Powering Sustainable AI in the United States: The Resurgence of U.S. Electricity Demand

Rémi Paccou - Sustainability Researcher | Energy System Analysis, Sustainable AI/Digital, Data Centers & ICT | PhD Student at CIRED, Prospective Modeling for Sustainable Development.

This week at BloombergNEF’s Summit in New York, Schneider Electric North America President Aamir Paul highlighted a stark projection: AI‑related computing could add between 42 GW and 90 GW of continuous load to the U.S. grid by 2030.

That projection isn’t just a headline—it marks a turning point. In the U.S., AI is moving rapidly from hype to infrastructure reality. Its energy appetite is growing just as fast as its adoption. While conversations around responsible AI models and ethical use are gaining traction, the underlying physical demands—on power grids, on land, on materials—remain largely unaddressed.

Two structural shifts are now reshaping the conversation. First, the U.S. grid was never designed for AI-scale loads; it was built for a mid-20th century consumer economy, not for data centers drawing gigawatts. Second, electricity itself is becoming a constrained resource, with implications that ripple across decarbonization goals and economic competitiveness.

Rushing to build for short-term needs risks derailing efforts to electrify and decarbonize the economy. The US has waited too long. The AI demand shock must drive a shift to modernize infrastructure for resilience - not continued dependence on outdated systems. This is a systemic problem. It requires a systemic response.

To contribute to the conversation, we’ve drafted five short articles that explore a range of possible futures for AI and energy in the U.S.

1. Today: Contextualizing AI-Driven Energy Demand

2. Futures Framework: Four Divergent Scenarios

3. Systemic Risks: Limits to Growth, Techno-Excess, and Crisis Dynamics

4. Sustainable AI: Efficiency, Integration, and Grid Co-Optimization

5. Strategic Pathways: Policy, Planning, and Infrastructure Alignment

Let's start.

Electricity as the New Scarcity

The era of flat power demand in the U.S. has ended. From the post-World War II period until the early 2000s, U.S. electricity demand grew significantly, driven by economic expansion, population growth, and the adoption of new electric appliances like air conditioners, computers, and incandescent lighting (U.S. EIA, 2024). Peak annual growth rates reached 8.8% in the 1950s. However, this growth gradually slowed, leveling off at around 0.7% and 0.6% annually from the early 2000s to the 2010s. This deceleration was due to energy efficiency improvements (Compact Fluorescent Lamps and Light-Emitting Diodes), economic downturns (e.g., the 2008 financial crisis), and declining domestic manufacturing (Wilson et al., 2024).

The decade of the 2020s is a turning point. Newly released data shows seismic change: growth jumped from 0.6% in 2022 to an estimated 3.0% in 2024 (NERC, 2024). This acceleration is highlighted by the five-fold increase in the 2024 national electricity projection from that in 2022. Demand by 2029 is set to rise by 128 GW – equivalent in its power usage to thirteen New York Cities. This rapid and enormous change must command immediate notice, as it may cause heightened greenhouse gas emissions, escalating electricity bills, and stress on the aging U.S. grid (WRI, March 20, 2025).

Meeting these needs will require a profound expansion in generation and transmission capacity relative to recent decades (U.S. DOE, 2023). Five-year electricity growth projections have skyrocketed from 2.8% to 15.8% in only two years (NERC, 2024). That growth is fueled by the booming expansion of AI and data centers, federal policy aimed at domestic advanced manufacturing, transportation and heat electrification, growing power needs from oil and natural gas sources, and production of hydrogen fuel facilities (Wilson et al., 2024; NERC, 2024). Others predict major expansion in five-year growth to a new total of 128 GW—the equivalent of annual electricity usage by all homes in the U.S.

Can Power Grids and Data Centers Meet Surging Demand?

Beyond the hype, a consensus emerges across industries: the escalating energy demands of AI have moved beyond a distant threat to a tangible reality now reshaping infrastructure planning (Bain & Company, 2024), and current projections warn that data centers could double their share of U.S. electricity consumption, reaching 5-9% by 2030 (Shehabi et al., 2024). While precise annual growth estimates remain uncertain due to AI's dynamic scaling and efficiency, the trend is clear: regional grids face booming electricity demand as the US economy's structural electrification unfolds across sectors. High-growth scenarios of 10-15% are increasingly plausible, driven by grid infrastructure limitations and permitting bottlenecks (EPRI, 2024). Although some earlier power shortage alarms were exaggerated, the concentrated proliferation of data centers, coupled with the rising demand, undeniably places significant strain on regional electricity grids, demanding a fundamental shift towards a new electricity system paradigm.

The AI Grid Challenge: A Race Against Demand

AI-driven data centers are causing a significant increase in power demand, necessitating a massive modernization of the U.S. grid infrastructure, rather than just minor upgrades. The North American Electric Reliability Corporation (NERC) forecasts a 21.5% rise in U.S. winter peak load over the next decade, reaching an unprecedented 843 GW by 2034. This growth, combined with existing interconnection queues, supply chain issues, permitting delays, and labor shortages, directly threatens to increase energy costs, raise prices for consumers, and delay the deployment of necessary infrastructure upgrades.

Building upon the lessons of the mid-20th century's rapid grid expansion, today's infrastructure challenges present unique opportunities to rethink traditional approaches. This necessity for a new systemic perspective can emerge from constraints such as stricter permitting processes, the need for efficient land use, and ambitious decarbonization goals (WRI, 2025). However, a bottleneck emerges from the disconnect between the swift integration of new loads and the protracted timelines for expanding generation and transmission, potentially leading to economic imbalances that favor rapidly scaling AI-driven sectors.

Time to Meet Great Issues

"We cannot avoid meeting great issues. All that we can determine for ourselves is whether we shall meet them well or ill". Theodore Roosevelt's call to action in his 1899 speech, "The Strenuous Life", resonates profoundly today as the U.S. confronts the dual challenge of AI's short-term demand and electrical infrastructure modernization. The nation can no longer build as it did decades ago, leaving no option but a system vision over traditional ones.

But true systemic change involves the integration of practices, not just isolated efforts. The Rocky Mountain Institute's Power Couples model (Engel et al., 2025) demonstrates this by offering a blueprint for pioneering AI companies to swiftly access clean power for their data centers while safeguarding grid stability. In the US context, this means for instance strategically pairing AI Data Centers with new, on-site solar, wind, and battery generation located near existing grid infrastructure.

Risks are real. Our research reveals AI is set to drive 20–50% of US electricity demand growth between 2025 and 2030, and power demand could exceed current grid capacity by over 29 GW. Regional hotspots-like Texas, Northern Virginia, and California-are already showing signs of stress. Seven major US grid regions face potential reserve shortfalls by 2028.

Four Future Scenarios on AI Consumption of Electricity

We studied four counterfactual scenarios of AI electricity demand, confronting these with the dynamics of data center infrastructure expansion alongside grid capacity. We developed a system dynamics model to detect factors causing regional energy crunches and propose technology and policy recommendations to power sustainable AI in the US.

Scenario 1: Sustainable AI. Sustainable AI proponents promote extensive use of power‑efficient data centers, hardware and algorithms that generate a mutually supportive loop in which AI optimizes the grid and renewable energy integration to stabilize usage. Industrial coalitions cross‑utilize renewable capacity and schedule demand response.

Scenario 2: Limits to Growth. Expansion in AI is faced with availability of power, shortages of data, material shortages, regulation barriers and resistance from society, producing a constrained path characterized by mismatches between supply and infrastructure and reactive responses. Periodic curtailments and price peaks destabilize value chains and induce piecemeal responses.

Scenario 3: Abundance Without Boundaries. Techno-optimists promote unrestrained AI implementation in U.S. industries with the hope that efficiency improvements will address resource shortages. But lowered costs per unit unleash a Jevons Paradox rebound that spurs unmanageable demand that exceeds capacity and obstructs wider electrification.

Scenario 4: Energy Crisis. Panic caused by black‑swan AI load surges triggers grid shortages, economic shocks and stringent regulation as hurried adoption collides with too little planning, incorrect forecasting and uncoordinated leadership. Enforced curtailments bring AI deployments to a halt and leave systemic weakness revealed.

As Graham T.T. Molitor put it, “The future isn’t something we inherit. It’s something we wrestle into existence.” - The real test for AI in the US isn’t whether we can build powerful machines; it’s whether we can imagine a future that runs on purpose.

Which of these scenarios are you seeing play out in your work or region? Share your observations—and watch for our next piece on blind spots in AI load forecasting.

In our second article, we’ll take a closer look at how current forecasting methods capture—or miss—the scale and speed of AI-driven demand. See you there.

Full paper here > https://www.se.com/ww/en/insights/sustainability/sustainability-research-institute/powering-sustainable-ai-in-the-united-states/

#AIandEnergy #SustainableTech #FutureofAI #EnergyTransition