AI drives upward revisions to data center demand forecasts, with power and supply chains becoming the main bottlenecks

Overview

This report is a summary by J.P. Morgan’s hardware and networking team of key points from recent third-party industry conferences, with a core focus on AI-driven data center demand, U.S. power demand forecasts, evolving power supply models, supply chain lead times, and project execution risks. The report shows that AI infrastructure demand is changing the scale and number of data center campuses, with capacity per campus evolving from the traditional roughly 1 GW to more than 10 GW.

Core views

The core view is that demand for data center capacity continues to expand rapidly, with AI resetting the baseline for capital expenditures and driving a significant upward revision to forecasts for U.S. data center power demand by 2030. At the same time, power availability has become the most critical constraint, pushing projects toward off-grid gas, lower-power-cost regions, and some international markets. Long lead times for components, heavy electrical equipment, skilled labor, and grid interconnection imply that there may be a meaningful gap between proposed project pipelines and actual built capacity. Resistance from local and state governments is also increasing, raising the risk of project cancellations, delays, and execution issues.

Analysis framework

The report uses a third-party industry update synthesis approach, integrating perspectives from BloombergNEF, Gartner, and Datacenter Hawk webinars to cross-analyze demand forecasts, project pipelines, power supply constraints, component lead times, grid interconnection progress, and regulatory/community resistance.

Methodology notes

• Industry demand forecastingBase case and AI chip scenario

  A base case is formed using current project timelines and construction progress, while a more aggressive AI chip scenario is built using AI server forecasts from semiconductor vendors and OEMs.

  This framework is used to illustrate the potential upside elasticity of AI server demand on data center power demand, though the report also notes that the base case may still be conservative.

• Supply chain constraints analysisComponent lead-time framework

  Lead times are assessed separately for AI accelerators, optical modules, advanced packaging, HBM, DRAM/NAND, heavy electrical equipment, skilled trades, and grid interconnection.

  This framework is used to determine the real bottlenecks from planning to data center commissioning, showing that hardware, power supply equipment, labor, and grid approval all affect construction timing.

• Project execution validationGrid interconnection pipeline funnel

  Proposed capacity, contracted capacity, construction-approved capacity, and capacity actually under construction are compared in tiers.

  This framework is used to avoid equating proposed interconnection requests with actual construction progress, highlighting that the share actually under construction in the Texas project pipeline is very low.

Asset mapping & comparison

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

• AI server and accelerator supply chain

  Direct beneficiary of expanding AI data center demand

  Strengths

  AI is resetting the capital expenditure baseline, and the aggressive scenario suggests server demand could drive higher power and compute infrastructure buildout requirements.

  Weaknesses

  Although AI accelerator lead times are relatively short, long lead times for advanced packaging, HBM, and storage may constrain delivery pace.

  Comparison

  Compared with traditional data center construction, AI infrastructure is increasing single-campus capacity from about 1 GW to more than 10 GW.

  Risks

  If assumptions around hyperscalers’ infrastructure useful lives or other demand assumptions change, or if AI server forecasts fall short of expectations, demand could be below the aggressive scenario.

• Electric utilities and grid interconnection infrastructure

  Becoming the core constraint and a potential beneficiary in data center construction

  Strengths

  Under the base case, about 80% of new capacity still needs to be supported by the grid, driving demand for grid expansion and power equipment.

  Weaknesses

  Grid interconnection lead times are 48–84 months, and heavy electrical equipment lead times are 36–60 months, so actual construction speed may lag project filings significantly.

  Comparison

  The aggressive AI chip scenario requires the grid to absorb 19 GW per year, materially above the roughly 10 GW of annual additions in the base case.

  Risks

  Interconnection queues, equipment shortages, regulatory approvals, and local resistance could lead to project cancellations or delays.

• Off-grid gas and low-cost power regions

  Serving as alternative power supply and siting flexibility under grid constraints

  Strengths

  Off-grid gas campuses in West Texas are expanding rapidly, and lower-power-cost regions such as Northern Mexico are becoming more attractive.

  Weaknesses

  Off-grid power may face constraints related to fuel, environmental compliance, permitting, construction, and long-term reliability.

  Comparison

  In the base case, off-grid power supports about 20% of new capacity; in the aggressive scenario, on-site gas generation may play a larger role due to greater pressure on grid absorption.

  Risks

  Policy, emissions, gas supply, and cross-border construction risks may affect project feasibility.

• Data center developers and construction services

  Benefiting from an expanding project pipeline, but with increasing execution divergence

  Strengths

  Growth in project size and number creates long-term construction demand.

  Weaknesses

  An increasing number of projects are being handled by less experienced developers, while skilled trade lead times are 48–60 months, raising execution risk.

  Comparison

  The proposed pipeline is large, but only 2% of Texas’s proposed interconnection capacity is actually under construction, showing a sizable gap from planning to execution.

  Risks

  Rising opposition from local and state governments led to approximately $130 billion of projects being blocked or delayed in Q1 2026.

Key data

• U.S. data center 2030 power demand base case118 GWBloomberg’s base case based on current project timelines and construction progress, significantly up from the prior 2H25 forecast of 77 GW.
• U.S. data center 2030 power demand aggressive AI chip scenario160 GWBased on AI server forecasts from semiconductor manufacturers and OEMs.
• Annual capacity additions in the base caseApproximately 10 GW/yearUnder the 118 GW demand scenario, about 80% is supported by the grid, with the rest supplied off-grid.
• Annual grid absorption requirement in the aggressive scenario19 GW/yearThe report believes this target is difficult to achieve, so the role of on-site gas generation may expand.
• Proposed grid interconnection capacity in Texas434 GWGartner cites ERCOT data, covering all interconnection requests rather than only data centers, equivalent to about 5.1 times the state’s peak demand.
• Status of the proposed Texas interconnection pipeline31% contracted, 9% approved for construction, 2% actually under constructionEquivalent to 134 GW contracted, 39 GW approved for construction, and 11 GW actually under construction.
• AI accelerator lead time3–12 monthsRelatively short among IT-related components.
• Advanced packaging lead time24–36 monthsOne of the bottlenecks in the semiconductor supply chain.
• HBM lead time30–36 monthsA memory supply constraint for AI servers.
• DRAM/NAND lead time36–48 monthsLong lead times in the storage supply chain.
• Heavy electrical equipment lead time36–60 monthsIncluding transformers, turbines, switchgear, and similar equipment.
• Grid interconnection lead time48–84 monthsOne of the longest critical bottlenecks in the report.
• Scale of projects blocked or delayed in Q1 2026Approximately $130 billionNew data centers are facing resistance from local communities and state governments.

Impact & implications

From an investment perspective, the upward revision in AI data center demand is positive for industry chains related to AI servers, network equipment, optical modules, advanced packaging, HBM, storage, and power infrastructure, but constraints are expanding from computing hardware to power, grid interconnection, heavy electrical equipment, labor, and local permitting. Investment judgments should not rely only on the scale of proposed projects, but should also track contracted, construction-approved, and actually under-construction capacity, as well as whether the power supply plan is executable.

Risks

• Insufficient power availability and excessively long grid interconnection timelines may constrain the actual pace of data center commissioning.
• Long lead times for heavy electrical equipment, skilled labor, HBM, DRAM/NAND, and advanced packaging may create multi-stage supply bottlenecks.
• Proposed project pipelines may overstate actual construction progress, with only 2% of Texas interconnection requests actually under construction.
• Rising opposition from local communities and state governments increases the risk of project cancellations and delays.
• A growing share of inexperienced developers may raise construction execution risk.
• If hyperscalers change assumptions around infrastructure useful life, demand for new data centers may decline.

What to watch

• Whether U.S. data center power demand forecasts continue moving from the 118 GW base case toward the 160 GW scenario.
• Changes in the mix between grid-supplied and off-grid-supplied annual capacity additions.
• Conversion rates among proposed, contracted, construction-approved, and actually under-construction interconnection capacity in Texas and other core markets.
• Whether lead times for heavy electrical equipment, HBM, DRAM/NAND, advanced packaging, and skilled trades ease.
• Project progress in off-grid gas campuses in West Texas and lower-power-cost international markets such as Northern Mexico.
• Approval, restriction, cancellation, and delay trends from local and state governments for new data center projects.