A critical look at AI's electricity demands, the risks of overbuilding infrastructure, and the Southeast's stake in this high-powered gamble.

Southeastern utilities are now projecting some of the most aggressive growth rates tied to data centers. History suggests caution is warranted—especially when ratepayers bear the financial risk.

The AI revolution is undeniably reshaping our digital landscape, but a crucial question remains: how will we power it? In his thought-provoking new paper, “Artificial Intelligence Meets Natural Stupidity: Managing the Risks,” energy expert Amory Lovins presents a sobering analysis that challenges many of the prevailing narratives around AI and energy, with particular implications for the Southeastern United States.

Lovins, the Cofounder and Chairman Emeritus of the Rocky Mountain Institute, has advised major firms and governments in more than 70 countries for over 45 years on energy and its links to security, development, environment, and economy.

The AI-Driven Electricity Bubble Risk

Lovins identifies a potential trillion-dollar overbuild in AI infrastructure, driven by speculative investments and highly uncertain projections. Despite the hype, AI’s future electricity needs remain wildly unpredictable, influenced by:

• Rapidly evolving technology (efficiency quadruples roughly each year)
• Uncertain market adoption
• Volatile market conditions
• Limited trust in AI systems

This scenario bears a striking resemblance to the 1999 IT-driven electricity boom, when the coal industry claimed information technology would require half the nation’s electricity by 2020. That prediction proved spectacularly wrong, resulting in hundreds of unnecessary power plants that hurt investors.

The Reality Check: A Southeastern Focus

Before we commit to building massive new power infrastructure, consider these facts:

• Current Usage: Data centers currently use only about 4-5% of US electricity and 1.5% of global electricity
• AI Portion: Of that data center usage, only about a quarter powers AI
• 2023-2024 Trends: US grid electricity fell in 2023 and rose just 2% in 2024 (the fourth-fastest rate in the past decade) 
• Data Centers’ Share: Data centers’ share of US electricity crept up from just 4.4% to 4.5%

Here in the Southeast, we’re seeing particularly aggressive growth projections. In Georgia, data centers are driving a remarkable 80% of Georgia Power’s projected power sales growth through 2028. This dramatic forecast has prompted concerns from five major tech companies who challenged these projections in 2024. Despite these concerns, Georgia has already approved approximately $3 billion in new fossil-fueled power plants to meet this speculative demand.

Lovins emphasizes that while electricity growth is real in a few hotspots (like Northern Virginia’s “Data Center Alley” and increasingly in parts of Georgia), this pattern has been widely misreported as a national trend. The Southeast is becoming a key battleground for these competing narratives about energy needs.

Who Bears the Risk? Southeastern Ratepayers in the Crosshairs

Perhaps the most concerning aspect of this AI power rush is who ultimately bears the financial risk when projections don’t materialize. Lovins’ paper highlights a critical issue that directly affects Southeastern utility customers: ratepayer risk.

When utilities build new power plants for data centers that either don’t materialize or fail to remain operational long-term, existing customers often get stuck with the bill. Three major US credit-rating agencies warned in 2024 that utilities face “substantial credit risk” from inaccurate load forecasts, noting there is “a considerable risk that residential [and other non-data-center] customers may end up paying disproportionately.” 

This risk is particularly acute in the Southeast:

• Georgia: Ratepayers could be on the hook for $3 billion in new fossil-fueled power plants that may not be needed if AI growth projections prove exaggerated
• Virginia: An independent state study found household electric bills could rise $14-37/month by 2040 due to AI-related infrastructure, with a 15% rise (averaging $22/month) already proposed
• Regional Impact: Across the Southeast, data center developers often secure favorable rates through “economic development” discounts, with lost revenue made up by other customers

As Lovins points out, this risk transfer is particularly concerning given that tech firms seeking these rate benefits hardly need financial assistance: “The Magnificent Seven in March 2025 had nearly five times the market capitalization of America’s hundred biggest shareholder-owned utilities combined.”

A local parallel can be drawn to past energy infrastructure projects in our region that didn’t pan out as projected, leaving ratepayers bearing costs for generations.

The Renewable Solution in the Southeast

Contrary to the narrative that only traditional power sources can meet AI’s demands, Lovins presents compelling evidence that renewables—particularly solar power—are actually ideal for powering AI data centers:

• Speed: Solar uniquely matches AI’s rapid development pace
• Flexibility: Can be deployed almost anywhere
• Reliability: When properly implemented with storage, provides highly reliable power
• Cost-effectiveness: Offers competitive pricing without the long-term risks

This is particularly relevant for the Southeast, which has excellent solar resources. According to Lovins’ analysis, “solar power uniquely matches AI’s torrid pace; can go about anywhere; would need zero land-use to power the world; and is readily integrated with other resources to provide cost-effective, clean, firm, critical-uses supply.” 

The potential for renewable solutions in our region is substantial. Lovins notes that “in 2023, 60 countries or territories were 50–100% powered by wind, sun, and water—12 of those 98.4–100%—while 11 US states produced 53.2–118% as much electricity from those sources as all the electricity they used.” While the Southeast has historically lagged in renewable adoption, the economic case is becoming increasingly compelling.

Major tech companies seem to recognize this reality, having already contracted for approximately 40 GW of renewable energy for their data centers. In the Southeast, we’re starting to see examples of the “Power Couple” approach Lovins advocates—placing new data centers with renewable energy sources at underused gas plants—which could provide a model for future development.

The practical evidence can’t be ignored: Lovins cites Australia’s largest electricity user, mining giant Rio Tinto, which “just chose 2.7 GW of wind and solar, backed by 0.6+ GW of batteries” as “the cheapest and most reliable solution” to power its aluminum-smelting complex. If renewables can reliably power aluminum smelting—one of the most electricity-intensive industrial processes—they can certainly handle data centers.

The Efficiency Factor

One frequently overlooked aspect is AI’s potential to improve energy efficiency elsewhere. If AI can enhance building energy management, industrial processes, and transportation systems, it could potentially save more energy than its data centers consume.

However, Lovins cautions that we must also consider AI’s potential to accelerate fossil fuel extraction through improved exploration and production methods—possibly negating any efficiency benefits.

A Sensible Path Forward for Southeastern Utilities and Regulators

To navigate these complex challenges, Lovins suggests several approaches that have particular relevance for the Southeast:

1. Risk Management: Require data center developers to guarantee power payments with bonds or insurance, ensuring ratepayers don’t bear the risk of project failures. As Lovins argues, “a sensible but apparently overlooked protection for the other customers…would be to require all promised payments to the utility for the large load’s new power supply to be bonded or insured by a creditworthy counterparty.” This approach would be especially valuable in states like Georgia, where large infrastructure investments are already planned.
2. Demand Flexibility: Implement “flexiwatt” strategies that shift computing loads to times when clean energy is abundant. Lovins cites a February 2025 assessment concluding that such flexible data-center operations “could probably make existing power plants and grids sufficient to run all US data centers proposed for this decade.” This could be particularly effective in the Southeast, where peak demand often coincides with summer cooling needs.
3. Co-location: Place new data centers with renewable energy sources at underused gas plants (“Power Couples”). Lovins notes that this approach “can neatly and profitably satisfy all the conflicting goals” of data center power needs. The Southeast has numerous natural gas facilities that could be ideal candidates for this approach.
4. Market-Based Solutions: Let markets accurately price and allocate risks to the appropriate beneficiaries. Lovins notes that “if the risk of project failure is as small as developers claim, bonding should be very cheap.” 
5. Ratepayer Protection: Implement stronger regulatory oversight similar to Oregon’s approach, which makes large-load developers share forecasting risks, or enforce binding take-or-pay contracts as seen in Indiana, Michigan, and Ohio. Lovins points out that “some states hold other customers harmless by simple regulatory policies like Kentucky’s 35-year practice of forbidding rate discounts below cost, or beyond five years, or unless the utility has surplus capacity.” 

For Southeastern utility commissions, these strategies offer practical ways to fulfill their mandate to protect consumers while still enabling economic development.

Conclusion: Protecting Southeastern Ratepayers While Embracing Innovation

As we navigate the AI era, the stakes for energy policy are particularly high in the Southeast. Poor bets on electricity demand could waste billions in investments, lock in unnecessary fossil fuel infrastructure, and—crucially for our region—burden ratepayers with decades of unnecessary rate increases from stranded assets.

The Southeast has historically been cautious about renewable energy adoption compared to some other regions, but the economics and reliability of these resources have improved dramatically. As Lovins points out, when properly understood and implemented, renewable energy solutions can now “neatly and profitably satisfy all the conflicting goals” of powering data centers reliably, affordably, and cleanly. 

For Southeastern utilities and regulators, the key takeaway is clear: rather than rushing into massive infrastructure investments based on speculative projections, a more measured approach that accurately prices risk and protects ratepayers will better serve our communities. As Lovins concludes, “If the risk of project failure is as small as developers claim, bonding should be very cheap. If it’s not so small, it’s more important to avoid.” 

By ensuring that tech companies bear the financial risks of their own growth projections through bonds or insurance requirements, we can welcome AI innovation while safeguarding our region’s economic interests.

With disciplined foresight, accurate risk pricing, and market-led investment in proven solutions, we can support AI’s development while avoiding repeating costly historical mistakes. The key lies in ensuring that AI’s energy foundation is as intelligent as the technology it powers—and that our region’s ratepayers don’t foot the bill for speculative investments.

This post is based on Amory Lovins’ May 2025 paper “Artificial Intelligence Meets Natural Stupidity: Managing the Risks.” The full paper provides extensive data and detailed analysis on these critical issues.

Resources

xAI Brings the “Move Fast and Break Things” Mindset to Memphis – SACE

We Went to the Town Elon Musk Is Poisoning – More Perfect Union