Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

Scaling AI infrastructure is a synchronization challenge

Data centers require the right equipment, power, workforce, and approvals.

[Photo: Getty Images]

Profile photo of Glen Clark

CEO, DP World U.S./Mexico and Regional Head of Contract Logistics · DP World

AI infrastructure has entered its second act. The first phase was a land grab: secure sites, lock in power, buy chips, and build fast. The second phase is more complex—and far less forgiving. It’s about scaling across regions, timelines, and systems that don’t naturally move in sync.

The numbers are staggering. JLL estimates that nearly 100 gigawatts of new data center capacity will be added between 2026 and 2030, roughly doubling current global capacity. The sector is projected to grow at a 14% annual rate, requiring up to $3 trillion in investment.

But the issue is not just demand. It’s coordination—aligning power, equipment, permitting, workforce, and logistics across multiple sites and timelines.

A data center with servers but no grid connection is a warehouse. A site with power but no cooling is idle real estate. A facility with permits but no skilled technicians is a money pit.

Scaling AI infrastructure is becoming a synchronization challenge, not just a capacity challenge.

From single sites to global deployment

Early data center buildouts were measured site by site: secure power, acquire land, bring a facility online. That model no longer works.

The question is no longer, “Can we build this facility?” but “Can we bring 20 facilities online in the right sequence with the right equipment, power, workforce, and approvals?”

Demand is outpacing supply, with record-low vacancy and growing difficulty delivering capacity. Securing large-scale power—often 300 megawatts or more—has overtaken connectivity as the priority, while interconnection timelines in some markets now stretch for years.

Growth at that pace rewards orchestration, not fragmented execution. However, there are four areas where AI infrastructure starts to break down.

1. Logistics

      As I’ve written before, logistics is no longer a support function for digital infrastructure—it is core infrastructure.

      Each facility depends on precise delivery of high-value components—graphics processing units, racks, switchgear, cooling systems—moving across borders into tightly scheduled construction environments. A single delay can cascade across regions.

      A delayed transformer can stall an entire site. Equipment that arrives too early will sit idle, tying up millions of dollars in capital. Lead times for transformers and switchgear stretch into years, with utilities reserving production slots well in advance.

      Logistics is no longer focused on moving equipment. It is about protecting the deployment schedule.

      2. Energy

      Power is no longer just a site-selection issue. It is a sequencing issue.

      Deloitte notes that hyperscale data centers may need to act as grid partners, not just customers. Fewer than 5% of data centers currently participate in demand-response programs, where consumers voluntarily reduce power consumption during peak periods, yet pilots show 10% to 30% of electricity load can be flexed during peak demand events without disruption to data center operators.

      That means shifting workloads in data centers, not shutting down operations. Energy load can be flexed by rerouting non-urgent compute, delaying AI training, pre-cooling facilities, adjusting cooling intensity, or using on-site batteries—all without affecting uptime.

      The ability to time-shift workloads turns energy from a fixed constraint into a controllable variable. In effect, data centers are no longer just power consumers; they are becoming active participants in how the grid operates.

      3. Permitting and community alignment

      As data centers grow larger and more visible, so does community resistance. Municipalities and local residents are asking hard questions about power, water, land use, and long-term impact.

      Uptime Intelligence reported that two U.S. data center projects were canceled due to local opposition in 2023, a number that jumped to six in 2024 and 25 in 2025. So far in 2026, four more have been identified with another four expected to follow—all tied to public opposition—suggesting that 2026 could set another record.

      At the same time, the footprint for data centers is expanding. The U.S. now has more than 3,000 operational data centers and over 1,500 in development, many in rural areas that have never hosted this infrastructure, according to Pew Research Center.

      The pushback is no longer theoretical. In Fort Worth, Texas, residents recently called for a moratorium on new data centers over concerns about energy consumption, water use, and quality-of-life impacts.

      That combination of rapid expansion and rising resistance is where coordination breaks down.

      4. Workforce

      While AI is digital, the infrastructure it requires is not. AI infrastructure is built and operated by electricians, engineers, construction crews, and technicians.

      The U.S. construction industry needs 349,000 new workers in 2026 and 456,000 in 2027 to meet demand driven by workforce retirements and AI infrastructure growth.

      This is not an HR challenge. It is an infrastructure challenge.

      Projects risk falling out of sync

      When AI infrastructure fails, it doesn’t collapse. It drifts. Small delays across power timelines, equipment sequencing, and labor shifts compound into cost and schedule risk.

      Sightline Climate’s 2026 outlook illustrates how difficult it is becoming to bring AI infrastructure online at the pace demand requires.

      The firm is tracking 190 gigawatts of announced capacity across 777 data center and AI factory projects since 2024. Yet despite plans for at least 16 gigawatts to come online in 2026 across roughly 140 projects, only 5 gigawatts is currently under construction. Sightline estimates that 30% to 50% of the 2026 pipeline may not materialize this year.

      In other words, demand is not the limiting factor. Execution is.

      The issue is not individual bottlenecks. It is dependencies. A delay in power affects equipment installation. Equipment delays affect commissioning. Workforce shortages affect construction schedules. When one piece slips, the effects ripple across the system.

      How leaders can respond

      Leaders planning AI infrastructure across regions can take three actions.

      1. Design deployment as a system, not a series of sites. Every new facility should be mapped against dependencies across power, logistics, permitting, labor, and equipment. A site is not ready when land has been acquired. It is ready when the critical path is credible.
      • Build optionality early. That means alternate ports, backup carriers, regional staging hubs, prequalified suppliers, modular construction options, flexible warehousing, and energy pathways that reduce single points of failure. Optionality is not excess. It is the price of operating in a constrained environment.
      • Invest in people and partners together. The best digital tools will not compensate for a workforce that is unprepared or a partner network that is misaligned. Training, shared data, clear escalation paths, and region-specific operating playbooks will determine whether teams can execute under pressure.

      The AI conversation often focuses on models and chips. But it’s the physical layer that will determine how fast AI scales.

      Capacity may start the race, but coordination will determine who finishes.

      Glen Clark is CEO at DP World U.S./Mexico and regional head of contract logistics.