MTS

Wiki/Physical layer

Power and the Grid

The binding constraint on AI buildout is not silicon. It is electrons, copper, and grain-oriented electrical steel. Chips ship in days. A 500 MW substation upgrade ships in five years. A 765 kV transmission corridor ships in a decade. This page assembles the primary-source evidence that the physical electricity system, not the supply of GPUs, sets the upper bound on how much AI capacity the United States can stand up before 2030.

Numbers below are drawn from the LBNL/DOE 2024 data center energy report, FERC Order 2023 filings, ISO/RTO planning documents, GE Vernova investor disclosures, Wood Mackenzie's 2025 transformer survey, and reporting from S&P Global, Utility Dive, and Latitude Media.

1. The macro forecast: LBNL/DOE 2024

The single most-cited recent forecast is Lawrence Berkeley National Laboratory's 2024 United States Data Center Energy Usage Report, prepared for DOE and released December 2024. The report is the first official update of LBNL's data-center energy series since 2016, and it is what every utility planner, ISO, and congressional staffer is referencing.

Key numbers (LBNL 2024 report, executive summary and Section 4):

Year U.S. data-center electricity Share of U.S. grid
2014 60 TWh 1.9%
2018 76 TWh 1.9%
2023 176 TWh ~4.4%
2028 (low) 325 TWh ~6.7%
2028 (mid) 440 TWh ~9.1%
2028 (high) 580 TWh ~12.0%

The 2028 range is unusually wide because AI is the dominant swing factor. LBNL breaks out AI-accelerated servers separately and finds that "servers for AI workloads" account for roughly 60–70% of the 2023-to-2028 growth — i.e., AI alone could grow from ~30 TWh (2023) to 165–325 TWh (2028), a 5–10x increase in five years. Conventional cloud and enterprise loads grow modestly; the spike is entirely accelerator-driven.

LBNL is explicit that this forecast assumes hardware ships and gets powered. It does not model interconnection queues, transformer lead times, or local opposition. Those are the constraints this page documents.

Source: LBNL 2024 United States Data Center Energy Usage Report (LBNL-2024-DCEUR), Dec 2024 — https://eta-publications.lbl.gov/sites/default/files/2024-12/lbnl-2024-united-states-data-center-energy-usage-report_1.pdf (Executive Summary pp. v–viii; AI breakouts Section 4 pp. 30–45).

DOE press release: https://www.energy.gov/articles/doe-releases-new-report-evaluating-increase-electricity-demand-data-centers.

For context: Grid Strategies' National Load Growth Report 2025 shows utility five-year forecasts have been revised upward in each of the past three years; the cumulative revision is roughly +128 GW of peak demand vs. the 2022 baseline forecast (Grid Strategies, Dec 2025).

2. Interconnection queues, by ISO

Every U.S. ISO is in queue crisis. The problem is asymmetric: generation queues are clogged with renewables that can't get studied; load queues — newer, less standardized — are clogged with data-center applications that may or may not be real.

PJM (Mid-Atlantic; 65 million people; the AI epicenter)

  • Load forecast. PJM's 2025 Long-Term Load Forecast projects +32 GW of peak load 2024–2030, with data centers accounting for 94% of the increase. Energy demand grows from ~810 TWh today to ~1,415 TWh by 2040 — a 72% jump, with data centers responsible for 422 TWh of the 605 TWh of new annual load. (PJM Inside Lines, 2025 Year in Review.)
  • Queue depth. PJM's Transition Cycle 1 cleared roughly 72 GW of interconnection requests in 2024; Transition Cycle 2 holds another ~46 GW, with a further 63 GW projected to be studied through 2026. The active backlog is on the order of 200+ GW of generation seeking interconnection — most of it solar and storage, with ~46 GW of approved-but-unbuilt projects sitting in commercial limbo.
  • Reforms. FERC accepted PJM's Order 2023 compliance filing in 2024, moving the queue to cluster studies. In Feb 2025, FERC ordered additional reforms (the "Reliability Resource Initiative") to fast-track shovel-ready dispatchable generation. Capacity prices in the July 2025 PJM auction cleared at $329.17/MW-day — near the cap — and the Dec 2025 auction cleared 134,479 MW with prices remaining at or near administrative limits. (PJM press release 20251217.)
  • Dominion / Northern Virginia. Dominion Energy in 2024 disclosed that connection wait times for new large loads in Loudoun County had stretched from a typical 3 years to 4–7 years, citing high-voltage delivery constraints into Ashburn rather than generation shortfalls (Bloomberg, Aug 29 2024). Campus requests escalated from ~30 MW to "several gigawatts" per site beginning in 2025. Dominion paused some new large-load connections through January 2026 and is preparing for a ~70 GW pipeline of pending large-load requests in northern Virginia alone (DCD; Data Center Frontier, 2025).

ERCOT (Texas; the most aggressive permitting regime; now reforming)

  • Large-load queue. ERCOT was tracking 226 GW of large-load interconnection requests as of Nov 18 2025, up from 63 GW in Dec 2024 — a ~3.6x increase in one year. Data centers are ~73% of large-load applications. (Utility Dive, "ERCOT's large load queue jumped almost 300% last year"; ERCOT System Planning Update Dec 2 2025.)
  • 2025 long-term forecast. ERCOT's April 2025 long-term load forecast (Item 8.1, April 7 2025) showed peak demand climbing from the 2024 record of 85.5 GW to ~278 GW by 2029 and ~368 GW by 2032 under the TSP-provided forecast. ERCOT itself warned regulators against treating these as guaranteed outcomes — explicit acknowledgment of phantom queue risk.
  • Senate Bill 6 (2025). Signed by Gov. Abbott on June 21 2025 (effective immediately). Key provisions, per Baker Botts, McGuireWoods, and Bracewell client alerts:
    • Loads ≥75 MW must register with ERCOT/PUCT.
    • Any new large load interconnected after Dec 31 2025 must install equipment allowing ERCOT to directly curtail the load during firm load-shed events.
    • PUCT must create a competitively procured reliability service from large loads ≥75 MW (a paid-curtailment product).
    • ERCOT may order large loads to deploy backup generation or shed load under "reasonable notice" during emergencies.
    • Co-located generation (behind-the-meter) is subject to new transparency and grid-impact studies before islanding from the grid.
  • Net effect of SB 6. Texas remains the fastest place to break ground but is no longer the freest place to operate: the implicit deal is "build fast, but accept curtailability." This is the template other states are studying.

MISO, SPP, CAISO, NYISO

  • MISO. ~353 GW across 1,696 projects in queue. MISO has refused stakeholder calls to pause the 2025 queue. Launched an "Expedited Resource Addition Study" (ERAS) in late 2025, second cycle accepting ~6.1 GW of largely battery projects. Long-Term Resource Adequacy (Sept 16 2025 SPC deck) flags load growth concentrated in Indiana/Ohio data center clusters.
  • SPP. Joint-planning agreement with MISO on the seam (the JTIQ — Joint Targeted Interconnection Queue), targeting ~$1.65bn of seam-region transmission. SPP launched its own ERAS process for ~13 GW of fast-tracked projects in 2025.
  • CAISO. Persistent queue indigestion; suspension rates 46–79% in late-stage projects (Enverus 2025 Outlook). California has no equivalent data-center cluster but is constrained by transmission to absorb new load growth in Silicon Valley.
  • NYISO. ~26.6 GW in queue across 189 projects; relatively small but constrained by transmission into downstate. NYISO has had a large-load interconnection process for years — the only ISO besides ERCOT with formal large-load rules pre-2025.

The phantom queue problem

The single most important caveat to every number above: a large fraction of queued load and generation is not real. Data-center developers routinely submit 5–10x more interconnection requests than they plan to build, in order to hold queue position across multiple utilities. RMI and Latitude Media estimate that total queued U.S. data-center load already exceeds 120 GW — roughly 2x any plausible 2030 buildout. One major U.S. utility reported that ~30% of 2024 load-interconnection applications were canceled. FERC Order 2023's tightened readiness deposits and site-control requirements are designed to fix this, but the cleanup will take until 2026–2027.

The practical implication is that announced GW is not deliverable GW. The gap is the key variable for any AI-capex forecast.

3. Transformers — the silent crisis

Large power transformers (LPTs) — the multi-hundred-ton step-down units between transmission and distribution — are the single most underappreciated bottleneck. Wood Mackenzie's Q2 2025 transformer survey is the canonical source.

Equipment 2020 lead time 2024–25 lead time 2026 expected
Standard substation power transformer ~50 weeks 120–128 weeks 128+ weeks
Generator step-up (GSU) transformer ~50 weeks ~144 weeks rising
Largest custom LPTs (HV) 80–100 weeks up to 210 weeks (~4 years) unchanged
Distribution transformers (pad-mount) 6–12 weeks 60–80 weeks partial recovery

Sources: Wood Mackenzie Q2 2025 power-transformer survey; NIAC Addressing the Critical Shortage of Power Transformers (June 2024 draft); DOE Large Power Transformer Resilience Report to Congress (July 2024, signed by Secretary Granholm); PowerMag "Transformers in 2026" (2025).

Why it broke

Three forces collided:

  1. Demand surge. Wood Mackenzie reports power-transformer demand up +119% since 2019 and distribution-transformer demand up +34%. The drivers are AI load, electrification, IRA-driven generation buildout, and end-of-life replacement (much of the fleet dates to the 1970s).
  2. Grain-oriented electrical steel (GOES). A narrow set of producers worldwide make the high-permeability silicon steel that goes into LPT cores. Cleveland-Cliffs is the only domestic producer; Nippon Steel, JFE, ThyssenKrupp, Posco round out the global capacity. None of them expanded materially during the 2010s.
  3. Domestic manufacturing base. The U.S. has roughly six dominant LPT suppliers — Hitachi Energy, Siemens Energy, Hyundai Electric (Montgomery, AL), Prolec GE / SPX (Waukesha, WI), Delta Star, and Pennsylvania Transformer Technology / "PTT" (Canonsburg, PA). Plus smaller players: Howard Industries (Laurel, MS — distribution and medium power), Virginia Transformer Corp (Roanoke), Wilson Transformer (Australian, but selling in the U.S.), and Niagara Transformer. These six suppliers hold ~70% of the U.S. power-transformer segment (Newton-Evans, mid-2024 assessment). Combined U.S. domestic LPT output is on the order of 300–400 units per year against demand pushing toward 600+ annually.

Strategic Reserve

DOE's July 2024 Large Power Transformer Resilience Report to Congress (signed by Secretary Granholm 7/10/24) restated the case for a federal Strategic Transformer Reserve. NIAC's June 2024 recommendation was to create a "virtual reserve" — the federal government as buyer-of-last-resort to smooth manufacturer order books — rather than a physical stockpile. As of mid-2025 nothing of consequence has been appropriated. GAO Report GAO-23-106180 documents the gap. Industry runs informal spare-sharing programs (the Spare Transformer Equipment Program, STEP, plus EEI's RESTORE).

The bottom line: even if every other constraint relaxed, the U.S. cannot install enough LPTs to power the announced data-center buildout before 2028.

4. Gas turbines — sold out through the decade

If grid expansion is blocked, the response is on-site or near-site generation. Gas turbines became the default. They are now also sold out.

GE Vernova (Greenville SC + Belfort, France)

  • Q4 2025: booked 18 GW of gas turbine orders in a single quarter.
  • Year-end 2025 backlog: ~80 GW, with combined backlog + slot-reservation agreements at ~100 GW (Utility Dive, Dec 2025: "GE Vernova gas turbine backlog hits 100 GW as prices rise").
  • Lead times: orders placed now deliver in 2028; CEO Scott Strazik told investors GE Vernova is "in discussions with customers about 2030 slots" and expects to be sold out through 2030 by end-2026.
  • Capacity expansion plan: 20 GW/yr by mid-2026, with potential stretch to 24 GW/yr by mid-2028 at Greenville and Belfort.

Source: Utility Dive "GE Vernova expects to end 2025 with an 80-GW gas turbine backlog that stretches into 2029" (Dec 2025); GE Vernova Q4 2025 earnings call; Bloomberg "GE Vernova Builds Gas Turbine Backlog to Meet Power Demand Boom" (April 23 2025).

Siemens Energy and Mitsubishi Power

Siemens Energy has reported similar dynamics: ~50 GW of gas-turbine orders in backlog as of late 2025, with delivery slots into 2029–2030. Mitsubishi Power (Takasago, Japan) has indicated comparable order books, with M501JAC HA-class units the volume product into the U.S. market. American Public Power Association's mid-2025 piece on turbine backlogs notes that ~80% of new heavy-duty gas turbine capacity orderable through 2030 is already committed to specific customers — the spot market for a 2028 delivery slot is essentially gone.

Implication: marginal data-center developers in 2026 cannot order a new combined-cycle plant and expect operations before 2030. They have three options — wait, buy used turbines (rare and small), or pivot to aeroderivatives (LM6000-class), reciprocating engines (Wartsila, Caterpillar), or fuel cells.

5. Regional snapshots

Northern Virginia (Loudoun / Prince William / Fauquier)

The world's largest data-center cluster — ~15 GW of operational + pipeline. Dominion's connection queue: 4–7 year wait for greenfield 100 MW+ campuses; some sub-areas (eastern Loudoun, around Ashburn) effectively closed to new connections through 2026. Virginia's State Corporation Commission approved Dominion's new GS-5 rate class in November 2025 — explicitly designed to make large-load customers fund their own infrastructure. The PJM-approved 765 kV MidAtlantic Resiliency Link (Putnam County, WV → Frederick County, MD, ~260 miles) and the Valley Link (Campbell County, VA → Fauquier County, VA, ~155 miles) are the core of the response. Jointly developed by AEP, Dominion, and FirstEnergy, total cost ~$4.6 billion within a broader $6.7bn 2024–25 RTEP cycle. Earliest energization for new 765 kV: 2030–2032.

        WV                            MD
         \                            /
          \--- 765 kV (260 mi) ------/
              MidAtlantic Resiliency Link
                                   |
                            Loudoun cluster (15+ GW)
                                   |
          /--- 765 kV (155 mi) ----\
         /                          \
       Campbell Co VA          Fauquier Co VA
              Valley Link

Texas (DFW, Houston, Abilene/Stargate, San Antonio, Austin)

SB 6 (above) is the regulatory frame. Physically, Texas has spare interconnection capacity in West Texas (where wind sites + Permian gas + Stargate-style campuses converge) but limited transmission east. ERCOT's 765 kV plan — the "Permian Basin Reliability Plan," approved by PUCT Oct 2024 — adds ~$30bn in HV transmission, energizing 2030–2035. Texas's competitive retail market and ERCOT's unique status outside FERC jurisdiction give it speed advantages but no escape from physical limits.

Arizona (Maricopa County)

APS told regulators in mid-2025 that if all proposed data centers in its service area materialized, demand would be 19,000 MW vs. ~8,200 MW historic peak — a 2.3x system multiplier. APS is "turning away potential data center customers." Data centers projected to account for 94% of APS load growth 2023–25. SRP introduced a Large Customer Integration Process (LCIP) in 2025 requiring data-center customers to fund upfront infrastructure studies and upgrades. Mesa, Chandler, and Phoenix have begun zoning data centers — Chandler City Council voted 7-0 in Dec 2025 to reject a $2.5bn AI data center.

Atlanta / Georgia Power

Georgia Power's 2025 IRP (approved by PSC July 15 2025, expanded Dec 19 2025) authorizes 9,885 MW of new capacity through 2031, ~⅔ gas. The plan includes 268 MW gas uprate at Plant McIntosh, five new gas plants, ~4,000 MW renewables, ~1,500 MW storage. Georgia Power projects ~8,500 MW of load growth over six years. The IRP keeps existing coal units online past prior retirement dates specifically because of data-center load uncertainty. (Utility Dive, Renewable Energy World, SACE.)

Memphis / TVA — xAI Colossus

xAI began Colossus operations in June 2024 powered by as many as 35 unpermitted gas turbines. A second site in Southaven, Mississippi added 27 turbines (~495 MW). Shelby County Health Department in July 2025 issued an air-quality permit covering 15 turbines / 247.2 MW at the Memphis site (through Jan 1 2027). Earthjustice + NAACP + SELC suing under Clean Air Act; SELC studies estimate ~1,700 tons/yr of NOx, the largest single industrial NOx source in greater Memphis. TVA negotiating long-term firm supply. (Inside Climate News, July 17 2025; CNBC; E&E News.)

Pacific Northwest — Quincy WA, Hillsboro OR (BPA)

Hillsboro data-center inventory grew >600% 2020–24; Quincy/East Wenatchee inventory doubled. ~3 GW of data-center load projected for Hillsboro alone, requiring multiple 230/500 kV upgrades by BPA and PGE — estimated $2 billion+. BPA's 2025 Cluster Study queue reached 65 GW; BPA stated that projects in the 2025 CS likely receive no long-term firm transmission until well after 2030. BPA proposed 13 new transmission projects, combined ~$3bn, and approved a FY 2025 grid-reinforcement budget of $590M.

6. Transmission buildout

Transmission planning historically lags load growth by 5–7 years; AI is accelerating that gap.

Region Project Cost Energization
PJM (VA/WV/MD) MidAtlantic Resiliency Link 765 kV, 260 mi ~$4.6bn (within RTEP) 2030–32
PJM (VA) Valley Link 765 kV, 155 mi included above 2030–32
PJM (overall) 2024 RTEP $6.7bn 2027–32
MISO–SPP seam JTIQ portfolio ~$1.65bn 2028–30
ERCOT Permian Basin Reliability Plan, 765 kV first-in-TX ~$30bn 2030–35
BPA 13 new transmission projects (PNW) ~$3bn 2028–32
MISO Long Range Transmission Planning LRTP Tranches 1+2 $20bn+ 2028–33

FERC's relevant orders:

  • Order 2023 (July 2023): cluster studies, "first ready, first served," readiness deposits, financial penalties on delayed studies ($1,000–$2,500/business day). Effective Nov 6 2023.
  • Order 2023-A (March 2024): rehearing clarifications.
  • Order 1920 (May 2024): long-term regional transmission planning — utilities must plan for 20-year horizons including extreme weather, electrification, and large-load growth.
  • FERC's Talen-Amazon ruling (Nov 2024): rejected an amended ISA that would have expanded behind-the-meter load from 300 to 480 MW at Susquehanna nuclear — chilling effect on co-located BTM nuclear deals nationally.

7. Behind-the-meter and the bypass economy

When the grid says "five years," developers route around it. Behind-the-meter (BTM) generation is the dominant response. Three categories:

  1. BTM gas turbines. xAI Memphis is the high-profile case; Crusoe (Abilene/Stargate), Wyoming projects, and Permian-adjacent campuses are scaling on stranded-gas + on-site turbines. Aeroderivative units (LM6000, SGT-A45) get permitted faster but have shorter lives.
  2. BTM nuclear. The frontier deal type. Talen-AWS at Susquehanna (1.92 GW PPA, restructured from BTM to FTM after FERC's Nov 2024 ruling); Constellation-Microsoft at Three Mile Island Unit 1 restart (~835 MW); Energy Harbor / Vistra deals on existing fleet. SMRs (Oklo, X-energy, NuScale) targeted at this market but not commercial pre-2030.
  3. BTM solar + storage. Limited to ~20–40% of a 24/7 AI campus load without massive over-build; viable as a hedge layer or for inference-heavy facilities with curtailable workloads.

Why BTM wins economically. A BTM project (i) skips most or all of the transmission/distribution wires charges that compose 40–60% of an industrial customer's bill in most U.S. jurisdictions; (ii) avoids ISO interconnection queue waits because no new grid-side service is required (or only a small standby connection); (iii) bypasses local political review for distribution upgrades; (iv) lets the developer time-shift fuel procurement.

Why regulators are pushing back. Stranded-cost concerns: if AWS pulls 1.9 GW off Susquehanna into a BTM arrangement, that nuclear capacity no longer serves PJM-pool customers, and those customers see their reliability margin shrink without compensation. Hence FERC's Nov 2024 ruling and the SB 6 transparency requirements in Texas.

8. The new flexible-load frameworks

Both ERCOT and PJM are formalizing demand response as a structural product, not a peak-shaving curiosity.

  • ERCOT Controllable Load Resource (CLR) + SB 6 reliability service. SB 6 (PURA § 39.170(b)) requires ERCOT to develop a competitively procured demand-reduction reliability service for loads ≥75 MW. Combined with the existing CLR product, this creates a paid market for AI workloads to curtail during emergencies. Inference-heavy and batched-training jobs are economically curtailable; real-time inference is not.
  • PJM Emergency Load Response Program (ELRP) / Capacity Performance. PJM allows demand-response resources to clear in the capacity auction. The July 2025 auction cleared with prices near the cap, signaling that demand-side flexibility now has six-figure $/MW-day value.
  • CAISO Demand Response Auction Mechanism (DRAM) and SPP's reserve-sharing programs are smaller but evolving.

The strategic question for hyperscalers is whether to pay a premium for firm power or to accept curtailment in exchange for faster interconnection and lower energy cost. Anthropic's published frontier-training schedules and Microsoft's reported "interruptible" training contracts in Texas suggest the latter is becoming standard for training, while inference remains firm.

9. The deliverable-GW gap

Synthesis: announced data-center pipeline through 2030 is on the order of 80–100 GW in PJM alone, 138 GW projected by ERCOT, plus tens of GW each in MISO, SPP, and the southeast utilities. Total announced > 300 GW. Deliverable physical GW by 2030, given the constraints above, is likely 60–110 GW. The gap will be closed by some combination of:

  • Cancellations and consolidations as financing tightens.
  • Acceleration of BTM and stranded-gas approaches.
  • Reclassification: announced 1 GW campuses delivered as 200 MW phase-1's with phase-2's slipping to 2031–33.
  • Capacity-market price signals high enough to bring older units out of retirement (Brandon Shores, Eddystone, parts of the coal fleet).
  • A small contribution from new nuclear (Vogtle 5+6 unlikely; restarts like TMI, Palisades, Duane Arnold yes).

The two numbers to watch through 2026 are (a) the actual cancellation rate in the PJM and ERCOT queues as readiness deposits come due, and (b) GE Vernova's 2026 booked-orders disclosures, which will reveal whether the gas-turbine market has cleared at 2030–31 delivery or whether developers are giving up and reallocating capital.

Selected primary sources

  • LBNL (2024). 2024 United States Data Center Energy Usage Report, LBNL-2024-DCEUR. https://eta-publications.lbl.gov/sites/default/files/2024-12/lbnl-2024-united-states-data-center-energy-usage-report_1.pdf
  • DOE (July 2024). Large Power Transformer Resilience Report to Congress, signed by Secretary Granholm. https://www.energy.gov/sites/default/files/2024-10/EXEC-2022-001242%20-%20Large%20Power%20Transformer%20Resilience%20Report%20signed%20by%20Secretary%20Granholm%20on%207-10-24.pdf
  • NIAC (June 2024). Addressing the Critical Shortage of Power Transformers to Ensure Reliability of the U.S. Grid.
  • FERC Order No. 2023 (July 2023) and Order 2023-A (March 2024). https://www.ferc.gov/explainer-interconnection-final-rule
  • FERC Order No. 1920 (May 2024), regional transmission planning.
  • PJM. 2025 Year in Review: Planning Prepares for Burgeoning Electricity Demand (Inside Lines). PJM Capacity Auction release, Dec 17 2025.
  • ERCOT. Long-Term Load Forecast Update 2025–2031 (Item 8.1, Apr 7 2025); System Planning and Weatherization Update (Item 16.2, Dec 2 2025); 89th Legislative Session — ERCOT Status (Sept 2025).
  • Texas SB 6 (2025), signed June 21 2025. Analyses: Baker Botts, McGuireWoods, Weil, K&L Gates, Bracewell, Balch & Bingham, Butler Snow (all client alerts, summer 2025).
  • Wood Mackenzie (Q2 2025). Power transformer lead-times survey.
  • Newton-Evans Research (mid-2024). Assessment of the U.S. Power Transformer Industry.
  • GE Vernova Q4 2025 earnings; Utility Dive, Bloomberg, Power Magazine, RTO Insider coverage.
  • Bloomberg (Aug 29 2024). "Virginia Data Centers Face Seven-Year Wait for Power Hookups, Dominion Says."
  • Inside Climate News (July 17 2025). xAI Memphis turbines. SELC, Earthjustice, NAACP litigation filings.
  • Grid Strategies (Dec 2025). Power Demand Forecasts Revised Up for Third Year Running.
  • RMI (2025). The Interconnection Queue Continues to Be a Barrier to American Economic Competitiveness.
  • Latitude Media (2025). "Phantom data centers are flooding the load queue."
  • Georgia Power 2025 IRP, approved Georgia PSC July 15 2025; expanded Dec 19 2025.
  • Talen Energy investor releases (June 11 2025) on AWS expansion; FERC Susquehanna ISA ruling (Nov 2024).
  • BPA OATT (effective Oct 1 2025) and 2025 Cluster Study documents.