How AI and Data Centers Are Driving Load Volatility

For decades, electric load growth followed a familiar rhythm. As an industry, we built forecasting models around gradual increases, predictable peaks, and long planning horizons. That framework served us well for a long time.

But that era is ending.

The rapid rise of artificial intelligence, and the data centers required to power it, is introducing a fundamentally different kind of load behavior—fast, concentrated, and far less predictable. What’s becoming increasingly clear is that the question is no longer whether AI-driven data centers will impact the grid, but how that impact will show up in load volatility—and how we respond.

The New Load Profile: Large, Fast, and Spiky

AI workloads are fundamentally different from traditional commercial or industrial demand. Training large language models, running inference at scale, and supporting real-time AI-enabled applications requires massive computing power. That power is often deployed in hyperscale data centers that can rival small cities in electricity demand.

What makes this challenging isn’t just the size of the load; it’s the behavior.

• Rapid ramping: AI workloads can spin up and down quickly, driving sharp changes in demand over minutes or hours.
• High coincidence: Data centers tend to cluster geographically, concentrating demand and creating localized stress on transmission and distribution assets.
• Uncertain utilization: Many facilities are built ahead of actual demand. As planners, we’re asked to account for load that may or may not materialize at full scale.

This combination turns traditional load forecasting on its head. Instead of slow, incremental growth, we’re facing steep ramps and new forms of volatility that legacy planning models were never designed to handle.

Why AI Makes Volatility Worse Than “Normal” Data Center Growth

Data centers themselves are not new. What’s new is the computational intensity and operational profile of AI.

AI training runs can consume enormous amounts of power for short, intense periods. Inference workloads are steadier, but they scale rapidly as AI becomes embedded in consumer products, enterprise software, and even grid operations.

Together, these patterns introduce:

• Higher peak-to-average ratios
• Less correlation with traditional time-of-use patterns
• Greater sensitivity to market prices, cooling requirements, and algorithmic scheduling decisions

In many ways, AI-driven demand behaves less like a factory and more like a financial market: responsive, optimized, and inherently volatile.

Infrastructure Risk: Overbuilding vs. Underpreparing

When faced with rising data center demand, the instinct is often to build: new substations, upgraded transmission lines, accelerated generation interconnections.

But history offers a cautionary tale. During the early days of internet expansion, companies massively overbuilt fiber optic networks in anticipation of explosive growth. Capacity far outpaced actual usage, stranding assets for years.

We face a similar risk today if we plan exclusively around worst-case AI load scenarios without considering flexibility.

Overbuilding carries real consequences:

• Higher costs for ratepayers
• Long permitting and construction timelines
• Reduced agility as technology and demand patterns evolve

At the same time, underpreparing risks reliability failures, interconnection backlogs, and missed economic development opportunities.

From my perspective, the challenge isn’t choosing between growth and caution—it’s finding the balance. And that balance depends on better data and smarter modeling.

Load Volatility Is a Planning Challenge, Not Just a Supply Challenge

AI and data centers are forcing us to confront a core truth: volatility cannot be solved with generation and traditional planning methods alone.

We need a more dynamic planning approach that integrates:

• Granular load modeling that captures ramp rates, coincidence, and uncertainty
• Scenario-based forecasting rather than single-point estimates
• Flexible resources such as demand response, energy storage, localized generation, and managed load agreements

In conversations I’ve had with utilities and data center operators, I’ve seen growing openness to creative solutions. Some operators are willing to curtail or shift load in exchange for faster interconnection or more favorable rates. Others are investing in behind-the-meter generation and storage that can meaningfully reduce grid impacts—if utilities have visibility into how those assets will operate.

Without high-resolution data and models that reflect real-world behavior, those opportunities are easy to miss.

The Role of AI in Solving the Problem It Creates

There’s an irony here. The same AI technologies driving load volatility can also help us manage it. 

Advanced analytics and AI-driven modeling can:

• Improve short- and long-term load forecasts
• Identify where volatility poses the greatest system risk
• Optimize the placement and operation of flexible resources
• Stress-test infrastructure plans against a wide range of future scenarios

This shifts planning from a static, periodic exercise to a continuous, adaptive process—one that’s far better suited to an energy system shaped by rapid technological change.

What Utilities Should Be Asking Now

As AI-driven data center growth accelerates, I believe utilities need to reframe the questions they’re asking:

• Where is load volatility most likely to emerge on our system?
• How much of projected demand is firm versus speculative?
• What flexibility options exist before we commit to long-lived infrastructure?
• Do our models reflect how modern loads actually behave?

Answering these questions requires tighter integration between planning, operations, and data, along with tools that can evolve as quickly as the technologies reshaping demand.

At Resource Innovations, we’ve built our Grid360 Grid Impact Assessment System to analyze existing and emerging loads (including AI-driven data centers) alongside flexibility options to identify the most cost-effective, reliable ways to accommodate change.

Preparing for a Volatile Future

AI and data centers are not a passing trend. They represent a structural shift in how electricity is consumed and valued. Load volatility is a natural consequence of that shift, and one of the defining grid challenges of the next decade.

Utilities that succeed won’t treat volatility as an anomaly to smooth out. They’ll treat it as a reality to plan for. With the right data, models, platforms, and strategies, we can support AI-driven growth without overbuilding, compromising reliability, or losing sight of affordability.

The future load is volatile. But with the right approach, planning for it doesn’t have to be.

In Part 2, our VP of Grid Edge, Chris Porter, explores what this means from a load flexibility perspective—and how utilities can turn volatility into a strategic advantage rather than a liability.