U.S. grid bottlenecks for AI-driven power demand growth are broadly manageable; PJM is the key exception

Overview

This report is the fourth installment in HSBC's 'Powering AI—U.S.' series, focusing on whether the growth in AI and data centers will be constrained by U.S. grid interconnection bottlenecks. The report's core view is that while overall U.S. grid constraints are evident, they can usually be managed by filtering out speculative applications, using modern technology, streamlining processes, and adding staff and resources; the real standout exception is the PJM region, which hosts a large share of data center applications, adds new power connections slowly, and has seen capacity market prices and costs rise sharply.

Core views

The report argues that the lengthening of overall U.S. interconnection timelines reflects simultaneous surges in demand-side and supply-side applications, and does not necessarily mean that AI and data center growth will be systemically blocked. HSBC expects new U.S. renewable and gas-fired capacity to rise from 60-70 GW per year in 2024/25 to around 100 GW per year by 2030, enough to match 2-3% annual load growth. The problem in PJM is more concentrated: data centers account for about 40% of applications, demand forecasts have been revised upward, new capacity connections have stalled, and interconnection timelines exceed eight years, causing the capacity market to fail to incentivize new supply in time and pushing up capacity prices and total costs quickly.

Analysis framework

The report uses a regional power-system comparison and bottleneck decomposition approach, comparing the U.S. overall with PJM, CAISO, ERCOT and other regions. It focuses on load growth forecasts, the credibility of data center applications, interconnection queue size, annual capacity additions, peak capacity addition ratios, PJM capacity market prices and capacity costs, and further distills actionable short-, medium- and long-term solutions.

Methodology notes

• Power system analysisInterconnection queue vs. capacity additions comparison

  Compare the scale of interconnection applications with actual annual capacity additions and the share of peak capacity to assess whether grid bottlenecks will create a material supply constraint.

  The report notes that the U.S. interconnection queue is close to 2,000 GW, but actual deployable capacity is far below the queue size; the key is to identify the more reliable 20-50% load applications and the optimal 300-500 GW of additional capacity over the next five years.

• Regional comparisonISO/RTO cross-market comparison

  Compare capacity additions and interconnection efficiency across PJM, CAISO, ERCOT and other power regions.

  CAISO and ERCOT have significantly increased capacity additions over the past five years, while PJM additions remain only 2-4% of peak demand, indicating that the PJM bottleneck is region-specific.

• Market mechanism analysisCapacity market incentive mismatch

  Compare the timing gap between PJM interconnection timelines and the capacity market's advance procurement cycle.

  PJM's interconnection timeline is about eight years, far longer than the capacity market's 2-3 year forward procurement cycle, so the capacity market mainly constrains existing generation resources, lifting capacity prices without bringing new supply online in time.

Asset mapping & comparison

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

• U.S. electric utilities

  Benefit from AI and data center load growth, but with significant regional differences.

  Strengths

  Overall demand growth and higher grid investment needs support capacity expansion, transmission upgrades and utility capital expenditure.

  Weaknesses

  If load applications are overstated or interconnection processes remain congested, investment pacing and cost recovery could come under pressure.

  Comparison

  Capacity additions have improved more visibly in CAISO and ERCOT, while PJM's new connections lag behind.

  Risks

  Rising capacity prices, regulatory intervention, cost pass-through disputes and forecast errors.

• Existing generation assets in the PJM region

  One of the direct beneficiaries of rising capacity market prices.

  Strengths

  Slow new capacity connections make the capacity market rely more heavily on existing generation assets, supporting capacity revenues.

  Weaknesses

  Over the longer term, if regulators cap prices more tightly or interconnection speeds up, excess returns could normalize.

  Comparison

  Compared with regions where interconnection is smoother, existing PJM resources have greater pricing power in near-term capacity scarcity.

  Risks

  Adjustments to capacity price caps, substitution by long-term PPAs and policy pressure to control costs.

• Data centers and hyperscale cloud providers

  Both a source of new load and the party bearing the pressure of interconnection and power security requirements.

  Strengths

  With capital strength and power procurement capacity, they can improve certainty through self-supplied power, long-term PPAs and participation in grid upgrades.

  Weaknesses

  In regions such as PJM, they face connection delays and higher power costs.

  Comparison

  Project execution risk is lower in regions with more flexible grids or faster capacity additions.

  Risks

  Higher application hurdles, rising cost burdens and longer queue times.

• Advanced transmission technology and grid software vendors

  May benefit from reforms aimed at improving interconnection efficiency in PJM and other regions.

  Strengths

  Dynamic line rating, advanced power flow control, topology optimization and third-party interconnection study software can shorten study times and unlock existing grid capacity.

  Weaknesses

  Adoption depends on regulation, RTO processes and utility execution capability.

  Comparison

  Compared with traditional large-scale transmission upgrades, advanced transmission technologies may relieve local bottlenecks faster and at lower cost.

  Risks

  Insufficient technology adoption, lack of standardization and lower-than-expected savings.

Key data

• Share of data center applications in PJMabout 40%PJM covers several of the U.S.'s major data center hubs and is the power region with the highest concentration of data center interconnection requests.
• PJM interconnection timelinemore than 8 yearsIn 2024, the average time from generation interconnection application to commercial operation in PJM rose from less than 4 years in 2000-2020 to more than 8 years.
• Size of the U.S. interconnection queuenearly 2,000 GWFar above the annual new capacity of 60-70 GW in 2024/25, reflecting a very large pool of applications that still needs screening and queuing.
• HSBC's U.S. load growth forecast2-3% per yearBelow the more than 5% annual growth implied by utilities' aggregate forecasts over the next five years; the report believes data center applications may be overstated.
• U.S. new capacity forecastabout 100 GW/year by 2030Up from 60-70 GW per year in 2024/25; the report believes this is sufficient to support 2-3% annual load growth.
• PJM capacity pricesabove USD270 and later above USD300/MW/dayApplies for 2025 and beyond, a significant increase from the roughly USD40/MW/day average in 2022-24.
• PJM capacity costsaverage about USD16bn/year in 2025-27Double the roughly USD8bn/year average in 2015-21; the 2022-24 average was about USD3bn/year.

Impact & implications

For investors, the implication of the report is not simply to turn bearish on AI-driven power demand growth, but to distinguish between nationwide capacity-supply capability and localized grid governance bottlenecks. The broader U.S. power value chain may still benefit from AI and data center load growth, but in PJM the combination of capacity prices, interconnection delays and transmission upgrade costs could reshape the set of winners: existing generation assets, resources that can connect quickly, advanced transmission technology providers, and hyperscale cloud companies with the ability to self-procure power are more likely to benefit, while projects dependent on long interconnection queues and large grid-upgrade costs face higher risk.

Risks

• Applications for U.S. data centers and AI load may be overstated, with actual load growth coming in below expectations.
• PJM's demand forecasts have been inaccurate in the past; if forecasts continue to deviate, capacity planning and price signals may be distorted.
• PJM's long interconnection timeline may continue to delay new generation capacity coming online, amplifying capacity price and electricity cost pressures.
• Capacity price caps, long-term PPAs or policy intervention could alter the earnings path for existing generation assets.
• Excessively high transmission upgrade costs could cause projects to leave the queue, weakening the pace at which new supply is delivered.
• If advanced transmission technologies, process optimization and delegation of responsibilities are not implemented effectively, bottleneck relief could be slower than expected.

What to watch

• Whether PJM accelerates its interconnection study process and adopts third-party software or adds resources.
• Whether PJM expands 'as available' interconnection, cuts unnecessary grid upgrades or allows more flexible load absorption.
• How policy tools such as fast-track applications, capacity market price caps and long-term PPAs are implemented in PJM.
• Whether the division of responsibilities among PJM, state governments, utilities and data centers changes, especially whether hyperscale cloud providers are required to self-supply power.
• Whether the gap between annual capacity additions, peak capacity share and interconnection timelines in CAISO, ERCOT and PJM narrows.
• Whether actual U.S. load growth is closer to HSBC's 2-3% forecast or to the higher growth expectations in utilities' aggregate forecasts.