The global power sector is entering a period of sustained change. According to the Electricity 2025 report from the International Energy Agency (IEA), electricity consumption is expected to grow steadily through 2027, driven by electrification across transportation, buildings and industry.

Within that broader shift, data centers stand out. Research from Goldman Sachs suggests that global data center power consumption could increase by roughly 50% by 2027 and as much as 175% by the end of the decade. Much of that growth is tied to artificial intelligence (AI), which is rapidly changing both the scale and the profile of data center loads.

What matters here is not just the magnitude of demand, but how it behaves. Data centers are no longer just large electricity consumers. In many cases, they are beginning to influence how power systems are planned, operated and even designed.

A different kind of load

The U.S. grid has always had to deal with growth. What’s different now is the type of load being added. AI-driven data centers combine several characteristics that rarely appeared together in traditional industrial loads:

• Very high-power density (GPU-intensive racks now draw 50-100 kW or more – several times the historical norm of 10-20 kW)
• Continuous, 24/7 operation
• Tight tolerances for power quality
• The ability to scale quickly

That combination changes the planning equation. In regions with heavy data center development, utilities are seeing energy use rise faster – and in more concentrated ways – than in the past.

Instead of broad, gradual load growth, demand is showing up in clusters. Transmission systems and emerging on-site generation must deliver large amounts of power into specific areas, while local distribution networks are expected to support sustained, high-capacity loads with little margin for disruption.

This is forcing a shift in how infrastructure is planned. Forecasting is more uncertain. Timelines are tighter. And in many cases, the traditional sequence of plan-permit-build is struggling to keep up.

AI: More than just demand

AI is often described as an energy-intensive technology, and that’s true as far as it goes. Training large models requires substantial computer resources, often running continuously across thousands of processors. At scale, even inference workloads that handle everyday queries add up quickly.

But this view only captures part of the picture. AI is also becoming a tool for managing the systems it puts pressure on. Inside data centers, it’s already being used to optimize workload placement, improve cooling performance and identify equipment issues before they lead to failures.

At the grid level, similar approaches are being explored for power consumption forecasting and system balancing, especially as more variable renewable generation comes online. So, while AI is clearly contributing to load growth, it is also helping operators make better use of the infrastructure they already have.

Both dynamics are happening simultaneously.

Transmission and distribution under pressure

The effects of data center growth show up differently across the grid. On the transmission side, the challenge is scale. Moving large blocks of power over long distances takes time to plan and even longer to build. Permitting, siting and cost all play a role, and lead times can stretch into years.

On the distribution side, the timeline is shorter and often more urgent. Substations, feeders and transformers in high-growth areas can approach capacity limits quickly – sometimes faster than expected.

Several issues are becoming more common:

• Localized congestion in data center hubs
• Increased sensitivity to power quality issues
• The need for faster infrastructure upgrades

One key challenge is that these pressures are not evenly distributed. They tend to be highly localized and they don’t always align with long-term planning assumptions. That makes coordination between utilities, developers and regulators more important – and more complex.

Data center capacity and utilization trends

These shifts are already influencing how equipment is specified, deployed and maintained across both utility and data center environments.

Another dimension of AI-driven load growth is how quickly data center capacity is being absorbed. Industry insights, including analysis from Goldman Sachs Research and similar firms across the sector, show that AI workloads are taking up a growing share of total capacity.

The share of total capacity is projected to grow from roughly 14% to more than a quarter of total data center power demand by 2027. Meanwhile, overall data center power consumption could rise by roughly 175% by 2030 compared with 2023 levels.

Utilization levels provide additional context. Occupancy has already pushed into the mid-90% range in key markets, leaving virtually no headroom for new high-density workloads. Under these conditions, even incremental increases in demand can quickly translate into added strain on power delivery and cooling systems.

For grid operators, this creates both urgency and uncertainty. High utilization requires new supply to sustain growth. At the same time, variability in AI adoption complicates planning timelines. In regions with concentrated data center development, small forecasting gaps can have outsized impacts on infrastructure readiness, reliability and cost.

These dynamics reinforce a central point: as utilization rises, operational intensity becomes a key factor in how the grid is planned and managed. Aligning capacity expansion with evolving demand will be critical to maintaining system stability and supporting continued growth in AI-driven workloads.

From load centers to energy nodes

At the same time, data centers themselves are changing. Many large facilities are no longer designed as purely passive loads. Instead, they are incorporating additional energy capabilities that allow them to interact with the grid in more flexible ways:

• On-site generation resources
• Long-term renewable energy agreements
• Battery storage systems
• Participation in demand response programs

In practice, this means data centers can adjust how and when they draw power. In some cases, they can reduce peak demand or shift load in response to grid conditions.

This does not eliminate the need for new infrastructure. But it does introduce a more dynamic relationship between large energy users and the grid – one that wasn’t common even a decade ago.

Rethinking power distribution inside the data center

Changes aren’t limited to how data centers interact with the grid. They are also happening inside the facility. As power densities increase, traditional low-voltage distribution approaches become harder to scale efficiently. This has led to a growing interest in medium-voltage (MV) architectures.

By distributing power at higher voltages deeper into the facility, operators can reduce losses, simplify cabling and better support high-capacity deployments. What used to be a secondary design decision is now central to how large data centers are built.

There is also increasing attention on converting more of the supply to load power architecture, using more DC vs. traditional AC systems. While still early, enabling technologies are being explored in both research and pilot environments, particularly to reduce conversion steps and align more directly with DC-based computing loads.

There are still open questions – particularly around protection coordination, standards harmonization and interoperability across equipment from different manufacturers. In North America, for example, data center MV systems are typically specified under IEEE standards (notably C37.74 for padmount switchgear). But the rapid growth of the market has attracted IEC-tested equipment that is evaluated under different testing standards.

As both standards frameworks coexist on the same sites, questions of compatibility, certification and long-term maintainability are becoming more pressing. For emerging system architectures that rely on DC, the challenges are even more fundamental: protection schemes, fault interruption methods and safety standards are still maturing.

Even so, the direction of travel is becoming clearer. As computing loads evolve, so do the expectations placed on electrical architecture – and so does the urgency of resolving these technical foundations.

The growing importance of system visibility

As both grid conditions and data center operations become more complex, visibility into electrical systems is becoming more important. Operators require more detailed, real-time insight into system performance – how power is flowing, where constraints are emerging, how equipment is operating under load and where risks may be developing. This has led to broader adoption of monitoring, sensing and analytics capabilities across electrical infrastructures.

These tools support:

• Earlier identification of potential issues
• More informed operational and maintenance decisions
• Improved alignment between facility demands and grid conditions

In practice, the demand for visibility is outpacing what many installed systems were designed to provide. Data center operators increasingly expect remote monitoring to be a standard capability, not a premium add-on. Fault detection, automated diagnostics and integration with facility-wide energy and power management systems (EPMS) are also becoming baseline requirements.

At the same time, the commissioning process itself is under pressure. The expectation for rigorous factory and site acceptance testing has intensified, even as project timelines compress. The gap between what operators need in system visibility and what the supply chain currently delivers at scale remains one of the less discussed but more consequential constraints in the market.

Taken together, these developments point toward a more integrated model in which electrical systems are expected to respond with the same level of flexibility as the digital workloads they support.

Looking ahead

Demand for data center capacity is expected to remain strong through the end of the decade. Even as new facilities are built and efficiency improves, the overall trajectory points upward.

Meeting this level of consumption will require coordinated efforts across several areas:

• Expanding and modernizing grid infrastructure
• Integrating additional low-carbon generation
• Deploying energy storage and flexible resources
• Continuing to improve efficiency at the facility level
• Addressing equipment lead times and supply chain bottlenecks that constrain the pace of deployment

AI will play a role here as well, not just in driving demand, but in helping manage it.

An evolving grid

The relationship between data centers and the power grid is changing. What was once a relatively straightforward model – centralized generation supplying large but predictable loads – is becoming more dynamic. Data centers are growing in scale, but they are also becoming more flexible, more integrated and in some cases, more active in how they interact with the grid.

The shift has implications beyond any single technology. It affects how infrastructure is planned, how power is distributed and how resilience is built into the system.

In that sense, the rise of data centers is not just a demand story. They are part of a broader transition toward a grid that is more adaptive, more interconnected and increasingly shaped by the digital systems it supports.