Energy Gang Podcast Summary

Episode Title: Data centers for AI will need to embrace flexibility if our electricity system is going to cope. How can large loads support the grid?
Podcast: Energy Gang by Wood Mackenzie
Release Date: September 16, 2025
Host: Ed Crooks
Guests:

• Amy Myers Jaffe (NYU Energy, Climate Justice, and Sustainability Lab)
• Shanu Matthew (Lazard Asset Management)
• Varun Sivaram (Founder & CEO of Emerald AI)

Episode Overview

This episode investigates how the explosive growth of AI data centers—soon to be the largest electric loads in the U.S.—is straining today’s grid, driving energy costs, and creating security concerns. The panel debates how embracing “flexibility” at large data centers could mitigate infrastructure needs, lower costs for consumers, and accelerate clean grid transition. With perspectives from industry, policy, investment, and tech, the discussion explores the regulatory, technical, and economic hurdles—and opportunities—of making AI power-hungry but grid-friendly.

Key Discussion Points & Insights

1. The AI Data Center Boom and Grid Strain

• AI data centers’ electric demand is skyrocketing. According to Varun Sivaram, “AI is maybe the most exciting thing I have ever seen happen to this energy sector. But it’s becoming one of the largest and soon the largest consumer of American electricity.” (03:43)
• The demand growth is “doubling every year,” risking overloads, reliability threats, and rising consumer bills if infrastructure doesn’t keep up. (38:00)

2. A National Security and Economic Challenge

• Amy notes U.S. Secretary of Energy Chris Wright calls the grid “very concerning” and urges a “Manhattan Project, number two” for electricity supply to meet both AI and broad economic needs. (05:52, 06:12)
• “Everything has to be done to maximize electricity supplies… stakes could not be higher,” Ed summarizes. (06:12)

3. Flexibility: The Game-Changer for Fast Data Center Growth

• The panel agrees that if data centers become flexible—able to dial down power briefly during grid strain— they can connect faster and avoid exacerbating bottlenecks.
• “100 gigawatts of available stranded capacity… could be unlocked by just modestly flexible data centers,” says Varun, referencing research by Tyler Norris. (11:23)
• The “immovable object” (utility reliability requirements) is meeting the “unstoppable force” (tech/data center demand for uptime). Both sides must nudge their models. (11:55)

4. Tech Sector Reluctance and Emerging Solutions

• Shanu explains: “For a tech company… being told you can’t run that [investment] for every maximum hour is a really negative NPV calculation. That’s why they’re so hesitant.” (01:04, 14:15)
• Yet, recent shifts in software and hardware (distributed, interconnected data centers, improved orchestration) are making flexibility more viable—and in some cases, inevitable. (14:15, 15:55)
• “If flexibility is part of the equation, what do they get in return for that?” Investors want assurance of ROI if asked to provide flexibility. (14:15)

5. Regulatory Push: Texas and PJM’s Tough Choices

• Texas (ERCOT) leads with laws enabling the grid to “turn a data center off completely” if residential reliability is at risk—setting a precedent for other states. (16:36)
• PJM proposes “NCBL (Non Capacity-Backed Load)” rules—flexible loads risk being the first curtailed, unless they bring backup capacity. Industry and politicians worry about fairness and costs. (43:08, 46:15)
• The panel expects further push and pull as market incentives, speed of interconnect, and regulatory frameworks are negotiated. (47:20)

6. The Role of Distributed and Interconnected Data Centers

• Technological advances now allow for workloads to be shifted between sites. “If you have a networked set of data centers, it becomes even easier to… coordinate operations,” enabling both grid support and service reliability. (31:11)
• Amy uses an analogy: “It’s as if I had human beings, electronic human beings, in different locations… I could tell one of them to take lunch at a different time… but we’d still meet our target.” (27:38, 33:44)
• Real-world demonstration: Oracle’s Phoenix data center reduced consumption by 25% at peak, simulated to prevent blackouts—a model for scale-up. (27:38)

7. Market & Investment Implications

• Private equity is pouring money into data centers and energy infrastructure, reminiscent of past energy “booms.” However, participants argue the AI-driven wave has more staying power due to the relentless growth of computation. (37:04, 38:00)
• AI’s energy needs could reach 12% of U.S. electricity by 2030, “maybe 25% mid-next decade” (38:00), far surpassing other compute revolutions because efficiency improvements are lagging compared to compute growth.

8. Barriers and Next Steps

• The path to grid-friendly, flexible AI centers requires:
  1. Demonstrations of flexibility at scale
  2. Regulatory reforms (faster/priority interconnection for flexible centers)
  3. Technical advances in orchestration, hardware, and software (47:20)
• Varun: “A virtual power plant of AI data centers… could be the most powerful demand-side solution we’ve ever seen.” (48:39)

Notable Quotes & Memorable Moments

• On urgency and stakes:
  “We have to move fast. It’s a Manhattan Project, number two.”
  — Amy Myers Jaffe, quoting Secretary Chris Wright (06:12)
• On the two cultures clashing:
  “The immovable object is the incumbent electric power industry… The unstoppable force is the AI and technology industries where… almost perfect 100% uptime.”
  — Varun Sivaram (11:55)
• On flexibility’s potential:
  “$4 trillion of new investments in AI data centers. If we can get this right on both sides.”
  — Varun Sivaram (13:10)
• On distributed data centers:
  “If you have a very well networked set of data centers, it becomes even easier… to pause workloads in one location and make sure they get picked up somewhere else.”
  — Varun Sivaram (31:11)
• Real-world analogy:
  “I could tell one of them to go take lunch at a different time than a different one, but we would still meet our target…”
  — Amy Myers Jaffe (33:44)
• On grid operator priorities:
  “People will be angry, regulators will lose their jobs, politicians will lose their jobs. Their hard constraint: they have to keep the lights on.”
  — Ed Crooks (29:30)
• On the investment frenzy:
  “We’re seeing, you know, massive, massive partnerships between private capital providers and energy developers… an insatiable thirst to finance these deals.”
  — Shanu Matthew (39:46)
• On the promise of flexible data centers:
  “A virtual power plant of AI data centers... most powerful demand side solution we’ve ever seen.”
  — Varun Sivaram (48:39)

Timestamps for Key Segments

• 00:01–04:05: Opening; stakes, guest intros, Varun’s background & Emerald AI’s mission
• 05:09–07:43: Recap of Council on Foreign Relations event, national security implications
• 09:00–13:28: The “flexibility” concept, research implications, overcoming industry inertia
• 14:15–16:36: Why tech companies are cautious about flexibility; investment logic
• 16:36–20:53: Regulatory precedents (Texas law), incentives to avoid forced shutdowns, real solutions emerging
• 25:07–27:38: Technical breakdown: timescales, solutions for spikes, real-world demonstration in Phoenix
• 27:38–35:22: Distributed computing analogy; resilience and security considerations
• 37:04–42:59: Investment cycle vs compute growth; can efficiency gains offset surging AI demand?
• 43:08–47:20: PJM’s NCBL proposal, market friction, policy tradeoffs
• 47:20–49:37: Steps for scaling flexible data centers; “virtual power plants” concept
• 49:37–50:41: Closing reflections and outro

Conclusion

The episode delivers a nuanced look at one of the decade’s biggest energy challenges: how the AI revolution’s insatiable data center appetite can be squared with grid stability, decarbonization, and affordability. While “flexibility” offers hope—both for accelerating data center deployment and protecting consumers—the solutions will require new industry mindsets, investment, technical innovation, and regulatory adaptation. If it works, the U.S. could lead the world in turbocharging AI while keeping the lights on—for everyone.