Not long ago, the first questions in data center development centered on land availability, tax incentives, and fiber connectivity. Those conversations haven’t gone away, but they’ve become secondary. Today, the first question is almost always the same: What’s your power solution?
That shift didn’t happen overnight, but it has accelerated faster than most of the industry anticipated. AI workloads are driving exponential growth in compute demand, and the power infrastructure required to support that growth simply hasn’t kept pace. Understanding why, as well as what it means for developers, investors, and operators, is no longer optional but table stakes.
A supply problem decades in the making
Unfortunately, the electricity grid wasn’t built for this moment. Across much of the U.S., baseload generation capacity has been contracting for the better part of a decade. In many of the deregulated markets, without clear market signals and amid real volatility since deregulation, many generators have become reluctant to build new plants without a signed power purchase agreement in hand. The result is a market caught in a circular bind: you need power to develop a site, and you need a committed offtake agreement to actually deliver that power.
Meanwhile, interconnection queues have become severely backlogged. The Lawrence Berkeley National Laboratory has tracked almost two-and-a-half terawatts of generation and storage projects stuck in queues nationwide — supply that exists on paper but cannot yet reach the grid. Permitting timelines, interconnection studies and supply chain constraints compound the problem further. New baseload generation, even when fully financed, can take years to energize.
Demand is real and complicated
On the demand side, the picture is equally nuanced. AI-driven load growth is mostly genuine and substantial, but utility forecasts are being distorted by speculative or duplicative load requests from data center developers staking multiple sites simultaneously. These phantom load requests are causing or accelerating regional forecasts of generation shortfalls. These loads also operate differently from the demand grid operators have traditionally served. They operate in large blocks (100MW, 500MW or more) and can spike in their demand or decrease in those large chunks creating challenges for maintaining reliability and adding another layer of complexity for utilities planning capacity.
The upshot: forecasted demand and actual demand are two very different numbers, and the gap between them is causing real problems for grid planning and market design alike.
The Constraints Are Structural, Not Temporary
It’s tempting to treat power access as a transactional problem that can be financed or contracted around. That instinct is understandable, but it’s also one of the most persistent and inefficient misconceptions in the market today. While capital can build island solutions that are as reliable as the grid, the cost is substantially higher as excess generation built for redundancy sits idle most of the time. In addition, the regulatory framework governing the U.S. electricity system is a patchwork of overlapping utility territories, state rules and regional structures. Every service territory has its own process, its own timeline and its own set of stakeholders. Even the most well-capitalized developers cannot simply buy their way through a system that moves at a regulatory pace, not a commercial one.
Another regulatory challenge is that grid investments are paid for by all of a utility’s customers on monthly electricity bills. So even though data centers want to pay their fair share, in practice, determining the right amount and the mechanisms for large-load allocations to protect customers from increasing costs is complicated and evolving.
Off-grid solutions are emerging as a workaround, but they come with real tradeoffs: reduced efficiency, and substantially higher cost to deliver the same level of resilience and reliability. In some states, there are also serious legal constraints.
What This Means for the Market
The winners in this cycle will not necessarily be those with the most capital or the largest land positions. The ones who solve the power constraint are those who can translate demand into deliverable supply. That requires flexibility, early engagement with interconnecting utilities, and a willingness to work with the grid rather than around it.
As of today, power access is going to get harder before it gets easier. The developers and investors who recognize that now and build their strategies accordingly will be the ones positioned to meet demand when it matters most.
To meet that demand means rethinking what site viability actually looks like. It means understanding the full cost of power, including the cost of constructing and operating new baseload generation and required grid investments. And it means approaching utility relationships as partnerships rather than procurement exercises. Helping the utility help you isn’t a soft strategy; in many cases, it’s the only viable path to getting a project energized on a timeline that matters.
There is no cookie-cutter solution here. Each region, utility and project requires a tailored commercial and regulatory approach. The sooner that becomes the default assumption, the better the industry will be positioned to meet the scale of demand ahead.
In our next piece, we’ll explore how more efficient market design and updated regulatory policy could meaningfully accelerate the path for new generation to come online, as well as what that could mean for the broader AI and data center build-out.
Navigating the power landscape requires more than a standard playbook. If you’re working through a power strategy for your next project, we’d welcome the conversation. Contact usto get started.
Editor’s note: This is the first in a series of articles from ASG discussing power, market design and policy. This piece is by Allison Clements with contributions from Dustin Wertheimer.
