Podcast Summary: "AI Exchanges: Power Problems?" (Goldman Sachs Exchanges)

Date: April 2, 2026
Host: Alison Nathan (A), Co-Host: George Lee (B)
Guest: Brian Singer (C), Head of GSUSTAIN, Goldman Sachs Research

Episode Overview

This episode of "Goldman Sachs Exchanges" dives into the mounting power and resource demands triggered by the rapid expansion of AI and data centers. Hosts Alison Nathan and George Lee are joined by Brian Singer, head of GSUSTAIN, to break down the latest forecasts, challenges, and implications for companies, investors, and broader economies as hyperscalers pour unprecedented capital into AI infrastructure. The discussion bridges economic, technological, and political perspectives on how the world will keep up with AI’s voracious appetite for power.

Key Discussion Points & Insights

1. The Tremendous Growth in AI-Driven Power Demand

• Capital Spending Surge:
  • Hyperscalers' combined capital and R&D budgets for 2026–2027 have increased by over $300 billion.
    • “You’re growing off of a upwardly revised budget of $300 billion plus, and that is going to trickle down into power and a whole host of other uses.” — Brian Singer (01:23)
• Supply-Demand Imbalance:
  • Despite increased spending, a major supply-demand imbalance persists in compute and power (02:04).
  • Comparison to the shale revolution: a massive investment leading eventually to overcapacity, but AI is not there yet.

2. Drivers of Runaway AI Demand

• Enterprise & Agentic Machine Demand:
  • AI demand is driven not just by consumer interactions but also by “agentic” machine-to-machine operations.
  • “I think what we’ll see will ultimately swamp the amount of human AI traffic is agentic, machine-to-machine traffic. So I think that’s kind of the runaway demand driver.” — George Lee (03:38)
• Upward Revised Projections:
  • Expected 220% growth in global AI and non-AI data center power demand from 2023 to 2030 (prior forecast: 175%). This equates to adding a new “top 10” power-consuming country (03:56).

3. The Six Ps Framework: Growth Drivers and Constraints

• Brian Singer’s Six Ps:

  1. Pervasiveness of AI
  2. Productivity of models/servers
  3. Price of power
  4. Policy
  5. Parts
  6. People

• Some Ps fuel growth (pervasiveness, productivity), while others constrain it (people, parts, policy, price). (05:23)

4. The Human Capital Bottleneck

• Labor Shortages:
  • “We think we’re going to need to see 500,000 new jobs in the U.S. alone … 300,000 for building generation and 200,000 for transmission and distribution.” — Brian Singer (06:23)
  • The biggest concern is transmission/distribution electricians, who need 4 years to train and apprenticeships are lagging by 20k–25k.
  • If grid upgrades are slow, companies invest in “behind the meter” solutions.

5. Behind-the-Meter Solutions & Resource Mix

• Short-Term Workarounds:
  • Many data centers deploy on-site (behind-the-meter) natural gas generators due to grid constraints, even if they’re less efficient.
  • “We’re starting to see greater deployment of behind the meter solutions. And that may last for a period of time, probably not forever…” — Brian Singer (06:23–07:54)
  • These are mostly natural gas, split between more and less efficient types.
• Longer-Term Mix:
  • Until 2030: ~60% power from thermal (mostly natural gas), 40% renewables, plus some nuclear (12:30).
  • By mid-2030s, nuclear and renewables expected to increase as tech, policy, and supply chains mature.

6. Pricing, Policy, Politics, and Populism

• Rising Costs & Political Tensions:
  • Some states consider data center moratoria due to affordability concerns; risk of higher consumer prices.
  • Alignment is growing between hyperscalers, regulators, and communities for take-or-pay contracts to insulate consumers (09:27).
  • “No one wants to have to pay more, especially if their demand isn’t necessarily the precise cause of it.” — Brian Singer (09:27)
• Affordability & Impact:
  • Even if power becomes $40/MWh more expensive, it’d reduce hyperscalers’ EBITDA by just 2.5% by 2030, less than 1% on cash return—affordable for them, but critical for the public (10:34).

7. Financial Capacity & Risk

• Are Hyperscalers Overextending?
  • Currently redeploying up to 87% of cash flow plus R&D into CapEx—upward trend, but still manageable due to strong balance sheets (14:43).
  • Balance sheets remain flexible, with low net debt (15:44).
• Utilities’ Challenge:
  • Utilities lack the financial firepower of hyperscalers and rely on regulatory support, capital markets, and robust contracts (16:46).
  • Nuclear development is hindered by hesitance to be “first mover” (17:36).

Notable Quotes & Memorable Moments

• On AI Power Demands:
  • “To put what that means into perspective, that would be like adding another top 10 consuming country to the mix.” — Brian Singer (03:56)
• On Human Constraints:
  • “The number of electricians … four years of skilling … the number of energy apprentices that … needs to rise by another 20 to 25,000.” — Brian Singer (06:23)
• On Resource Mix Strategy:
  • “We are in a yes and environment here, not a no or environment.” — Brian Singer, on using all available energy resources (14:06)
• On Corporate Impact:
  • “That would only have an impact of about 2.5% on their 2030 EBITDA.” — Brian Singer, on increased energy costs for hyperscalers (10:34)
• On Nuclear & Utilities:
  • “Everyone wants to hold the door open for someone else … no one wants to be first, second, third, or fourth.” — Brian Singer, on nuclear investment hesitation (17:36)

Timestamps for Key Segments

• 01:23: Capital spending surge & its trickle-down effects
• 03:15: Runaway demand from agentic systems
• 03:56: Revised power growth forecasts; country-scale impact
• 05:07: Introduction of the Six Ps framework
• 06:23: Labor shortages & human capital constraints
• 07:54: Behind-the-meter solutions and efficiency tradeoffs
• 09:27: Political, pricing, and populist pressures; take-or-pay contracts
• 10:34: Affordability analysis & corporate profit impact
• 12:30: Energy source mix through 2030 and beyond
• 14:06: "Yes and" approach to energy sourcing
• 14:43: Financial health of hyperscalers; CapEx ratios
• 16:46: Utilities' capital challenges & nuclear project hesitation

Conclusion

The episode features a data-rich, nuanced exploration of the looming power and infrastructure challenges posed by exponential AI growth. The discussion balances optimism around technology and energy innovation with grounded awareness of policy, workforce, and market risks. The consensus: AI power demand will continue to outpace expectations, triggering strategic, technical, and political shifts as industry and society race to keep up.