Power surge: AI, renewable energy, and the future of electricity

Episode Summary

Executive Summary: The episode argues that generative AI is triggering a generational surge in electricity demand, led by data centers, electrification, and industrial reshoring. Goldman Sachs expects an all-of-the-above supply response—especially natural gas, renewables, and batteries—while noting major constraints in permitting, grid interconnection, equipment, and affordability. Europe faces a similar but larger electrification-driven opportunity and infrastructure challenge.

Main Topics: AI and data centers as a new driver of power demand (Priority: 5/5): The hosts explain that generative AI materially increases electricity use because AI queries and model training are far more energy intensive than traditional search and computing, making data centers a major new load on the power system. Broad-based acceleration in electricity demand (Priority: 5/5): Beyond AI, the discussion highlights electrification of transport, homes, and industry, plus reshoring and industrial growth as additional forces pushing U.S. and global power demand higher after years of stagnation. Supply mix: natural gas, renewables, and storage (Priority: 5/5): The panel argues new demand will be met through an all-of-the-above approach, with natural gas providing dispatchable reliability, solar and other renewables meeting decarbonization goals, and batteries helping balance intermittency. Utility investment, grid buildout, and financing (Priority: 4/5): Utilities are expected to face a major capex cycle to build generation and transmission, supported by debt, equity, and potentially direct contributions from data center customers through special contracts and rates. Europe’s parallel electrification and grid modernization challenge (Priority: 4/5): Alberto Gandolfi describes Europe as lagging the U.S. but facing a potentially larger demand inflection from RepowerEU, AI, and electrification, alongside a very old grid requiring massive transmission, distribution, and renewable investment. Nuclear and direct power deals for data centers (Priority: 3/5): Carly Davenport explains that while new regulated nuclear plants are unlikely in the U.S., data centers may secure direct contracts with unregulated nuclear plants and may explore small modular reactors in the 2030s. Key bottlenecks: permitting, equipment, and raw materials (Priority: 4/5): The panel identifies long permitting timelines, interconnection queues, equipment lead times, and copper/raw material constraints as the main risks that could slow the power buildout.

Key Arguments: Generative AI and data centers are creating a level of electricity demand growth not seen in decades, shifting power demand from flat to structurally higher. A ChatGPT search can use about 10 times more power than a Google search because AI queries require far more computation and model processing. U.S. power demand is expected to grow at roughly 2.5% CAGR to 2030, with data centers contributing about 90 basis points of that growth. Data center power demand is projected to rise 160% by 2030 versus 2023, implying a load large enough to rank as a top-10 global power-consuming country if isolated. Past efficiency gains in data centers masked demand growth, but those gains are slowing as the shift to cloud and hyperscale matures. Future supply will likely be an all-of-the-above mix, with Goldman estimating roughly 60% natural gas and 40% renewables for new generation. Natural gas is favored for reliability and dispatchability, while renewables remain essential because hyperscalers have carbon-reduction commitments. U.S. utilities will need a major capital cycle, including about 50 GW of new generation and billions in transmission and grid upgrades. Europe may see an even larger percentage increase in power demand over the next decade because AI overlays an ambitious electrification agenda. In Europe, grid modernization is urgent because the medium-voltage and distribution network is aging and underprepared for data center connections. Nuclear is unlikely to be a major source of new regulated utility-built capacity in the U.S., but direct deals with unregulated plants can serve data center demand. The biggest execution risks are slow permitting, long interconnection queues, constrained equipment supply, and the need to keep customer bills affordable.

Data Points: Power demand growth in U.S. to 2030: ~2.5% CAGR - Carly Davenport on Goldman Sachs' forecast for U.S. electricity demand growth through 2030. Data centers’ contribution to U.S. power demand growth: ~90 bps of the 2.4%-2.5% growth - Carly breaking down how much of total U.S. demand growth comes from data centers. Data center power demand growth: 160% by 2030 vs. 2023 - Goldman Sachs estimate for data center electricity demand. AI search power use: 10x a Google search - Brian Singer explaining why AI queries are much more energy intensive. Energy intensity reduction in newer AI servers: 85% lower vs. two generations ago - Brian citing efficiency improvements in newer high-powered servers. Historical efficiency gains: >10% per year reduction in energy intensity - Carly describing why data center electricity demand was previously masked. New U.S. power generation needed: 50 GW - Estimate of new generation capacity required to support data center growth. New power generation investment in U.S.: ~$50 billion - Based on Goldman Sachs cost assumptions for generation alone. Utility investment outlook: 40% higher cumulative investment over next 4 years vs. prior 4 years - Goldman Sachs utilities coverage forecast. Natural gas / renewables split: 60% natural gas, 40% renewables - Goldman’s expected mix for new U.S. generation tied to demand growth. Average transmission permitting period: ~4 years - Carly discussing how long it can take to get transmission projects approved. Interconnection queue timeline: 40 to 70 months - Time often required to connect generation projects to the grid. Data center build time: 2 to 3 years - Comparison showing power connection can take longer than the facility itself. Europe demand increase over 10 years: 40% to 50% - Alberto Gandolfi’s estimate combining electrification and AI/data centers. Europe electrification impact: 30% to 35% increase in power consumption - Estimated effect of RepowerEU-style electrification when fully completed. RepowerEU timing: 2033-2035 implied completion - Alberto says the 2030 plan is likely 3 to 5 years delayed. European grid age: More than 40 years old - EU data cited by Alberto on the condition of the grid. European transmission and distribution investment need: €800 billion over 10 years - Estimate for grid modernization in Europe. European renewable investment need: ~€850 billion - Estimate for additional renewable buildout in Europe.

Pivotal Quotes: "We think it's really going to be an all-of-the-above approach because we're coming off of a time of no power demand growth and now seeing this acceleration from a number of different areas." — Carly Davenport: On how new electricity demand will likely be met by multiple supply sources rather than one dominant solution. "Our projection is that data center power demand will rise 160% by 2030 relative to 2023 levels." — Carly Davenport: Goldman Sachs' central estimate for the scale of AI/data center electricity growth. "The inflection point is in sight." — Alberto Gandolfi: On Europe’s electricity demand turning higher as electrification and AI begin to bite.

Implications: Utilities, generators, and grid suppliers may enter a multi-year capex supercycle, while hyperscalers will shape power markets through direct deals and decarbonization demands. The key winners will be firms that can add reliable capacity fast; the key risks are delays, bottlenecks, and affordability pressure.