AI data center energy demand is changing everything for power traders, not only because of the scale and speed with which it’s hitting the grid, but because of the locations of certain data center clusters, their “always on” load profiles, and more.
And while data center load growth is here to stay, major uncertainties remain:
- How many proposed data centers will actually make it through large load interconnection processes?
- How many will bring their own generation?
- What will mounting community and legislative resistance do to the pipeline of proposed projects?
- Will AI data centers continue to operate near full capacity 24/7/365, or is more load flexibility coming?
How traders think about these developments can make or break their strategy in the data center era. No one has perfect foresight, but leading power traders are doing everything they can to stay ahead of the changes.
Here are Amperon’s top 5 considerations for power traders in the era of AI data centers:
- Data center load size
- Data center locations
- Data center load profiles
- Data center load flexibility
- Managing data center uncertainties
The speed and scale of data center loads
It’s no secret that data center loads are massive. Meta’s Hyperion campus in Louisiana is slated at 5 GW, supported by 10 power plants. That’s roughly half the energy demand of the entire Entergy Louisiana service territory. One large data center project can transform the landscape for an entire utility.
Even more concerning is that data centers tend to be clustered in certain areas. More than 70% of global internet traffic already flows through Northern Virginia’s Data Center Alley, and the projections keep accelerating. PJM capacity auction prices have already skyrocketed in recent years, and Dominion data center load growth projections are now 400% higher than they were just a few years ago. Virginia’s entire load is projected to double by 2035, driven mainly by just three counties.
And Northern Virginia is not the only cluster. Google’s Columbus Cluster, the largest AI supercomputer on Earth as of 2026, is just a mile from Meta’s Columbus site. In Germany, more than 100 of the country’s roughly 500 data centers are in or around Frankfurt, effectively selling out grid capacity in the region for years to come. In London, some housing projects were told they’d need to wait until 2037 for a grid connection.
Data center locations and why they break the status quo
Data centers don’t show up in traditional load centers. In MISO, for example, the South has historically dominated load growth figures, but today, states like North Dakota, Minnesota, and Wisconsin are driving far higher growth in the North and Central regions.
Data center developers are looking for cheap land and cheap power, and those tend to be far from population centers. Moreover, cold climates help with data center cooling, which attracts them to the North. The geography of future growth will not resemble the past, and power traders must be aware of changing zone-level dynamics.
How data centers alter grid-level load profiles
Thanks to the “always on” nature of AI data centers, baseload is growing faster than peak load in many regions. This is a relatively new development for experienced market participants. In fact, some prominent analysts claimed in recent years that baseload is dead. Suddenly, AI is changing all that.
Baseload power demand is the lowest level of continuously expected demand, while peak load is the highest demand that a grid operator has to serve in a given period. In recent decades, grid planners have spent much of their time worrying about how to serve peak load. Now they also have to worry about baseload, which could mean nuclear and coal plant retirements will continue to be delayed, even at great expense.
AI data centers run at roughly 90% load factors, meaning they draw near-maximum power around the clock, all year round. In contrast, most industrial loads operate at roughly 60-80% load factors. Importantly, cryptocurrency miners have high load factors, yet they can be highly responsive to high prices or coincident peak periods. AI workloads are structurally less flexible, making them much more difficult to integrate into the grid.
Of course, all this assumes data centers will continue to be grid-connected. Hyperscalers, however, are increasingly building their own generation to bypass grid interconnection queues and get projects online more quickly. This is one of the many uncertainties surrounding data centers for power traders, as discussed further below.
Demand response and other step changes in data center load
Load flexibility is one of the hottest topics in the world of data centers. Data centers typically don’t “ramp” up or down the way large industrial loads to. Their demand changes tend to be much more rapid “step changes,” which can cause major headaches for grid operators and neighboring loads. Problems can arise in a few different ways:
- Data center planned curtailments: Many grid operators are now forcing or incentivizing data centers to flex load during critical hours, as has been recommended by key scholars. The resulting curtailments can create a step change in load that forecasts currently have no way of precisely modeling, although AI-driven forecasts continuously incorporate such changes into model retraining in a way that purely statistical methods can't.
- Data center missed curtailments: This threat is more theoretical than practical as of 2026, but cautionary tales abound about data centers refusing to curtail, even when ordered to do so. Hyperscalers’ profits can dwarf missed curtailment fees, so operators may simply refuse to interrupt their workloads. Time will tell if this threat has teeth.
- Data center unexpected outages: There are various reasons that data centers can unexpectedly trip offline, and considering the size of the load, this can have cascading effects on the grid. The most notable examples come from data centers sensing voltage instability on the grid and switching to backup power to protect their sensitive electronics, creating further voltage instability on the grid that could spiral out of control.
Managing data center uncertainties
Despite the obvious and growing impact of data centers on the grid, many uncertainties persist. There are three main layers of uncertainty surrounding data center developments:
- The grid interconnection queue: Power producers are familiar with the ever-growing queues of projects looking to connect to the grid. Now the same can be said for large power users. Data centers in congested areas may have to wait 4 to 7 years to receive grid power. Wait times like this can kill a project entirely, creating uncertainty about how many planned developments will reach fruition.
- Bring-your-own-generation: To bypass interconnection queues, many data centers are now building their own behind-the-meter power plants, from unsanctioned gas engines to nuclear plant re-starts. How far will this trend go? Will these plants provide excess capacity to neighboring customers, or will they remain walled off? Everyone is watching to find out.
- Local and state resistance: A data center development backlash is rapidly mounting from local communities, state legislators, and even governors. Whether data centers actually push rates higher for other customers is a nuanced question, but the perception of harm is real. The same is true for data centers’ impact on water supplies and other issues. As of July 2026, 100+ US localities have enacted restrictions or permanent bans on new data centers, including one statewide ban, and the trend is accelerating. Will this stop projects, or will they simply go elsewhere?
Ultimately, trading strategies depend not only on the answers to these questions, but on the risk profile of each firm. Some trading desks may anchor their positions on a central projection, betting that competing forces will balance each other out in the long run, while speculative traders may take much more aggressive positions to capitalize on potential volatility.
From intra-day to term traders, from position management to speculative trading, one thing is clear: data centers are changing how power traders operate. The key is to stay attuned to changing market dynamics.
Historical trends and decision heuristics may have little relevance to the load growth patterns of today. Changing loads, load profiles, and zone-level dynamics mean traders need responsive forecasting, a curious mindset, and trusted partners to help them navigate the many changes and uncertainties of the AI era.
Explore load forecasting software or explore detailed load growth dynamics in key power markets:
.svg)
%20(3).png)
%20(2).png)
%20(1).png)
.png)
.avif)
.avif)
.avif)
.avif)
.avif)
%20(15).avif)
.avif)
%20(10).avif)
.avif)
.avif)
.avif)
.avif)
