AI’s Infrastructure Future May Be Built on Its Industrial Past

AI is driving one of the largest infrastructure buildouts in modern history. Across the US, growing demand for AI capacity is placing pressure on power grids, cooling systems, and the physical infrastructure that supports the digital economy.

As a result, some of the most consequential discussions are taking place far from GPUs and training clusters. They are occurring within power markets, utility planning meetings, cooling system design reviews, and workforce development programmes.

Digital infrastructure has always depended on physical infrastructure. As AI capacity grows, the ability to build, power, cool and operate facilities at scale is attracting renewed attention. In the process, sectors that traditionally sat outside the technology industry are becoming increasingly important participants in the future of AI infrastructure.

Pressure Has Emerged Across Energy Systems

PJM Interconnection, which serves 67 million people across 13 states and manages one of the world’s largest concentrations of data centre development, has become a focal point for debates around infrastructure capacity. Wholesale electricity prices across the market rose 76% during the first quarter of 2026 compared with the previous year, while capacity costs increased by almost 400%.

The issue has attracted growing political attention. Recent reporting in the Los Angeles Times noted that federal officials have discussed whether PJM’s structure remains fit for purpose as electricity demand rises.

In New York, grid operator NYISO recently reported some of its lowest reserve margins in recent years. The organisation projects summer demand of 31,578MW against available generation capacity of 34,615MW. During prolonged heatwave conditions, demand could move close to operational limits.

Data centre development has become part of a broader debate around grid capacity, energy affordability, and how infrastructure planning keeps pace with growing demand.

Why Existing Power Assets Are Attracting New Attention

Utilities across North America are preparing for data centre-related load growth estimated at between 37GW and 66GW over the coming years.

According to planners, new data centre demand can emerge within 18 to 36 months. Many energy infrastructure projects operate on timelines measured in years rather than months. The result is a growing mismatch between the speed at which AI infrastructure is being deployed and the pace at which supporting energy systems can be planned, approved and delivered.

Historically, utilities and infrastructure providers have worked within relatively predictable demand cycles. The current wave of AI investment is compressing those timelines, requiring grid operators, energy providers, and infrastructure developers to respond to growth that is arriving faster than many planning frameworks were originally designed to accommodate.

For many operators, the question is not whether sufficient electricity can be produced. It is whether transmission capacity, substations, interconnection agreements, and supporting infrastructure can be delivered quickly enough to support planned growth.

When AI Workloads Meet Grid Reality

For much of the cloud era, connectivity, customer proximity, and land availability often shaped development decisions. Power availability now sits alongside them.

Terawulf’s Lake Mariner Data campus in New York illustrates why. Built on the site of a former coal-fired power station that ceased operations in 2019, the campus is projected to support up to 750MW of power demand at full buildout, drawing on existing industrial infrastructure and a regional electricity mix that is approximately 89% zero-carbon.

Projects such as Lake Mariner Data highlight a broader pattern emerging across the market. Operators are paying closer attention to locations where energy infrastructure already exists and where expansion can be achieved within commercially viable timeframes.

This has elevated the importance of what many operators describe as “time to power”. As Manish Kumar, Executive Vice President, Secure Power & Data Centers at Schneider Electric, recently observed, “As demand for AI infrastructure grows, time to power has become a defining constraint on growth.”

The appeal of locations such as Lake Mariner extends beyond energy infrastructure. Former industrial communities often bring a workforce with experience in power systems, maintenance, and large-scale operations.

Existing power infrastructure, industrial sites, and experienced workforces are becoming part of the same development equation. As data centre operators compete for increasingly scarce technical skills, the ability to retrain and redeploy workers from adjacent industries is becoming part of the broader infrastructure conversation.

Long-Term Energy Supply Shapes Expansion Decisions

Securing grid access is only one part of the equation. Operators are also placing greater emphasis on how long-term energy supply is sourced and managed.

BloombergNEF data shows that 20.9GW of corporate power purchase agreements were signed across the Americas in 2023, accounting for almost half of global corporate clean-energy procurement.

For many operators, access to long-term energy supply now forms part of the broader discussion around future growth capacity. Questions around energy supply are becoming more complex as operators gain a better understanding of how AI workloads behave in practice.

The way AI workloads draw power can differ significantly from traditional data centre operations. Large training environments perform highly synchronised operations across vast numbers of processors. The result is a demand profile often described as pulse-load behaviour.

According to Eaton analysis, power demand within AI facilities can fluctuate by as much as 50% of full load. In some environments, demand swings of up to 100MW can occur within “fractions of a second”.

Conventional generation assets typically respond at a much slower pace.

Utilities, energy providers, and operators are examining how these demand patterns interact with existing power systems and what additional measures may be required to maintain stability as AI infrastructure expands.

Cooling is Becoming a Strategic Capability

The challenge extends beyond electricity supply alone. Innovations that require higher rack densities have accelerated the adoption of liquid cooling.

The technical viability of liquid cooling is now widely understood. Operational execution is receiving far greater attention, with operators evaluating deployment models, maintenance requirements, retrofitting strategies, workforce training, and long-term operational consistency.

At facilities such as Lake Mariner, liquid cooling infrastructure forms part of the core design rather than being added as a supporting component. Therefore, cooling is no longer viewed as a facility management consideration. It is becoming part of the operational foundation required to support AI workloads at scale.

This reflects a broader pattern across the industry: as infrastructure becomes denser and more complex, operational expertise is becoming just as important as capital investment.

The People Problem Behind AI Growth

Electricians, cooling specialists, power engineers, HPC operators, and infrastructure technicians are now central to project delivery.

Schneider Electric’s research into emerging 800VDC architectures highlights how workforce readiness and technology adoption are becoming closely connected. The report notes that successful deployment depends on system design and training, operational experience, maintenance procedures, and supply-chain maturity.

The transition towards higher-density infrastructure introduces new technical requirements. It also introduces new skill requirements. In many cases, the availability of experienced personnel may influence project timelines just as much as equipment availability.

Therefore, the physical infrastructure required for AI growth cannot be separated from the people responsible for building and operating it. Similar questions are emerging outside the US, although the responses are beginning to diverge.

Two Regions, One Challenge, Different Responses

The pressures emerging in the US are not unique. Many of the same themes are appearing across Europe, although the response often looks different.

Across much of the US market, attention centres on securing power, reducing delivery bottlenecks, and bringing infrastructure online quickly enough to meet demand. Across Europe, there is growing interest in how data centres interact with wider energy systems.

As explored in Techerati’s recent analysis, policymakers and operators in Europe are examining how facilities can participate in flexibility markets, energy storage programmes, demand-response schemes and grid balancing initiatives.

The European Commission estimated that data centres currently account for approximately 415TWh of global electricity consumption. That figure could reach 945TWh by 2030 as AI and accelerated computing continue to expand.

Europe is also seeking significant growth in data centre capacity over the coming decade, placing greater emphasis on energy coordination, reporting frameworks, and efficiency standards.

The Infrastructure Behind the Infrastructure

The contrast between the two approaches is instructive.

Both regions are responding to the same underlying reality. AI infrastructure depends on systems that extend well beyond the walls of a data centre: the difference lies in where attention is being directed.

Power, cooling, workforce capacity, and energy procurement are often discussed separately. Across much of the industry, they are becoming part of the same planning conversation.

For years, digital infrastructure was largely discussed in terms of software, silicon and compute. AI is drawing attention back to the systems that make all three possible. It is also expanding the circle of organisations that influence how infrastructure is built and where capacity can be deployed.

Utilities, energy providers, industrial operators, and infrastructure developers are becoming more closely connected to the future of AI. In many cases, decisions about power availability, grid capacity, workforce readiness, and site development may prove just as influential as advances in hardware and models.

The next phase of AI infrastructure growth will be shaped by more than technology alone. It will also depend on how effectively physical infrastructure, energy systems, and industrial capabilities can be coordinated to support it.