and construction, I spoke with Steel Tube Institute Executive Director Dale Crawford about the growing expertise gap, the pressure that contractors and inspectors are facing, and why standardization and investment in people may become just as important to AI infrastructure as GPUs.
- The utility business is notoriously slow to change and often plagued by underinvestment. The AI industry is exactly the opposite, flush with cash and wanting tomorrow’s progress yesterday. Surely getting these two to work together can only end in tears?
The bigger issue isn’t incompatibility, it’s alignment at a very technical level. Projects are moving from design to installation before there’s a shared understanding of how these high-density systems are being implemented in the field.
When that shared understanding isn’t fully developed, the margin for misalignment across design, installation and inspection narrows, and that’s where challenges begin to surface.
From a steel conduit standpoint, that shows up on how raceway systems are specified versus how they’re installed and inspected under compressed timelines. Steel conduit is often selected for its durability and predictable performance, but if the team isn’t aligned on installation practices and code interpretation, even proven systems can become points of friction.
When that shared fluency isn’t there, the margin for misalignment narrows significantly. That’s where issues emerge. Because the system as a whole hasn’t developed a consistent, shared understanding at the same rate as the infrastructure is being deployed
- Can you dig deeper into the challenges contractors, inspectors and project teams are encountering on these data center projects?
The systems themselves have evolved quickly. High-density loads, redundant architectures and advanced distribution configurations have become standard in a relatively short period of time.
The challenge is not any one part of the project team. It is the speed, scale and density of these projects. Contractors are installing large raceway systems in tighter, more congested environments, designers are adapting to rapidly evolving load and redundancy requirements, and AHJs are reviewing highly complex installations on aggressive schedules.
When design intent, installation practices and inspection expectations are not aligned early, issues can surface at the handoff points.
The best way to reduce delays and rework is to build that alignment into the project from the beginning through clear specifications, proven materials, code-aligned installation practices and early communication among the project team and the AHJ.
- You mention a raft of solutions in an email you shared with me. On paper, they look great but they would take time to implement and if there’s one thing hyperscalers and the AI industry is short of, it’s definitely time.
There is a perception that standardization and education slow projects down, but in practice, the projects that stay on schedule are often the ones built around systems everyone already understands.
Well-established, code-aligned materials like steel conduit provide familiar performance characteristics and a common language across designers, contractors, inspectors and owners.
That consistency helps reduce interpretation gaps, supports a smoother review process and lowers the risk of late-stage changes or rework. In fast-moving data center construction, standardization is not a delay. It is one of the ways projects keep moving
- Part of the problem is that the current explosion in demand was not foreseen by anyone. It just happened, making it impossible to gather data sets and expertise that is often the driving force for long-term reliability of mission critical facilities. What are your views on that?
The pace and scale of demand, particularly tied to AI, accelerated beyond expectations, and the traditional pace of workforce development hasn’t kept up. That puts the industry in a position where systems are evolving faster than experience can accumulate, making continuous, structured education essential for deeper technical understanding.
From a conduit perspective, the applications themselves aren’t new, but the scale, density and integration of these systems are. At the same time, established standards and proven approaches help bridge that gap by providing a consistent framework that supports alignment even as systems evolve.
- Now, let’s be blunt. The industry needs experts and people with experience and we need them now. That will take years given the current environment. Should investment in training happen right now or could we end up with a bunch of experts twiddling thumbs after the AI bubble exploded?
This isn’t limited to data centers. The same complexity in electrical systems and the same reliance on robust, well-understood raceway solutions, such as steel conduit, are showing up in hospitals, food processing facilities and other mission-critical facilities.
This reflects a broader structural shift in electrical infrastructure, not a short-term cycle, so investing in training is about ensuring systems can be delivered safely and consistently.
The greater risk is the cost of operating without sufficient expertise in environments where performance, uptime and compliance leave very little margin for error.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.