Episode Summary
Executive Summary: The episode examines whether large data centers can be treated as flexible grid loads rather than fixed 24/7 demand. Guest Tyler Norris explains that very small curtailment levels could unlock massive new capacity—up to 98 GW—if load can shift away from system peaks. The discussion highlights the need for new tariffs, faster interconnection, and on-site/bridge power, while noting transmission and local deliverability remain major constraints.
Main Topics: Data centers as flexible grid load (Priority: 5/5): The core thesis is that large data centers are not truly inflexible; they may be able to curtail modestly, defer workloads, or use on-site power to reduce peak grid impact. Quantifying added capacity from flexibility (Priority: 5/5): Tyler Norris describes modeling across major U.S. balancing authorities showing that tiny curtailment allowances could enable tens of gigawatts of additional data center load. How curtailment would work in practice (Priority: 4/5): The conversation explores whether flexibility would resemble demand response, including notice periods, partial curtailment, weather-driven events, and shifting computational workloads across sites. Interconnection incentives and tariff design (Priority: 5/5): A key value proposition is faster time-to-power in exchange for flexibility, with emerging utility programs and tariffs beginning to formalize this tradeoff. Limits from transmission and deliverability (Priority: 4/5): Even if generation-side flexibility allows more load, local network constraints and deliverability requirements may still reduce what can actually be connected. Bridge power and temporary on-site generation (Priority: 4/5): The discussion notes that data centers already using temporary on-site generation or leased batteries could later convert those assets into low-utilization curtailment resources. Market and regulatory momentum (Priority: 4/5): Regulators and utilities are starting to require or request demand response participation from hyperscale loads, but many negotiations remain bilateral and confidential.
Key Arguments: Very small curtailment of new data center load can unlock large amounts of additional capacity because the grid is built for rare peaks, not average demand. The paper modeled 22 of the largest U.S. balancing authorities, covering about 95% of U.S. load, to estimate feasible additions under flexibility assumptions. At 0.5% curtailment, the model suggests up to 98 GW of new data centers; at 0.25%, up to 76 GW. Most curtailment would be partial, not full shutdowns: at 0.25% curtailment, curtailment would be needed in 85 hours, but in most of those hours at least half of the new load could remain online. Data center flexibility is more than just classic demand response because the load should be planned as flexible upfront rather than added later as firm load. Fast interconnection in exchange for flexibility is likely the most practical near-term market structure. Transmission and local deliverability constraints may erode some of the theoretical capacity gains, but the exact amount depends on jurisdiction-specific network conditions. On-site generation, batteries, and bridge power can provide a temporary path to both speed to power and later curtailment capability.
Data Points: Projected new data center capacity at 0.5% curtailment: 98 gigawatts - Estimated across 22 large U.S. balancing authorities, representing about 95% of U.S. load. Projected new data center capacity at 0.25% curtailment: 76 gigawatts - Modeled flexibility case for new data center load. Coverage of modeled balancing authorities: 22 balancing authorities / ~95% of U.S. load - Scope of the Duke University modeling exercise. Current U.S. data center load: sub-30 gigawatts - Used as a comparison to show how large the modeled additions are. Flexibility hours at 0.25% curtailment: 85 hours - Average number of hours per year when some curtailment would be required. Hours with at least 50% load retained: 73 of 85 hours - Most curtailment events would be partial rather than full. Hours with at least 75% load retained: 50 of 85 hours - Shows curtailment is often shallow. Average load factor across modeled balancing authorities: 53% - Average consumption over peak consumption. Unused power system capacity in 90% of hours: More than 30% unused - Illustrates that the grid is designed around rare peak events. Potential single driver of U.S. electricity growth: 44% of all U.S. load growth - Cited as an AI/data center growth forecast over the next five to seven years. Estimated curtailment duration of extreme peaks: 2.5 to 5 hours - Typical duration of the peak events driving flexibility needs. Grid-flexible device capacity from EnergyHub example: 3.4 gigawatts - 2.5 million devices aggregated into dispatchable virtual power plant capacity. Duke Energy requirement threshold: Above 100 megawatts - New hyperscale loads over this size will be required to participate in demand response.
Pivotal Quotes: "What really opened our eyes were some of these forecasts suggesting that AI specialized data centers are going to be likely the single largest driver of U.S. electricity load growth for the next five to seven years." — Tyler Norris: Explaining why flexibility research focused on data centers now. "At 0.5% flexibility or curtailment of the new data center load, you could add up to 98 gigawatts of new data centers across the U.S." — Tyler Norris: Summarizing the headline result of the modeling study. "I don't know where this idea came from of 24-7. They are not 24-7 loads, and it's critical to clarify this." — Tyler Norris: Clarifying that data centers have substantial non-compute flexibility.
Implications: If utilities and data centers formalize flexibility-based interconnection, grid capacity could expand much faster without waiting for full network buildout. Expect more demand response mandates, bespoke tariffs, and hybrid on-site/bridge power models.