Energy Gang – “What Does DeepSeek AI Mean For Energy?”
Date: February 4, 2025
Host: Ed Crooks (A)
Guests: Amy Myers Jaffe (B), Melissa Lott (C)
Episode Overview
This episode explores two converging events with seismic implications for the energy sector:
- The return of Donald Trump to the US presidency and his dramatic shifts in energy policy.
- The release of China's DeepSeek, a new large language AI model that may be orders of magnitude more efficient than previous models.
Ed Crooks and regulars Amy Myers Jaffe and Melissa Lott analyze how DeepSeek could disrupt expectations for power demand from AI, how markets are reacting, and what all this means against the backdrop of US policy turbulence. The discussion moves through hype vs. reality in energy predictions, infrastructure bottlenecks, debates over "baseload power," and concludes with insightful “free electrons” from each panelist.
1. Breaking News: DeepSeek AI and Energy Demand
[02:00–16:10]
A. DeepSeek’s Disruptive Efficiency
- DeepSeek, a new Chinese AI model, reportedly requires just 3–10% of the power needed to train previous-generation models like ChatGPT.
- Potential implications: May upend predictions of endless exponential electricity demand from AI-driven data centers.
Quote:
"Deepseek's power consumption for training the model could be just 10% or 5%, maybe even 3%, on some calculations..."
—Ed Crooks, [02:00]
B. Markets React to AI/Energy Hype
- Energy and tech stocks whipsawed:
- Constellation (major power generator) and EQT (large gas producer) saw dramatic movements as expectations for surging demand shifted.
- Investors question: Is the power demand hype overblown, or is it just moving to different sectors?
Quote:
"Their stock was $347 on Jan 24. It went to a low of 274... maybe all this power demand hype was too much hype."
—Amy Myers Jaffe, [04:00]
C. Local vs. Global Impact
- Electricity growth is not smooth; localized impact matters more than aggregate demand projections.
- Data center demands and manufacturing have highly region-specific effects, creating pressures on grid infrastructure which may differ widely across the US.
Quote:
"Does it change the overall story for me... No. Could it have local impacts? Yes. When I look at the numbers..."
—Melissa Lott, [07:00]
2. Revisiting Fundamentals: Grids, Infrastructure, and Technology Choices
[10:37–30:00]
A. The Aging Grid and Investment Imperative
- US and global grid infrastructure is aging; requires major investment regardless of specific AI-driven demand.
- Virtual Power Plants (VPPs) and storage seen as faster, more nimble solutions to local disruptions.
Quote:
"We still have an aging grid in the US, still have issues with reliability. We still have those problems that require investment to fix."
—Melissa Lott, [11:30]
B. Reality Check on Data Centers and "Stranded Gas"
- Grand ideas (e.g., building new gas plants for isolated data centers) run into real-world constraints: fiber capacity, backup/redundancy needs, environmental commitments (trust in eventual CCS), and lack of perfect sites.
C. Hype vs. Reality
- Panel agrees much of the market movement reflects hype and a rush to capitalize—while the real bottleneck remains physical infrastructure rather than theoretical demand.
Quote:
"Does this fundamentally change things... if you're a government regulator? It doesn't..."
—Melissa Lott, [13:00]
3. Big Ideas: Jevons Paradox, Rosenfeld Effect, and the Limits of Efficiency
[16:10–23:47]
A. Jevons Paradox: Efficiency Increases Can Spur Demand
- Satya Nadella (Microsoft CEO) tweets on Jevons Paradox: As energy efficiency improves, total energy use often rises, not falls.
- Example: UK coal in the 19th century; may apply to AI/data centers.
Quote:
"As energy efficiency increases, energy consumption will rise because you can do more with the energy."
—Ed Crooks, [17:00]
B. Rosenfeld Effect: When Efficiency Actually Flattened Demand
- California’s experience: Per capita demand stabilized for decades through aggressive efficiency standards (Rosenfeld effect/negawatt concept).
- But global context matters—growing economies lacking basic energy access may not see net reductions until basic needs are met.
Quote:
"Efficiency is really great, but it doesn't change the fundamentals of increasing energy demand."
—Melissa Lott, [18:25]
- Debate:
Amy: "Globally, the link between economic growth and energy has shrunk over time because of efficiency..." [19:27]
Melissa: Argues the distinction is sharper when considering populations just now gaining access.
4. Energy System Constraints: Demand vs. Supply, and the Role of Trade
[23:47–30:03]
A. Biggest Bottleneck: Supply-Side, Not AI’s Demand
- The constraint is infrastructure: grid upgrades, siting and permitting new generation, transformers, turbines.
- Tech companies will take what power they can get—that supply limit is binding.
Quote:
"The constraint here is going to be on the energy and electricity supply side, not on the demand side from AI."
—Ed Crooks, [24:47]
B. Energy/Electricity Trade
- Interstate/intercountry electricity trading is a crucial yet overlooked part of solving localized or regional deficits.
- Examples: US–Canada power trade; barriers within Canada; African Union plans for continental grids; Gulf region analysis (cited research: David Wogan, KAPSARC).
5. Policy Shock: Trump Admin, National Security, and "Baseload"
[30:10–47:51]
A. Administration’s Focus
- Trump’s new team emphasizes baseload/reliable power (read: gas, nuclear, possibly coal), seeing AI as a "dual use" (civilian&military) tech arms race vs China.
- Executive orders are jettisoning climate policy in favor of maximizing energy supply and "national security."
Quote:
"They see artificial intelligence as dual-use technology... because AI needs power, developing the power industry is an important part."
—Ed Crooks, [32:00]
B. The Baseload Debate
- What is “baseload” today? Old concept: plants that run steadily, 24/7. New reality: system needs flexibility, resilience, and diversified assets (renewables, storage, demand management).
- Amy: “Baseload should be retired”—focus should be on minimizing congestion and system risk, not 24/7 operation.
- Melissa: “Baseload” is operational, not a technology. The system needs a mix for reliability, not a single solution.
Notable Exchange:
Melissa: "Baseload is not used properly… it's a description of how a plant operates…" [36:05]
Amy: "This whole baseload thing needs to be retired. What we're really talking about is how do I eliminate system congestion..." [34:23]
- Key Point: Diversity in generation—including cheap renewables, firm/dispatchable resources, storage, and VPPs—is essential for system resilience and cost control.
6. Practical Realities: Speed, Costs, and the IRA/Tariff Uncertainty
[49:08–53:06]
A. What’s in the Queue?
- 93% of power generation projects seeking connections are wind, solar, or storage (per LBNL).
- Speed: Gas or nuclear plants can't ramp up fast enough to meet near-term needs, especially given permitting delays; wind and solar can.
Quote:
"If you want wind and solar, you can actually do that pretty quickly. The equipment is available now."
—Ed Crooks, [50:00]
B. Costs & Incentives Run Deep
- Amy describes a classroom exercise where solar PPA bids turned unprofitable without federal tax credits—but some projects could still work, especially as costs continue to fall.
- Tariffs on Chinese panels and debates over the Inflation Reduction Act (IRA) tax credits add uncertainty.
7. State vs. Federal Flashpoints & National Security
[54:17–55:24]
- Some executive orders specifically flag "dangerous local policies"—implying potential legal battles between the federal government and states such as CA & NY over clean energy, gas bans, and plant closures.
- Panel predicts these will become battlegrounds under the national security policy umbrella.
8. Notable Quotes & Memorable Moments
- On upgrading infrastructure:
- "The constraint is infrastructure... the tech industry will take as much power as they can get, but they won't be able to get all the power they want."
—Ed Crooks, [24:47]
- "The constraint is infrastructure... the tech industry will take as much power as they can get, but they won't be able to get all the power they want."
- On hype vs. reality:
- "An enormous amount of hype... people were looking for an excuse to take some profits and let air out of some of those companies."
—Ed Crooks, [13:41]
- "An enormous amount of hype... people were looking for an excuse to take some profits and let air out of some of those companies."
- On Jevons/Rosenfeld:
- "It is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to a diminished consumption."
—Ed Crooks quoting Jevons, [17:13]
- "It is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to a diminished consumption."
- On baseload:
- "I think this whole baseload thing needs to be retired."
—Amy Myers Jaffe, [34:23]
- "I think this whole baseload thing needs to be retired."
9. Free Electrons
[55:44–end]
Short “what’s catching your eye” energy-podcast tradition, rich with personality:
- Melissa: Recommends Ed’s recent energy policy newsletter and a Guardian explainer [57:00], as tools to keep track of the swirl of recent executive orders.
- Amy: Shares news of the first data center on the moon—a Lone Star Data Holdings pilot aboard a SpaceX flight, powered by solar, with the State of Florida, Valkyrie AI, and Imagine Dragons as customers. Points to the science fiction–esque implications and practical challenges like data transmission and radiation. [58:15]
- Ed: Recommends Jonathan Eig’s biography King on Martin Luther King, Jr., for insights on social change, leadership, and movement-building. [61:14]
10. Key Takeaways
- DeepSeek AI may dramatically change projections around AI-driven energy demand, but local impacts and underlying grid constraints may matter more.
- AI/data center energy forecasts are wildly uncertain; market reactions often overshoot substance.
- The energy transition's biggest bottleneck remains on the supply/infrastructure side: slow permitting, aging grids, and materials bottlenecks.
- The "baseload" debate is deeply political but technically outdated—the grid of the future demands asset diversity, flexibility, and resilience, not a single type of plant.
- Policy, especially under the Trump administration, is rapidly shifting focus to energy "reliability" and "national security," with potential legal clashes looming between federal and state governments.
- Technological and economic realities (cost trends, project queues, and speed-to-deployment) make renewables and storage hard to ignore, regardless of political rhetoric.
- The only certainty is continued, rapid change—and the need for more investment, humility in forecasts, and honest education to separate hype from reality.
Panel’s Parting Message:
This is a period of historic inflection for energy. Expect more volatility—and more conversations busting hype, drawing on data, and exploring the messy, fascinating reality of the energy transition.
