Energy Gang Podcast Summary

Episode: How can the grid help AI, and how can AI help the grid? | Live from NYU at New York Climate Week
Date: September 30, 2025
Host: Ed Crooks, Wood Mackenzie
Guests: Craig Sundstrom (AWS), Josh Parker (Nvidia), Xijo (Wood Mackenzie), Amy Myers Jaffe (NYU)
Location: New York University, Climate Week NYC

Overview

This live Energy Gang episode tackles two urgent, intertwined trends: the explosive growth in electricity demand from AI and data centers, and the opportunities AI creates for decarbonizing and modernizing the electric grid. Broadcasting from NYU during Climate Week, the expert panel explores rising tensions around data center energy use, innovations in grid management, the future of clean energy supply (especially nuclear and small modular reactors), and the politics and economics reshaping the U.S. power sector. The lively discussion brings in perspectives from industry leaders at AWS and Nvidia, energy analysts, and academic researchers, offering depth and cutting-edge insight balanced by on-the-ground realism.

Key Discussion Points & Insights

AI’s Surging Electricity Demand: Economic and Political Fallout

• Historic Surge in Load:

  • Xijo highlights the scale of the change: “We have about 116 gigawatts of data centers… that's about 15% of the U.S. peak load, coming in the next two to three years.” (12:42)
  • For decades, U.S. electricity demand was flat; this sudden burst in new demand is a paradigm shift the industry isn’t prepared for.

• Economic Development vs. Public Backlash:

  • Josh Parker frames AI-driven electricity demand as an economic opportunity: “There seems to be broad consensus that having electricity demand driven by something valuable like AI is actually really good for, number one, the greening of the grid, and number two, economic development.” (00:01, 09:40)
  • Headlines and local pushback (such as the canceled Indianapolis data center project) reveal deep anxieties about rising electricity prices and environmental impact (01:09).

• Political Flashpoints and Blame Games:

  • Amy Myers Jaffe draws the analogy: “Is electricity going to be the new gasoline?” (16:27)
  • The panel agrees that pricing and reliability could become potent political issues, with tech companies, renewables, utilities, and policymakers all potential scapegoats (16:52).

Rate Design, Tariffs, and the Regulatory Gauntlet

• Major Issue: Who Pays for Infrastructure?

  • Craig Sundstrom (AWS): “Rate design is now the spiciest issue in data center development as far as energy goes.” (00:34, 19:10)
  • Utilities are experimenting with minimum demand charges, longer contract terms, and other mechanisms to allocate the costs of massive infrastructure upgrades (21:32).

• Fragmented, Complicated System:

  • “Every state is different… you have different venues weighing different issues.” (23:58 – Craig Sundstrom)
  • Even experts note the complexity: "You might need a degree to read [the power bill]." (22:50 – Xijo)

Supply Crunch: Technology Solutions and Supply Timelines

• New Tech: Small Modular Reactors (SMRs) & Renewables

  • AWS is “moving into the nuclear space, particularly through SMRs… deploying 5 gigawatts of new advanced SMRs by the mid-2030s.” (07:23 – Craig Sundstrom)
  • But Xijo warns, “We need that energy, like right now… Our first SMR unit in the U.S. comes online in 2031. …But we still have the problem of the next five years.” (16:18, 51:05)
  • Clean energy (solar, wind, batteries) is still expensive and slow to build in the U.S. compared to other regions (14:52).

• Efficiency Gains off the Charts, but Not Enough Alone

  • Josh Parker (Nvidia): “Over the past decade we’ve… reduced the energy consumed for a particular inference task by 100,000 times.” (25:01)
  • Yet, absolute demand rises as efficiency improvements make new AI uses feasible (Jevons paradox) (26:06).

AI as Grid Optimizer & Flexibility Enabler

• Grid Coordination & Market Efficiency

  • AI streamlines system operations, from speeding up interconnection queues (GridUnity-AWS + Southwest Power Pool) (29:51), to optimizing battery storage dispatch (47:30).
  • Amy Myers Jaffe lists energy economy-wide optimizations enabled by AI: building HVAC, logistics (e.g. ‘Waze for shipping’), digitized permitting, and manufacturing (28:08).

• Flexibility: Data Centers as Demand Response Assets

  • Xijo shares data: a Meta data center dropped from 200 MW to 20 MW during a price spike, showing demand is already more flexible than assumed (34:03).
  • Josh Parker: “If you can ramp down by 25%, potentially unlock 100 gigawatts of energy… [data centers] become grid assets.” (38:38)
  • Craig Sundstrom is cautious: flexibility can't be one-size-fits-all since many essential services require high uptime (44:00).

Barriers and Skepticism

• Tech is Not a Silver Bullet

  • Some applications of AI are already delivering real-world energy gains, but student exercises with large language models highlight continued inconsistency, need for guardrails (32:24 – Amy Myers Jaffe).
  • Industrial/commercial applications of AI (e.g., digital twins) show more consistent gains, but consumer-facing AI is less reliable (33:04 – Josh Parker).

• Location, Regulation, and the Limits of Decentralization

  • Only 6% of data centers are sited with their own generation (“co-location”) due to regulatory, technical, and economic barriers (58:12 – Xijo).
  • Centralized grids persist because they remain efficient for large, inflexible loads (59:06).

Future Opportunities and Challenges

• AI and Grid Modernization

  • Panelists agree that AI is “perfectly poised” to optimize renewable-variable and battery-rich grids—if policies, market structures, and investments keep pace (45:18, 47:30).
  • There’s cautious optimism: “I like to be optimistic, and I think we have more reason to be optimistic than ever before.” (45:18 – Josh Parker)

• SMR & Advanced Nuclear: A Long Play

  • AWS is taking a first-mover, risk-sharing approach—equity investment in technology and specialty fuels, partnerships with utilities (49:00-50:44).
  • But meeting near-term (5-10 year) demand remains a major risk—potential for a “power supply crisis” if rapid action isn’t taken (51:36, 53:07 – Xijo).

• Thermal Networks: Turning Waste to Community Value

  • Possibilities for using data center heat in district heating networks, making data centers “assets to the communities where they’re at.” (67:01 – Josh Parker)

Notable Quotes & Memorable Moments

• On the Moment’s Tension and Opportunity

  • “There seems to be broad consensus that having electricity demand driven by something valuable like AI is actually really good for… the greening of the grid [and] economic development.”
    — Josh Parker (00:01, 09:40)

• On Load Forecasting Difficulties

  • “I don't think anyone… has been able to, with any level of certainty, project that far out. But we're trying to get as good as we can on that question.”
    — Craig Sundstrom (21:32)

• On Explosive Growth and Potential Crisis

  • “We have about 116 gigawatts of data centers… that’s about 15% of the peak load in the U.S.… a giant mentality shift for the industry.”
    — Xijo (12:42, 51:05)

  • “We might have a power supply crisis if we don’t do something differently.”
    — Xijo (51:36)

• On Data Centers as Flexible Load

  • “If just look at that data, it went from 200 megawatt load to 20 megawatt load… led us to believe is that probably data centers are more flexible… than we traditionally believe.”
    — Xijo (34:03)

• On Rate Design

  • “Rate design is now like the spiciest issue in data center development… as far as energy goes.”
    — Craig Sundstrom (19:10)

• On Jevons Paradox in Practice

  • “As the chips become more powerful and more efficient, people think of new things to do with them… we end up actually using more energy.”
    — Ed Crooks (26:06)

• On Practical AI Energy Benefits

  • “They’re seeing 30 to 40% reduction in energy just by using AI to optimize how the building is heated and cooled.”
    — Josh Parker (28:08)

• On Thermal Networks

  • “They use district heating to take the heat from the data center and to heat the local communities… data centers will actually become not only assets to the grid, but assets to the communities.”
    — Josh Parker (67:01)

Timeline of Major Segments

| Timestamp | Segment | |------------|----------------------------------------------------------------------------------------------------------| | 00:01 | Intro: AI as driver for economic growth and grid greening (Josh Parker) | | 09:40 | Public anxieties: Grid constraints, canceled data centers, political blame (Ed Crooks, Amy Myers Jaffe) | | 12:42 | Scale of data center demand and industry adaptation challenges (Xijo) | | 19:10 | Rate design, tariffs, and cost allocation for new infrastructure (Craig Sundstrom) | | 25:01 | Nvidia’s leaps in chip energy efficiency; Jevons Paradox; context for net load growth (Josh Parker) | | 28:08 | Broader AI-enabled energy savings across sectors (Amy Myers Jaffe, Josh Parker) | | 29:51 | Case study – AI expediting interconnection in power markets (Craig Sundstrom) | | 34:03 | Data center load flexibility during price spikes, new market dynamics (Xijo) | | 38:38 | Startups turning data centers into grid assets – flexibility and “pre-emptible” workloads (Josh Parker) | | 47:30 | Batteries, virtual power plants, and AI-enabled grid optimization (Amy Myers Jaffe, Josh Parker, Craig Sundstrom) | | 49:00 | AWS’s investments in SMRs and First-of-a-Kind deployment challenges (Craig Sundstrom, Amy Myers Jaffe) | | 51:05 | Near-term power supply crunch vs. long-term nuclear build-out (Xijo) | | 58:12 | Limits of microgrids and behind-the-meter deployment for data centers (Xijo) | | 67:01 | Thermal energy networks: Turning data center heat into a resource (Josh Parker) |

Audience Q&A Highlights

• Decentralization & Microgrids: Only 6% of data centers use co-location with power assets; remaining regulatory and economic roadblocks (58:12–59:06).
• Geothermal’s Future: Seen as promising, but more a 2030s technology than a fix for the current 5-year crunch (60:54).
• International Data Center Buildout: While Southeast Asia and Latin America offer cheaper build and energy costs, constraints still hinge on power availability and regulatory frameworks, not just CAPEX (61:44, 64:15).
• Thermal Energy Networks: Growing opportunity to recycle data center waste heat to community heating (67:01).

Conclusion

This episode captures a turning point: AI and data center growth are no longer niche technical questions—they have become headline political, economic, and environmental issues. The dialogue revealed both a sense of “perfect storm” opportunity and an undercurrent of urgency and risk: grid bottlenecks, regulatory lag, infrastructure cost, and potential backlash. Key takeaways include the need for cross-sector partnership, accelerated investment in new technology, smarter regulation, and creative market design—backed by the optimism that AI and digitalization could, if harnessed well, deliver a cleaner, more efficient, and more resilient energy system.