U.S. AI power demand can generally be absorbed by new capacity, but PJM grid interconnection bottlenecks are the key exception

Overview

This report is part of HSBC's Powering AI - US series and focuses on whether AI and data center load growth will be constrained by U.S. grid interconnection bottlenecks. The report's core judgment is that U.S. grid bottlenecks can generally be managed, and new generation capacity is likely to keep rising and meet demand; however, PJM is an important exception because it has concentrated a large share of data center interconnection applications, interconnection processes have slowed materially, and capacity prices have risen sharply.

Core views

First, U.S. load growth expectations are at a 25-year high, but data center applications may be overstated; HSBC forecasts U.S. annual load growth of about 2-3% over the next five years, below the more than 5% annual growth implied by aggregated utility forecasts. Second, the U.S. interconnection queue is close to 2,000 GW, while actual new capacity of 60-70 GW in 2024/25 is expected to rise to about 100 GW by 2030; the key task for system operators is to separate genuine load demand from the most attractive new capacity additions. Third, CAISO and ERCOT have doubled and tripled annual new capacity over the past five years, respectively, showing that interconnection bottlenecks are not insurmountable. Fourth, PJM accounts for about 40% of U.S. data center applications, but new interconnections have not increased materially, and the time to connect new capacity has risen to more than eight years, making it the main regional bottleneck in the transmission of AI-related power demand.

Analysis framework

The report starts from the U.S. supply-demand framework, comparing load growth forecasts, interconnection queue size, annual new capacity, regional new capacity additions, and changes in PJM capacity market prices. It then uses PJM as a regional case study to explain the mechanisms behind longer interconnection timelines, capacity market failure, and rising costs, and proposes solutions at the process, technology, market, and responsibility-sharing levels.

Methodology notes

• Power system supply-demand analysisMatching load growth with new capacity additions

  Match the load growth from AI and data centers against new capacity from renewable energy, gas-fired generation, and the interconnection queue to assess whether grid bottlenecks will become a systemic constraint.

  HSBC uses a 2-3% annual load growth forecast for the next five years as a benchmark and compares it with the nearly 2,000 GW U.S. interconnection application queue, 60-70 GW of annual new capacity in 2024/25, and about 100 GW of expected new capacity by 2030.

• Regional electricity market analysisPJM capacity market and interconnection process diagnosis

  Analyze the mismatch between PJM's forward capacity procurement cycle and the actual interconnection cycle for new capacity.

  PJM's capacity market typically procures 2-3 years ahead, but the interconnection cycle for new capacity has stretched to more than eight years, leaving the capacity market reliant mainly on existing generation assets and pushing up capacity prices and total capacity costs.

• Policy and technology solution assessmentInterconnection process optimization and advanced transmission technologies

  Improve the ability of the existing grid to absorb new load through software, process changes, planning, and transmission technologies.

  The report discusses third-party interconnection study software, as available interconnection, fast-track application channels, top-down grid upgrade planning, dynamic line rating, advanced power flow controls, and topology optimization.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• U.S. electric utilities

  AI and data center load growth increases power demand and drives grid planning, the capacity market, and new generation investment.

  Strengths

  Utilities have tools to manage interconnection queues by raising upfront costs, optimizing processes, using modern technology, and deploying more staff and resources; the capacity addition improvements seen in CAISO and ERCOT provide positive examples.

  Weaknesses

  Grid constraints, oversized interconnection queues, and uneven load application quality slow project execution and increase planning complexity.

  Comparison

  CAISO and ERCOT have doubled and tripled annual new capacity over the past five years, respectively, while PJM's new interconnections have remained relatively stagnant.

  Risks

  If interconnection processes do not improve enough, load growth may materialize more slowly than expected, or rising capacity prices may trigger regulatory and political pressure.

• PJM capacity market and regional power assets

  PJM is one of the U.S. regions with the highest concentration of data center interconnection applications, and rising capacity prices directly affect power costs and generation asset returns.

  Strengths

  High capacity prices may improve returns for existing generation assets and strengthen the economic signal for new capacity and grid upgrades.

  Weaknesses

  The mismatch between an eight-year-plus interconnection cycle and a 2-3 year capacity market procurement cycle means market mechanisms cannot incentivize new supply in time.

  Comparison

  Compared with CAISO and ERCOT, PJM's new capacity relative to peak demand remains around 2-4%, and no similar acceleration has occurred.

  Risks

  Excessively high capacity prices may lead to price caps, long-term PPA substitution, regulatory intervention, or disputes over cost pass-through.

• Data centers and hyperscalers

  The expansion of AI computing power drives electricity demand, but data center projects need to participate more actively in generation and grid upgrade planning.

  Strengths

  Large hyperscalers have the ability to sign long-term PPAs, bring their own power, and share the costs of grid upgrades.

  Weaknesses

  Applications may be overstated by 5-10x, and speculative applications increase screening and interconnection burdens on the system.

  Comparison

  PJM's short-term demand forecast has been lowered because of interconnection delays, but long-term growth expectations remain higher.

  Risks

  If queue delays, upgrade costs, or regulatory requirements increase, some data center projects may be delayed, canceled, or relocated.

• Advanced transmission technologies and grid-enhancing technologies

  Technology solutions can improve utilization of the existing transmission network and ease the long cycle and high cost of traditional grid upgrades.

  Strengths

  Dynamic line rating, advanced power flow controls, and topology optimization may shorten interconnection times and release additional capacity.

  Weaknesses

  They require system operators to adopt them more systematically in the study process, and real-world implementation is affected by regulation, cost allocation, and operating standards.

  Comparison

  Compared with traditional case-by-case grid upgrades, advanced transmission technologies may increase capacity access faster and at lower cost.

  Risks

  Slow technology adoption, inconsistent review processes, or unclear division of responsibilities may limit the actual effect.

Key data

• U.S. annual load growth forecast2-3%HSBC forecasts U.S. annual load growth of 2-3% over the next five years, below the more than 5% annual growth expected in aggregated utility forecasts.
• U.S. interconnection queue sizeclose to 2,000 GWThe report says the U.S. grid interconnection queue is at a record high.
• U.S. annual new capacity60-70 GW in 2024/25, about 100 GW in 2030HSBC expects new capacity from renewable energy and gas-fired generation to rise, enough to meet 2-3% annual load growth.
• PJM share of data center applicationsabout 40%PJM covers several major U.S. data center hubs and is the power region with the highest concentration of data center interconnection applications in the United States.
• PJM connection cyclemore than eight years in 2024The average time from a PJM generation interconnection application to commercial operation has risen from less than four years in 2000-2020 to more than eight years in 2024.
• PJM capacity pricesUSD270 to more than USD300/MW/dayCapacity prices in 2025 and beyond are significantly higher than the 2022-24 average of about USD40/MW/day and also above the 2015-21 average of about USD110/MW/day.
• PJM capacity costsaverage about USD16bn in 2025-27Capacity costs are about twice the 2015-21 average of USD8bn; the 2022-24 average was about USD3bn.
• PJM long-term peak load outlookabout 3-4%/year over the next 10 years, about 2-3%/year over the next 15-20 yearsBased on PJM's 2026 demand forecast, but the report emphasizes that historical aggregated demand forecasts for PJM have been uncertain.

Impact & implications

For investment research, the report reframes the AI power theme from a story of insufficient U.S. supply overall to one about regional grid interconnection bottlenecks and capacity market design. Overall U.S. capacity additions may still support AI and data center load growth, so the value chain benefit thesis is not negated; however, in PJM, interconnection timelines, capacity prices, and cost allocation will directly affect data center project execution, hyperscaler power procurement strategies, utility planning responsibilities, and the returns of related power assets.

Risks

• U.S. data center load applications may be overstated by 5-10x, creating a gap between demand expectations and actual implementation.
• PJM has historically forecast annual demand growth of 1-3%, but actual demand has been broadly flat since 2010, highlighting uncertainty in load forecasts.
• PJM's interconnection cycle of more than eight years may continue to slow the rollout of new generation capacity and limit the capacity market's ability to incentivize new supply.
• Rising capacity prices and capacity costs may trigger regulatory intervention, price cap adjustments, consumer burden disputes, or changes to market mechanisms.
• If advanced transmission technologies, fast-track application channels, or delegated responsibilities do not advance sufficiently, the PJM bottleneck may persist.

What to watch

• Whether PJM speeds up its interconnection study process, and whether third-party software, staff, and resource投入 can materially shorten review times.
• Whether PJM adopts as available interconnection, fast-track application channels, and more systematic top-down grid upgrade planning.
• Whether PJM capacity prices remain elevated after 2025-27, and how capacity market price caps and long-term PPA policies change.
• Whether CAISO and ERCOT continue to lead in new capacity additions and provide replicable lessons for other regions.
• Whether hyperscalers take on more responsibility for their own power supply, long-term power purchase agreements, and grid upgrade costs.
• Whether PJM's demand forecast is revised again, especially the extent to which AI and data center demand is actually realized after short-term interconnection delays.