The numbers behind the AI buildout, from the power company's side
Most of what the market knows about AI infrastructure demand comes from hyperscaler capex guidance and vendor announcements. Southern Company's August 14 earnings call offered a different vantage point: the utility actually delivering the electricity. Data center power usage across Southern's footprint rose 55 percent compared to the same quarter last year. Operational data center load now exceeds 1.2 gigawatts, up 500 megawatts year over year, a growth rate that would be remarkable for almost any other category of industrial electricity demand.
The forward pipeline is larger still. Southern's contracted large load pipeline, meaning deals already signed with committed capacity, grew by 6 gigawatts in a single quarter to reach 17 gigawatts total. Another 8 gigawatts sits in late stage development, and the company describes a prospective pipeline beyond that exceeding 75 gigawatts. Even accounting for the reality that not every prospective megawatt converts to a signed contract, the scale of demand utilities are fielding right now is difficult to reconcile with any narrative of AI infrastructure spending slowing down.
The OpenAI deal in the numbers
Inside that pipeline sits one deal worth naming specifically: Georgia Power, a Southern Company subsidiary, signed a 25 year agreement with OpenAI for a 3.2 gigawatt data center facility near Savannah. A quarter century power purchase commitment is an unusually long horizon even by utility standards, and it tells you something about how OpenAI is thinking about its own infrastructure planning: in terms of decades of guaranteed capacity, well beyond the timeline of any single model generation.
For Southern Company, a deal of that length and size effectively locks in a predictable revenue stream that utilities have not had access to at this scale from any other industrial customer category in recent memory. It also illustrates why utilities across the Southeast and Texas are increasingly willing to build dedicated infrastructure for a single customer, a departure from the traditional utility model of aggregating diverse demand across many smaller customers.
Addressing the moratorium backlash directly
Southern's CEO Chris Womack used the earnings call to respond directly to the growing wave of local and state pushback against data center construction, the same pressure that has produced moratoriums in New York and an interconnection audit in Texas over the past six weeks. Womack said there is 'noise all across the country about data centers' and that the industry has 'got to do a better job' countering misinformation about their impact on communities and rates, a remark aimed as much at Southern's own investors as at the public debate playing out in statehouses.
That framing, misinformation rather than legitimate concern, is a notable choice of words for a utility executive speaking to investors rather than to a public hearing. It suggests utilities benefiting financially from data center demand are increasingly willing to push back publicly against the regulatory skepticism building in states like Texas and New York, rather than staying quiet and letting hyperscaler customers absorb the political heat alone. Expect more utility executives to adopt similar language on upcoming earnings calls as the moratorium wave spreads to additional states.
Who is actually financing this buildout
Southern secured a $26.5 billion loan from the Department of Energy in February 2026 specifically to fund the power infrastructure this demand requires, a reminder that the capital behind AI data center power is not coming exclusively from hyperscaler balance sheets or private credit funds. Federal loan guarantees are underwriting a meaningful share of the generation and transmission buildout utilities need to keep pace with contracted demand like Southern's 17 gigawatt pipeline.
That mix of federal financing, decades long private power purchase agreements, and utility rate base growth is quietly becoming the actual capital structure of the AI boom's physical layer, distinct from and often invisible next to the headline numbers on hyperscaler capex. Understanding who bears the financing risk in that structure, DOE loan guarantees carry different risk allocation than a corporate power purchase agreement, matters for anyone trying to model how durable this buildout actually is.
Why Georgia looks different from Texas right now
Georgia's regulatory posture stands in visible contrast to the pause playing out in Texas and the moratorium already in place in New York. Southern's earnings call described continued momentum rather than a slowdown, and Womack's public defense of the industry suggests state regulators and the utility are aligned on continuing to approve large load connections rather than freezing them pending review. That divergence matters for site selection: two states with heavy data center activity are currently sending opposite signals about how predictable new capacity timelines will be.
Whether Georgia's approach survives the same political pressure building in Texas and New York is an open question. Southern's 75 gigawatt prospective pipeline is large enough that even a fraction of it converting to signed contracts would strain water and grid resources in ways that could eventually invite the same audit style scrutiny Texas regulators are now applying. For now, Georgia remains a comparatively lower friction option for enterprises evaluating where new AI infrastructure capacity is likely to come online on schedule.
What this means for the reader's planning
If you operate in the Southeast, Southern's 17 gigawatt contracted pipeline and 75 gigawatt prospective pipeline are useful signals for regional capacity planning, both for your own facilities and for understanding how much competing demand your cloud vendors' regional data centers may face for grid connections. A utility fielding this much committed demand is likely to prioritize its largest, longest term contracts, meaning smaller enterprise loads could see longer connection timelines even where a utility reports abundant future capacity on paper.
More broadly, treat utility earnings calls as a legitimate, underused source of ground truth on AI infrastructure demand. Hyperscaler capex guidance tells you what vendors intend to spend. Utility contracted pipelines tell you what is actually getting built and financed, on what timeline, and increasingly, how utilities themselves are choosing to defend that buildout publicly against the same regulatory pushback now surfacing in Texas and New York. Add your regional utility's quarterly earnings call to the same watch list you already keep for hyperscaler capex updates.
