Podcast Summary: "The Energy Cost of AI"
The AI Policy Podcast, Center for Strategic and International Studies
Guest: Joseph Majkut, Director, Energy Security and Climate Change Program, CSIS
Host: Gregory C. Allen
Release Date: October 2, 2025
Overview
In this episode, Gregory C. Allen sits down with Dr. Joseph Majkut for an in-depth discussion about the intersection of artificial intelligence (AI) growth and the U.S. electricity grid. The conversation covers the surprising demands placed on America's energy infrastructure by AI data centers, potential bottlenecks, the evolving energy mix, and the use of AI to optimize grid operations. Drawing from several recent CSIS papers authored by Majkut, the episode gives policy, technical, and economic context for how the U.S. – and the world – can adapt to the coming wave of energy-hungry AI systems.
Key Discussion Points
Joseph Majkut’s Background and Expertise
- (02:37-11:17) Dr. Majkut shares his journey into energy policy, starting with studies in mathematics, working on risk and uncertainty modeling, and an academic focus on climate science and the value of Earth observations.
- Key insights into how complex systems, uncertainty quantification, and computational modeling underpin his approach to energy and climate policy.
- Expresses a preference for the faster pace and tangible impact of public policy over academic publication cycles.
The Current State of the U.S. Electric Grid
- (11:17-18:36) For nearly 20 years, overall electricity demand in the U.S. has stagnated (0% CAGR since 2007), thanks largely to energy efficiency improvements and some deindustrialization.
- Grid is defined by three major components: generation (power plants), transmission (power lines), and distribution (to users).
- Utilities have little recent experience with large-scale demand growth. Regional variation exists – some areas like Texas and the Southeast have seen growth, but most of the system hasn’t.
- Notable Quote [17:40, Joseph Majkut]:
"Our utility industry... are untrained and inexperienced in a growth environment. And that is actually part of the challenge that we're facing."
The AI Demand Shock: Data Centers & Grid Growth
Magnitude of Projected Growth
- (26:33-34:31) AI data center electricity demand projected to surge from 4 GW in 2024 to 84 GW by 2030 – a 2,100% increase.
- Context: 84 GW is more than the entire UK grid (71.7 GW in 2024), and comparable to Saudi Arabia’s total grid (91.1 GW).
- Data centers are “single-point loads,” unlike EVs or homes, which are diffuse; a single data center can demand more power than many existing industrial facilities.
- Notable Quote [32:13, Joseph Majkut]:
"They're like black holes of electricity... I have yet to find a verified example of something that is taking 2 gigawatts of electricity in a single concentrated facility."
Geographic Concentrations
- Most new demand forecasted to cluster in Virginia, Texas, and increasingly in parts of the Midwest (e.g., Ohio, Pennsylvania).
- Significant regional impact: localized demand may increase at far higher rates than national averages.
Bottlenecks: Grid Connection and Speed to Power
Transmission & Interconnection Delays
- (19:19-25:34) Grid must remain constantly in balance; connection of massive new data centers is a complex, slow, and highly human/manual process.
- Median time from grid interconnection request to operation now ~5 years ([49:36]).
- Notable Quote [23:52, Joseph Majkut]:
"There are consultants that operate here. There's humans sending each other's PDFs, right? ... This is clearly a problem that AI can help with."
The XAI Memphis Case Study
- (35:09-41:09) Elon Musk’s XAI Memphis data center: built and powered in months by assembling hundreds of portable gas generators and acquiring a near-defunct power plant across state lines, circumventing regulatory and grid delays.
- Illustrates that engineering/construction are less of a bottleneck than interconnection and regulation.
- Notable Quote [39:05, Joseph Majkut]:
"What you'll see is... more sort of duct tape together, temporary solutions are important for speed to power in building gigawatt scale facilities."
Political and Structural Challenges
- Momentous new AI load collides with existing political and regulatory friction.
- Large new loads spark local political opposition, especially if grid upgrades appear to benefit tech firms or raise rates for consumers.
- Potential for “zero sum” regional politics around grid capacity and new industrial projects.
Energy Mix: How Will the U.S. Power the AI Boom?
Short- and Medium-Term Trends
- (52:46-56:52) For the next ~5 years, almost all new marginal electricity generation is expected to come from natural gas plants, solar, and batteries. Some wind power, and delayed retirements of coal plants, will also play a role.
- Notable Quote [54:22, Joseph Majkut]:
"Natural gas, solar and batteries are like 90 plus percent of the story for the next five years."
Longer-Term Strategic Questions
- Sustaining growth beyond 2030 will require greater investment in firm and clean generation (especially nuclear). Existing coal plants provide short-term slack but can't support long-term needs.
- Policy debate emerging: whether to treat the AI energy boom as an opportunity for a U.S. nuclear renaissance.
- Challenge: rebuilding the industrial capacity and workforce for nuclear plant construction.
- Notable Quote [60:09, Joseph Majkut]:
"They [the Biden administration] want to see 10 gigawatts of new nuclear under advanced development or under construction by the end of the administration... If you can do the 10 and have those under construction, then in 2035, you might be able to do, you know, 20 or 40 if that ends up being what we need."
AI for the Grid: Using AI to Solve Energy System Challenges
- (63:35-75:47) AI is being rapidly deployed to tackle energy sector challenges.
- Grid Interconnection: AI-powered tools can radically speed up project assessment and connection (e.g., Google-sponsored Tapestry).
- Transmission Optimization: Machine learning can allow dynamic line rating based on real-time and forecasted conditions, using weather and sensor data to operate closer to true capacity while remaining safe.
- Renewables Forecasting: ML improves weather prediction for solar and wind, helping operator manage variability.
- Demand Flexibility: Large new data center loads may be required to offer “demand flexibility” – temporarily reducing consumption during grid stress periods to support reliability.
- Economic incentives and new regulatory regimes may shape this future practice.
- Notable Quote [74:06, Joseph Majkut]:
"...if you were able to do this under certain conditions, you could add something like 100 gigawatts of data centers onto today's grid."
International Perspective: China and the Global AI-Energy Race
- China has rapidly expanded both grid capacity and its ability to deploy new power plants, including coal, renewables, and nuclear—contrasting with slower U.S. expansion.
- Other countries (UAE, Saudi Arabia, Japan, Norway, Greece) vying to attract AI infrastructure with abundant or fast-growing electricity systems.
- Relative model efficiency (compute per watt) may soon become a competitive factor, particularly if U.S. bottlenecks persist and firms must consider offshoring.
- Notable Quote [77:09, Joseph Majkut]:
"The concentration of data center demand that we talk about in our paper... this is the most American way to approach the problem—just throw an incredible amount of compute at it."
Notable Quotes & Memorable Moments
-
On the “black hole” nature of data center demand:
[32:13] Joseph Majkut:"They're like black holes of electricity."
-
On utility industry readiness:
[17:40]"Our utility industry... are untrained and inexperienced in a growth environment."
-
On speed to power:
[39:37]"You can build the... shell of the data center and pack it with chips in three months. It's going to take you years to get it interconnected. That's the speed to power."
-
On pragmatic policy and the nuclear future:
[60:09]"They want to see 10 gigawatts of new nuclear under advanced development or under construction by the end of the administration... If you can do the 10 and have those under construction, then in 2035, you might be able to do, you know, 20 or 40..."
-
Regarding AI-powered efficiency gains:
[65:36] Allen paraphrasing Google experience:"DeepMind, I want to say I can't remember the exact numbers, but it was something like they gave them two weeks and they increased efficiency by 10%..."
Timestamps for Key Segments
| Segment | Timestamp | |------------------------------------------------|------------| | Majkut’s career journey | 02:37–11:17| | State of the US electric grid | 11:17–18:36| | Data centers’ projected growth | 26:33–34:31| | The “speed to power” bottleneck | 35:09–41:09| | Political and structural challenges | 44:32–46:33| | Current/future US electricity mix | 52:46–60:09| | Nuclear’s role in long-term strategy | 60:09–61:15| | How AI can optimize the grid | 63:35–75:47| | International energy/AI race | 76:06–81:44| | Upcoming CSIS research directions | 82:02–85:19|
Structure of Conversation
- Context-setting: Host introduces Majkut and the rationale for the episode, referencing CSIS’s recent research.
- Deep-dive into energy policy and the grid: Majkut explains the grid’s technical makeup, how stagnation set the stage for the current AI-driven demand shock, and why existing processes can’t meet this new pace.
- Case studies: Example of XAI Memphis data center shows workarounds and exposes underlying system issues.
- Policy, politics, and possible futures: Discussion blends market-driven realities, regulatory barriers, and national strategic imperatives, especially around nuclear.
- AI as a solution, not just a consumer: Turns tables to explore how AI can help the energy sector address its own limitations, drawing on early success stories and new research.
- Global implications: Places the U.S. story in a wider geopolitical and industrial context—how energy policy and AI intersect globally and what the U.S. can learn or fear from international competitors.
- Outlook: Majkut previews upcoming CSIS studies looking at China-specific dynamics, regional U.S. differences, and macro questions of AI’s impact on overall energy intensity.
Takeaways
- The U.S. electric grid is facing an unprecedented surge in demand driven by AI data centers—comparable in scale to adding an entire new national grid.
- Existing infrastructure and institutional frameworks are unprepared for rapid scaling; major bottlenecks are often regulatory or process-based, not engineering.
- Natural gas, solar, and batteries will carry most of the immediate load, but a longer-term transition to nuclear and other firm, low-carbon resources is necessary for sustained growth and decarbonization.
- AI itself offers important solutions to optimize and accelerate energy system upgrades, from grid connections to transmission utilization and demand flexibility.
- The U.S. faces stiff international competition – especially from China and energy-rich nations – but can leverage its current AI and industrial strengths if it can adapt its energy policy and infrastructure quickly enough.
For more detail, read the referenced CSIS reports:
- Electricity Supply Bottleneck on US AI Dominance (Mar 2025)
- AI Power Sur: Scenarios for Gen AI Data Centers through 2030
- AI for the Grid: Opportunities, Risks and Safeguards (Sep 2025)
End of Summary
