Energy & power systems

Energy & power systems

Industry ID: 05 | Slug: energy-power | Researched: 2026-09-15 | Analyst: agent

1. Definition and boundaries

In scope. Electricity generation across all fuels (gas, coal, nuclear, hydro, wind, solar, geothermal, fuel cells, fusion); transmission and distribution networks and the equipment that builds them; grid interconnection processes and queues; wholesale and retail power markets, capacity markets and pricing; resource adequacy and bulk-system reliability; natural gas and oil upstream/midstream where it bears on power; energy storage; and — the defining question of this sector in 2026 — data-centre electricity demand and interconnection.

Explicitly out of scope. Carbon markets and carbon pricing, ESG and sustainability reporting, and climate adaptation and resilience finance. Those belong to sector 20 (climate/sustainability). Transport electrification demand sits with sector 05 only as a load input; the vehicles and batteries themselves belong to the EV and battery sector. Semiconductor manufacturing, including the chips inside data centres, belongs to semis. Data-centre construction, cooling and real estate belong to digital infrastructure; only their electrical interconnection and power procurement are ours.

Boundary disputes worth naming.

1. The climate/energy split is the central one. This sector was deliberately separated from climate/sustainability on the thesis that their capital flows are diverging. Section 8 tests that thesis against evidence and reports a qualified verdict: the driver has diverged decisively, the deployment volume has not yet.
2. Data centres sit across four sectors at once. The power question is ours; the compute, the buildings and the AI business models are not. Analysts who treat "AI power demand" as a single sector produce double-counted market sizes.
3. Behind-the-meter generation is contested between this sector and distributed-energy/cleantech. We claim it, because in 2026 the decisive facts about it are interconnection rules and curtailment obligations, not the technology.
4. Oil. Only the power-relevant part. Brent at $91/bbl matters here because it moves gas and retail bills; refining margins do not.

2. Subcategories

3. Market structure

Concentration varies by layer, and margin sits where scarcity sits.

• Networks are regulated monopolies. Transmission and distribution earn an allowed return on rate base. This is the least glamorous and most reliable margin in the sector, and it is expanding: EEI members plan $1.4 trillion of capital investment through 2030, up from a prior $1.1 trillion plan, with 2026 spending projected at $238.8 billion against $204.1 billion in 2025 — a 17% increase and a fourteenth consecutive record year (EEI, 2026-05-27). Growth in rate base is the business model, which means utilities have a structural interest in a large demand forecast.
• Generation is fragmented but regionally concentrated. In competitive markets, merchant generators compete into RTO auctions; in vertically integrated states, utilities own the plant. Pricing power currently sits with whoever owns dispatchable capacity in a constrained zone.
• Equipment manufacturing is a tight oligopoly, and this is where the pricing power has moved. Three firms — GE Vernova, Siemens Energy, Mitsubishi Power — supply large-frame gas turbines, and their output is effectively committed through 2030. GE Vernova's gas slot reservation agreements rose from 100 GW to 116 GW in a single quarter with a 125 GW year-end target; total backlog reached $176 billion on Q2 orders of $24.2 billion, up 88% organically; and it raised 2026 free cash flow guidance to $11.5-12.5 billion from $6.5-7.5 billion (GE Vernova, 2026-07-22). Transformers are similarly concentrated: power transformer lead times of 128 weeks, GSU transformers 144 weeks, with prices up 77% and 45% respectively since 2019 (POWER, 2026-01-02).
• Fuel supply is competitive and currently abundant. US natural gas production is forecast at a record 122.5 Bcf/d in 2026, up from 118.5 Bcf/d in 2025, with Henry Hub at $3.43/MMBtu (EIA, 2026-08-12; STEO 2026-09-09). Cheap gas is the reason the US electricity price problem is a network and capacity problem rather than a fuel problem — unlike Great Britain, where Ofgem attributed the October 2026 cap rise squarely to conflict-driven wholesale gas.

Barriers to entry are exceptionally high and have risen: interconnection (median over five years from request to operation), equipment lead times (2.5-3 years), capital intensity, and — new in 2026 — large-load tariff collateral and minimum-take obligations.

Where margin actually sits, ranked: (1) grid equipment manufacturing, temporarily, on scarcity; (2) regulated network rate base, durably, on volume; (3) dispatchable generation in constrained zones; (4) fuel supply, thinly; (5) merchant renewables, thinnest and most policy-exposed.

Market-size figures. We deliberately cite investment and revenue flows with named modellers rather than a synthetic "market size". BloombergNEF put global energy transition investment at $2.3 trillion in 2025, up 8% (BNEF, New Energy Outlook 2026, 2026-05-19) — a commercial research estimate, not an audited figure, and its deal-level basis is not publicly auditable. EEI's $1.4 trillion is a member-reported industry plan from a trade association that lobbies for rate-base growth; treat it as intent, not commitment. PJM's 2028/29 capacity auction cost of $16.4 billion for one delivery year is an audited market outcome and is the most trustworthy single number of the three.

4. Who matters

Leading companies. GE Vernova (https://www.gevernova.com/) · Siemens Energy (https://www.siemens-energy.com/) · Mitsubishi Power (https://power.mhi.com/) · Hitachi Energy (https://www.hitachienergy.com/) · Constellation Energy (https://www.constellationenergy.com/) · NextEra Energy (https://www.nexteraenergy.com/) · Dominion Energy (https://www.dominionenergy.com/) · Exelon (https://www.exeloncorp.com/) · American Electric Power (https://www.aep.com/) · Talen Energy (https://www.talenenergy.com/) · Vistra (https://www.vistracorp.com/) · Westinghouse (https://www.westinghousenuclear.com/) · Bloom Energy (https://www.bloomenergy.com/)

Notable startups. Fervo Energy (https://fervoenergy.com/) — enhanced geothermal, first non-recourse project financing for a first-of-a-kind geothermal plant · Commonwealth Fusion Systems (https://cfs.energy/) · TerraPower (https://www.terrapower.com/) — first NRC construction permit for a commercial non-LWR · X-energy (https://x-energy.com/) · Kairos Power (https://kairospower.com/) · Crusoe Energy Systems (https://www.crusoe.ai/) · Holtec International (https://holtecinternational.com/)

Active investors. Brookfield Asset Management (https://www.brookfield.com/) · Cameco (https://www.cameco.com/) · Google and Amazon as strategic investors and offtakers · the nine commercial banks underwriting Fervo's Cape Station facility (Barclays, BBVA, HSBC, MUFG, RBC, Société Générale, Bank of America, J.P. Morgan, Sumitomo Mitsui Trust) · DOE Loan Programs Office, functionally the largest single lender to US nuclear.

Platforms and standards bodies. PJM (https://www.pjm.com/) · ERCOT (https://www.ercot.com/) · MISO (https://www.misoenergy.org/) · CAISO (https://www.caiso.com/) · SPP · NYISO · ISO-NE · NERC (https://www.nerc.com/) · ENTSO-E (https://www.entsoe.eu/) · IEEE.

Regulators. FERC (https://www.ferc.gov/) · NRC (https://www.nrc.gov/) · PUCT (https://www.puc.texas.gov/) · PUCO (https://puco.ohio.gov/) · Ofgem (https://www.ofgem.gov.uk/) · ACER · China NEA (https://www.nea.gov.cn/).

Research institutions. Lawrence Berkeley National Laboratory EMP (https://emp.lbl.gov/) · Duke Nicholas Institute (https://nicholasinstitute.duke.edu/) · EPRI (https://www.epri.com/) · Monitoring Analytics, PJM's independent market monitor (https://www.monitoringanalytics.com/) · Idaho National Laboratory · MIT CEEPR.

Trade organisations. Edison Electric Institute (https://www.eei.org/) · Electric Power Supply Association (https://epsa.org/) · Nuclear Energy Institute (https://www.nei.org/) · SEIA (https://www.seia.org/) · American Clean Power (https://cleanpower.org/) · Data Center Coalition (https://www.datacentercoalition.org/).

Consumer and civil-society groups. Citizens Utility Board · Ohio Manufacturers' Association · Sierra Club · IEEFA (https://ieefa.org/) · GridLab · Global Energy Monitor (https://globalenergymonitor.org/) · NRDC · state consumer advocates including the Maryland Office of People's Counsel.

Named individuals. John Moura (NERC, Director of Reliability Assessments) · Pablo Vegas (ERCOT, CEO) · Drew Maloney (EEI, President and CEO) · Stu Bresler (PJM, COO) · Todd Snitchler (EPSA, President and CEO) · David Ulrey (Fervo, CFO) · Nikhil Kumar (GridLab).

5. Products, business models, technologies, customers

Major products. Bulk electricity (energy, $/MWh); capacity (the obligation to be available, $/MW-day); ancillary services (frequency and voltage support); transmission and distribution service; power purchase agreements and clean-attribute certificates; generation and grid equipment; and — genuinely new in 2026 — interconnection priority itself, which is being traded as an asset.

How money is made today, and how it is changing.

The classical model is a regulated return on network rate base plus merchant energy and capacity margin. Three changes are live:

1. Capacity revenue has become dominant in constrained markets. PJM's 2028/29 auction cost $16.4 billion against $2.2 billion for the 2024/25 delivery year, and the market monitor attributes $29.4 billion of additional capacity revenues across the last four auctions to data-centre load growth. A 10 MW industrial customer's monthly capacity charge rises from about $6,000 in 2024 to about $70,000 in 2028.
2. Speed-to-power is being priced as a separate product. Bloom Energy's pitch to data centres is a 55-day deployment against five-year grid interconnection; fuel cells are gas-burning assets sold at a premium for time, not carbon. Slot reservation agreements do the same for turbines.
3. Flexibility is becoming a contractual product. FERC's RM26-4 docket asks whether flexible large loads accepting curtailment could clear interconnection studies in 60 days, and its June 2026 show-cause orders require every RTO to create a flexible-load transmission service class.

Technologies that matter. Heavy-duty combined-cycle gas turbines; high-voltage transformers and GSUs, and the grain-oriented electrical steel upstream of them; lithium-ion grid storage (US capacity reached nearly 52 GW by June 2026 after 70% average annual growth over three years, EIA 2026-08-07); light-water reactor restarts and uprates; enhanced geothermal drilling; solid-oxide fuel cells; grid-forming inverters and synchronous condensers for stability; HVDC; and the load-modelling and protection-setting work NERC now requires of large computational loads.

Customer segments and what they buy on. Hyperscalers and AI developers buy on speed and firmness, then price — they will pay a large premium to energise a year earlier, and they are now buying 24/7 clean attributes as a secondary criterion (Google's 396 MW Fervo contract, Microsoft at Crane, Google at Duane Arnold). Industrials buy on price and reliability and are the constituency most damaged by capacity price increases they did not cause. Residential customers buy nothing — they are captive, which is why the affordability question resolves politically rather than commercially. Utilities buy equipment on delivery date, not price.

6. Geography

Demand growth is concentrated in a small number of places: Northern Virginia (the densest data-centre cluster in the world, and the site of the July 2026 3.8 GW load-trip event), Texas, Ohio, Georgia and Arizona in the US; and globally in China, India and Southeast Asia. The IEA forecasts 3.6% average annual global demand growth to 2030 after 3% in 2025, with emerging economies accounting for nearly 80% of additional consumption and China alone close to 50% (IEA, 2026-03-26).

Capital concentrates in the US and China. US investor-owned utilities alone plan $1.4 trillion through 2030. Europe remains the structurally weak leg, consistent with the macro picture of European venture funding.

Manufacturing is more dispersed than the demand: gas turbines from the US, Germany and Japan; transformers from Switzerland, Germany, Mexico, Korea and increasingly the US as roughly $1.8 billion of announced North American expansion comes online 2027-2029; solar modules overwhelmingly from China, which is now a tax-credit disqualifier in the US under the January 2026 foreign-entity-of-concern rules.

Regulation is the most fragmented layer and increasingly the decisive one. The US splits federal (FERC, NRC) from state (PUCT, PUCO), and in 2026 states led federal action on large loads — Texas SB 6 and the Ohio tariff preceded FERC's show-cause orders.

Non-US market: China (regional sources). China's National Energy Administration reported in September 2026 that installed solar capacity surpassed coal capacity for the first time (Global Times/NEA, 2026-09-01). That headline needs the counterweight: Global Energy Monitor found China curtailed 360 TWh of wind and solar in H1 2026, up 49% year on year — more than the entire increase in Chinese power demand over the period — while commissioning 30 GW of new coal (up 43%) against 2.7 GW retired, with a 274 GW coal pipeline and coal generation up 3.4% (GEM, 2026-08). The lesson generalises: installed capacity is not delivered energy, and a capacity milestone can coexist with rising coal output.

Non-US market: Great Britain and Europe. Ofgem raised the October 2026 cap 4% to £1,723, naming Middle East conflict-driven wholesale gas, with gas bills up 8% and electricity flat only because the government removed VAT (worth roughly £45). ENTSO-E's March 2026 expert panel report on the April 2025 Iberian blackout attributed it to oscillations, gaps in voltage and reactive power control, differing regulation practices and rapid generator disconnections — and concluded the remedies are "already technologically deployable", i.e. operational and coordination failures rather than a technology gap.

7. Historical trend patterns

Over 25 years this sector has cycled through a series of confident forecasts that did not happen. Being specific about those is what separates analysis from news.

• The 2000-2002 merchant gas build-out and bust. Deregulation triggered a construction boom on forecast demand growth that did not materialise; overbuild, Enron's collapse and merchant bankruptcies followed. This is the closest historical analogue to 2026 and the reason the 13% queue completion rate matters.
• "Nuclear renaissance" of 2007-2012. Dozens of announced US reactors; four started; two finished. Vogtle 3 and 4 came in years late and billions over budget. Every current nuclear forecast is made by an industry that has already produced one identical forecast that failed.
• Two decades of flat demand, 2005-2022. Efficiency, offshoring and LED lighting held US electricity demand roughly flat despite economic growth. PJM's 2021 ten-year forecast was 0.3% annual growth. An entire generation of planners, market rules and capacity constructs was built for a no-growth world, and that is why the system is responding badly to growth.
• The smart grid wave, ~2009-2014. Stimulus-funded AMI rollouts delivered metering but far less of the demand-side flexibility promised. A caution directly relevant to today's VPP and flexible-load claims: installing the hardware is not the same as dispatching the resource.
• Distributed generation and "utility death spiral", ~2013-2016. Rooftop solar was widely forecast to disintermediate utilities. It did not; net metering reform and the sheer scale of network costs prevented it. Utilities are structurally more central in 2026 than in 2013.
• The 2010s energy storage cost curve — a true positive. Not every forecast failed. Battery costs fell as predicted and US storage capacity is now compounding at roughly 70% a year. Storage is the sector's clearest case of a hype wave that delivered.
• Bitcoin mining load, 2018-2022. A large-load boom that arrived, interconnected, then partly departed on a price crash — leaving stranded interconnection in Texas and upstate New York. The nearest precedent for a data-centre demand reversal, and the one utilities cite privately.
• Hydrogen for power, 2020-2024. Hydrogen-ready turbines, hubs and enormous forecast volumes; very little power generation resulted. A recent, well-documented false positive in this exact sector, and the pattern — announcement volume far exceeding docketed projects — is the one to check SMRs against.
• Offshore wind, 2021-2026. Contracted, then repriced by inflation and rates, then cancelled in part, then subjected to federal stop-work orders in December 2025, then reinstated by courts across all five projects under construction (5.8 GW) in January-February 2026. A demonstration that in this sector, political risk is now as material as technical risk and moves faster.

The recurring pattern: this sector reliably over-forecasts demand inflections and new technology timelines, and reliably under-forecasts cost declines in modular manufactured technologies (solar, batteries). Apply that asymmetry to 2026: the demand forecast is more likely wrong than the storage cost curve.

8. What is changing now (as of 2026-09-15)

The macro frame. Per the shared macro brief, the FOMC held at 3.50-3.75% on a 9-3 vote in July 2026 with three dissents in favour of a hike, and named tariff pass-through, Middle East conflict disruption and "the surge in demand related to the AI buildout" as inflation drivers. It also flagged financial-stability risk from leveraged financing of infrastructure buildout. This sector is therefore in an unusual position: it is both a named cause of the inflation that is keeping capital expensive, and the most capital-hungry sector in the economy. Every thesis below must survive a higher discount rate. The September 16, 2026 FOMC decision is unresolved as of this research date.

Why the FOMC named energy — the specific mechanism. Two separate channels, and conflating them is the commonest error:

• Fuel channel (global). Brent swung from above $100/bbl at the start of Q2 2026 to $118 on 29 April and $72 on 26 June, with average daily swings of $4/bbl against $1/bbl a year earlier, before the 17 June US-Iran memorandum on resuming Strait of Hormuz shipping (EIA, 2026-07-15). This is what Ofgem passed through to British households: a 4% cap rise, gas bills up 8%.
• Capacity and network channel (US-specific). US fuel costs are falling — Henry Hub at $3.43/MMBtu on record production. The US electricity price increase is driven by capacity scarcity and network capex: PJM wholesale costs +46% to $56.7 billion in seven months, average price $116.53/MWh against $79.57/MWh. Residential prices are forecast at 18.2 c/kWh in 2026, +5%, with East Coast growth up to 7% annually — against PCE inflation of roughly 3.7-4.1%.

The four operative facts of the sector right now.

1. Forecasts moved up; deployment did not move with them. NERC's ten-year summer peak increase went to 224 GW, 69% above the prior year, with 13 of 23 assessment areas at elevated or high risk. Meanwhile PJM's capacity auction cleared at the cap and fell 6,831 MW short of its reliability requirement — the first RTO-wide shortfall in its history — procuring only 525 MW of new resources, of which 208 MW were uprates to existing plant.
2. The binding constraints are physical and industrial, not financial. Turbines committed to 2030; transformers at 128-144 weeks; interconnection at a five-year median. Capital is not scarce; delivery slots are.
3. Regulators are constructing an entirely new customer class in real time. FERC's June 2026 Section 206 orders to all six RTOs, its December 2025 co-location order to PJM, Texas SB 6 and the PUCT's 30-minute zero-compensation curtailment standard, and the Ohio 85%-minimum-take tariff. The rules being written in 2026 will determine returns for a decade.
4. The first honest down-revisions have arrived. PJM trimmed 2027-28; AEP Ohio halved its forecast; Exelon says 22% of its 65 GW pipeline will materialise; ERCOT suspended its Batch Zero process against 474 GW of requests, ~90% data centres, more than five times its historical peak, and said its own forecast is probably too high.

Testing the sector-split thesis: are energy and climate capital flows diverging?

Verdict: supported on drivers and policy, contradicted on deployment volume. The split is justified, but not for the reason usually given.

Evidence for divergence (strong).

• Interconnection queue composition inverted: gas requests +86% to 253 GW while solar −19%, storage −16%, wind −19% (LBNL, 2026-06-01). This is the single cleanest datapoint in the dossier.
• Federal clean-energy tax support is ending: OBBBA set a 4 July 2026 beginning-of-construction deadline, Treasury removed the 5% safe harbour, and FEOC rules from January 2026 disqualify Chinese equipment.
• Coal retirements slowed to the least in 15 years — 2.6 GW in 2025 against 13.7 GW in 2022 — with 4.8 GW postponed and 1.1 GW cancelled outright.
• The $1.4 trillion utility capex plan is justified by load growth and reliability, not decarbonisation; EEI's own framing is "electrification, industrialization, manufacturing onshoring, and data centers/AI".
• Buyers are paying premiums for gas-fired fuel cells explicitly on speed, with carbon not a stated criterion.

Evidence against divergence (also strong).

• EIA expects a record 86 GW of US capacity additions in 2026, led by 43.4 GW of utility-scale solar and 24.3 GW of storage. Storage reached ~52 GW by mid-2026 on 70% three-year average growth.
• Offshore wind's five projects under construction (5.8 GW) all won their court challenges and resumed.
• BNEF put 2025 global energy transition investment at a record $2.3 trillion, up 8%.
• China's installed solar capacity overtook coal in September 2026.
• The largest new clean-power deals of 2026 are demand-driven, not policy-driven: Google's 396 MW Fervo geothermal contract, Google's Duane Arnold nuclear PPA, CFS's fusion offtake into PJM.

Synthesis. The two sectors have not diverged in volume — clean generation is being built at record rates. They have diverged in causation, and that is the more important split for a trend platform. Clean energy in 2026 is being procured because it is cheap, fast to build, or firm and 24/7 — not because of climate policy, which has been actively withdrawn in the US. The selection criterion has changed from carbon to firmness, speed and interconnectability, and technologies that score well on those (geothermal, nuclear restarts, storage, gas) are winning regardless of which side of the old divide they sit on. The right way to read this sector is therefore not "fossil vs clean" but "dispatchable and deliverable vs neither" — and on that axis, energy and climate genuinely are different businesses with different buyers, different regulators and different risk. One caution: the record 2026 renewables build reflects a pipeline safe-harboured under the old regime, with a four-year placed-in-service limit. The divergence in volume is coming, around 2029-2030; it has not happened yet, and anyone declaring it already visible in installation data is misreading a lagging indicator.

9. The five lists

Five most important current trends

1. System-operator load forecasts re-rated by an order of magnitude (T-05-01)
2. Electricity prices outpacing inflation and entering the macro picture (T-05-02)
3. Gas turbine capacity sold out through 2030 (T-05-03)
4. Capacity markets clearing at administrative caps with physical shortfalls (T-05-04)
5. Interconnection queues rotating from renewables to gas (T-05-05)

Five fastest-growing signals

1. Co-location and behind-the-meter supply as the interconnection bypass (T-05-09)
2. Load flexibility traded for speed-to-power (T-05-10)
3. On-site fuel cells as bridge power (T-05-11)
4. Enhanced geothermal reaching project-finance bankability (T-05-12)
5. Data-centre load transients as a bulk-system stability problem (T-05-13)

Five trends most likely to affect businesses

1. Gas turbine capacity sold out through 2030 (T-05-03) — overlaps list 1
2. Grid equipment lead times as the hard physical constraint (T-05-06)
3. Large-load tariffs and the reallocation of interconnection cost risk (T-05-07)
4. Capacity markets clearing at caps (T-05-04) — overlaps list 1; a 10 MW industrial customer's PJM capacity charge rises ~12x by 2028
5. Federal clean-energy tax credit support withdrawing (T-05-16)

Five trends most likely to affect consumers

1. Electricity prices outpacing inflation (T-05-02) — overlaps list 1
2. Electricity affordability as a political constraint on load growth (T-05-14)
3. Capacity market costs flowing into bills from June 2028 (T-05-04) — overlaps lists 1 and 3
4. Coal retirement schedule stalling, with ratepayers funding units that barely ran (T-05-17)
5. Data-centre load transients as a stability risk (T-05-13) — consumers experience this as outage risk, not price

Acknowledged overlaps. T-05-04 appears on three lists and T-05-02 on two. This is not padding: capacity market design is simultaneously the most important current dynamic, the largest business cost shock, and the main consumer bill driver from 2028. Its recurrence is a finding.

10. Overhyped / overlooked / cooling / reversing

Most overhyped

SMRs as a near-term answer to data-centre power. The evidence that hype outruns substance is precise: zero commercial SMR megawatt-hours have been delivered anywhere in the US. The genuine 2026 milestones are TerraPower's first-ever NRC construction permit for a commercial non-LWR (March 2026, groundbreaking April) and Kairos's Hermes 2 groundbreaking (April 2026); TVA's BWRX-300 and Dow's Xe-100 applications are still under NRC review. DOE's own pilot programme targeted three designs reaching criticality by 4 July 2026 and had three of eleven projects with a Final Documented Safety Analysis as of May 2026. Meanwhile DOE's actual near-term nuclear plan — UPRISE, 2.5 GW by 2027 and 5 GW by 2029 — relies entirely on uprates, licence renewals and restarts, and the $80 billion federal partnership is built on AP1000 large reactors. Nuclear's share of US generation is flat at 18% in 2026 and 2027. The tell: capital is flowing (score 4) while revenue is zero (score 1). The 2023 NuScale/UAMPS cancellation at ~$89/MWh is the precedent nobody cites.

Federal Section 202(c) emergency orders as a reliability instrument. Presented as a necessary backstop; empirically and legally undercut in the same year. Four of eleven ordered units were not operating as of mid-2026. Net compliance costs reached $259 million for J.H. Campbell alone through 30 June 2026 after offsetting $239 million of MISO revenues. On 11 September 2026 the DC Circuit vacated that order, holding that "emergency" requires an immediate reliability crisis and that DOE's reading "invites frequent federal interventions that are unsupported by the statute". GridLab's Nikhil Kumar: "There is no emergency."

Honourable mention: the 474 GW and 2,061 GW headline numbers. Both are routinely cited as demand or pipeline. Both are option counts against a 13% historical conversion rate.

Most overlooked

Data-centre load transients as a bulk-system stability problem. NERC documented repeated unexpected load reductions of 1,000 MW or more across the Eastern and Texas interconnections in 2024-2025 and issued a Level 3 alert in May 2026; on 22 July 2026, 3.8 GW tripped offline in Northern Virginia, the largest such event in PJM history. This is physically the same failure mode ENTSO-E identified in the Iberian blackout — voltage and reactive power control gaps producing cascading disconnection. Attention has missed it because it is an engineering story with no equity ticker attached, and because the regulatory hook is incomplete: data-centre operators are not NERC-registered entities, so the alert reaches transmission planners, not the loads causing the problem.

Load flexibility as the cheapest available capacity. Duke's Nicholas Institute estimate of roughly 100 GW of headroom on the existing grid for loads willing to curtail a limited number of hours is larger than any supply-side option available this decade, and FERC has built it into RM26-4 (60-day studies for flexible loads) and the six show-cause orders. It is overlooked because it is a contracting problem rather than a construction problem and therefore has no vendor lobbying for it.

The transformer and GSU supply chain. 144-week GSU lead times sit inside every schedule in this sector and appear in almost no AI-infrastructure commentary. Grain-oriented electrical steel is the unfixed upstream bottleneck and attracts almost no new investment.

MISO as the control case. Capacity prices fell 42% in MISO in the same year PJM's hit the cap. Anyone extrapolating PJM to the US is wrong, and almost everyone does.

Cooling

Unbounded data-centre load forecasts. The indicator that turned: PJM cut summer 2027 by ~4 GW and 2028 by 4.4 GW (2.6%) after stricter vetting; AEP Ohio halved its forecast; Exelon disclosed 22% expected materialisation; ERCOT suspended Batch Zero on 7 August 2026. Note the honest caveat — PJM simultaneously raised its long-term growth rate from 3.1% to 3.6%. This is deflation at the margin, not a reversal.

Federal clean-energy tax credit support. Indicator: solar and wind interconnection requests both −19% in the 2026 queue data, even as installations hit records. The queue is the leading indicator.

Reversing

Coal retirements. Mechanism: federal emergency orders plus voluntary deferrals. 2025 coal retirements of 2.6 GW were the least since 2010 against 13.7 GW in 2022; 4.8 GW postponed, 1.1 GW cancelled. The reversal is now itself reversing after the DC Circuit ruling — which is the point: the retirement schedule has become a political variable and is no longer a reliable model input in either direction.

Trends that may reverse, and the mechanism.

• Data-centre load growth would reverse via an AI capex retrenchment. Mechanism: hyperscaler capex guidance cuts → queue withdrawals → capacity price collapse in PJM. Early indicator: withdrawal rates in the ERCOT large-load queue once Batch Zero resumes.
• Gas turbine scarcity would reverse via slot-reservation cancellations. Reservations are non-binding intent; the conversion rate is undisclosed and is the single most important unpublished number in the sector.
• Electricity price increases would reverse via political intervention — the UK VAT removal is a live template, and US state legislatures have shown they will act.
• The clean-energy policy withdrawal could itself reverse: a federal court has already struck down one IRS attempt to narrow credit eligibility, and a change of administration would reopen the question.

11. Risks and major uncertainties

Sector-specific risks.

1. Overbuild against a forecast that deflates. The 2000-2002 merchant gas bust is the template. Thirty-to-forty-year assets are being committed against a forecast revised 69% in one year and already trimmed in three jurisdictions.
2. Cost misallocation triggering political intervention. If data-centre costs land on residential bills in PJM states, legislatures will act, and they will act crudely.
3. A cascading outage caused by a load trip. 3.8 GW disconnecting in one event is already close to the largest contingency most planners model. The regulatory response to an actual blackout would be severe and fast.
4. Financing risk. The FOMC explicitly flagged leveraged financing of infrastructure buildout as a stability concern. Merchant projects underwritten at 2024 rates on 2026 capacity prices are exposed at both ends.
5. Equipment concentration. Three turbine makers and a handful of transformer suppliers is a single point of failure for national capacity expansion.
6. Regulatory whiplash. Stop-work orders issued and judicially reversed within six weeks; 202(c) orders issued and vacated within sixteen months; tax credits legislated and litigated. Political risk now moves faster than project risk.

Genuine unknowns — and the distinction matters.

Things we don't know but could find out: the conversion rate from turbine slot reservations to firm orders (GE Vernova has it and does not publish it); the true duplication rate in the ERCOT and PJM large-load queues (the operators are now measuring it); actual metered data-centre load as distinct from inferred commercial-sector growth (no operator publishes it, but it is measurable); the full net cost incidence of data centres on other ratepayers (no regulator has published one).

Things nobody can know: whether AI compute demand sustains for the fifteen-to-thirty-year life of the assets being built to serve it; whether SPARC achieves net energy gain in 2027; whether the Middle East conflict escalates or settles; the November 2026 and 2028 US election outcomes that determine tax, tariff and emergency-order policy; whether a 3.8 GW load trip eventually causes a cascading failure.

12. Scenarios to 2030

Base — "Constrained growth." Demand grows strongly but well below the forecast; roughly half of queued large load materialises. Turbines, transformers and interconnection meter the pace. Capacity prices stay high in PJM and moderate elsewhere. Retail prices continue rising 3-5% annually, staying above inflation. Nuclear adds 5 GW via restarts and uprates; SMRs remain non-commercial; geothermal reaches low single-digit GW. Falsifiable early indicator: PJM's January 2027 load forecast. A second consecutive near-term trim with a stable long-term rate confirms this branch.

Upside — "Delivery unlocked." Flexible-load interconnection works at scale: FERC's 60-day pathway is adopted, a meaningful share of the ~100 GW of headroom is contracted, and equipment expansions land on time in 2027-28. Load growth is served without proportional capacity build; prices stabilise. Indicator: the first RTO tariff with an operational flexible-load service class and at least 1 GW contracted under it by end-2027. Contracted megawatts, not announcements.

Downside — "Demand deflation." AI capex retrenches; queue withdrawals accelerate; PJM capacity prices fall below the cap; utilities are left with committed capex and minimum-take contracts against absent load; the 85% minimum-take tariffs become litigation. Indicator: two consecutive quarters of declining large-load queue volume in ERCOT after Batch Zero resumes, combined with a PJM auction clearing below the cap.

Disruption — "Firmness redefined." Enhanced geothermal and long-duration storage scale faster than expected and, together with contracted flexibility, displace the gas peaker as the marginal firm resource. Turbine backlogs convert poorly. Indicator: Fervo delivering Cape Station Phase 1 on schedule in early 2027 and a second developer reaching non-recourse project finance for EGS. One company is a project; two is a sector.

Regulatory — "The new customer class hardens." FERC finalises RM26-4 with a national large-load framework; states converge on minimum-take and curtailment terms; co-location is permitted but fully cost-allocated and curtailable. Siting becomes a compliance exercise. Indicator: a FERC final rule in RM26-4 before end-2027 that preempts or harmonises state large-load tariffs.

Failure — "Reliability event." A multi-gigawatt load trip coincides with a generation contingency and causes a cascading outage in PJM or ERCOT. Emergency federal and state intervention follows; data-centre interconnection is restricted by statute; the sector's political licence contracts sharply. Indicator: any load-loss event above 4 GW, or a NERC Level 3 alert escalating to a mandatory standard directed at large loads rather than transmission planners.

13. Data gaps and limitations

1. No metered data-centre-only load is published by any US system operator. Every attribution in this dossier — including NERC's, PJM's and EIA's — is inferred from commercial-sector totals and interconnection requests. This is the largest single gap in the sector and it undermines every "data centres caused X" claim, including ones we report.
2. Turbine slot-reservation conversion rates are undisclosed. GE Vernova's 116 GW is a company-defined non-GAAP pipeline metric with no published relationship to firm orders. The entire gas-scarcity thesis rests on a number whose meaning only the issuer knows.
3. True queue duplication rates are unmeasured. Exelon's 22% is one company's estimate; the 13% historical completion rate is a different measure over a different period. Nobody has published a like-for-like duplication analysis of the current large-load queues.
4. No regulator has published a full net cost incidence of data-centre load. Monitoring Analytics' 9% / $10.48-per-MWh figure is marginal wholesale cost in one RTO. The industry's counter-claim of net subsidy through contributed infrastructure is equally unaudited. This is the most consequential unresolved number in the sector and both sides are arguing from partial measures.
5. EIA STEO electricity detail pages did not resolve on retrieval. Residential price figures (18.2 c/kWh, +5% in 2026) come from Utility Dive reporting EIA's May 2026 STEO rather than from the EIA page directly, and are therefore one step removed from Tier A. The September 2026 STEO headline figures were retrieved directly.
6. Bloom Energy's 55-day Oracle deployment is single-source and company-originated, reported via investor-commentary media (Tier C). Its MW capacity is not disclosed. T-05-11 is marked single_source with evidence_quality capped at 2 accordingly.
7. Mitsubishi Power's 2026 backlog was not independently verified. Only GE Vernova's figures come from primary issuer disclosure; Siemens Energy and Mitsubishi are triangulated through trade press, one source of which (EUCI) is Tier C.
8. Total disclosed funding for TerraPower, X-energy, Kairos and Crusoe could not be verified from primary sources within budget. Those entity records carry "unverified"-equivalent language and medium confidence rather than a plausible figure.
9. Non-US coverage is thinner than it should be. China is covered via Global Energy Monitor and NEA-via-Global Times (state-affiliated); Europe via Ofgem and ENTSO-E. India, Japan, Korea and the Gulf are not covered at all, despite the Gulf being directly relevant to the oil price channel and India being a major share of global demand growth. No Chinese-language primary source was read directly.
10. The web search budget was exhausted at 200 calls before covering: on-site gas turbines at data centres (xAI Memphis and similar), 2026 electricity-affordability electoral politics, and independent SMR cost analyses. Those three areas are under-evidenced here.
11. Paywalled and licensed material. BloombergNEF's underlying deal-level data, IEA's detailed electricity statistics, and several trade datasets are licensed. Phase 2 needs a budget line for BNEF and IEA specifically.
12. Conflicting figures recorded rather than resolved: NERC vs Grid Strategies on reserve adequacy; PJM vs MISO on capacity price direction; ERCOT queue vs ERCOT forecast; LBNL queue decline vs EIA record installations; Sierra Club vs IEEFA on 202(c) costs (and those two are not independent — both oppose the orders). Each is in the relevant trend record's contradictions array.

14. Ranking scorecard

Total: 52 / 55. Recorded honestly: the two scores below 5 are the ones where this sector genuinely is not exceptional. Competitive turnover (criterion 1) is fast in policy and slow in assets — regulated monopolies do not change hands — and the paid-research opportunity (criterion 9) is real but structurally capped by the quality of free public data.

15. Sources

1. "Short-Term Energy Outlook, September 2026" — U.S. Energy Information Administration — https://www.eia.gov/outlooks/steo/report/ — 2026-09-09 — A
2. "Long-Term Reliability Assessment" — North American Electric Reliability Corporation — https://www.nerc.com/globalassets/our-work/assessments/nerc_ltra_2025.pdf — 2026-01-29 — A
3. "2026 PJM Load Forecast Report" — PJM Interconnection — https://www.pjm.com/-/media/DotCom/library/reports-notices/load-forecast/2026-load-report.pdf — 2026-01-14 — A
4. "PJM's Updated 20-Year Forecast Continues To See Significant Long-Term Load Growth" — PJM Inside Lines — https://insidelines.pjm.com/pjms-updated-20-year-forecast-continues-to-see-significant-long-term-load-growth/ — 2026-01-14 — A
5. "PJM Capacity Auction Procures 138,318 MW of Generation Resources" — PJM Inside Lines — https://insidelines.pjm.com/pjm-capacity-auction-procures-138318-mw-of-generation-resources-as-work-continues-to-address-growing-electricity-demand/ — 2026-07-14 — A
6. "ERCOT Releases Preliminary Long-Term Load Forecast for Years 2026-2032" — ERCOT — https://www.ercot.com/news/release/04152026-ercot-releases-preliminary — 2026-04-15 — A
7. "Queued Up: 2026 Edition — Characteristics of Power Plants Seeking Transmission Interconnection" — Lawrence Berkeley National Laboratory — https://emp.lbl.gov/queues — 2026-06-01 — A
8. "GE Vernova Releases Second Quarter 2026 Financial Results" — GE Vernova — https://www.gevernova.com/news/articles/ge-vernova-releases-second-quarter-2026-financial-results — 2026-07-22 — A
9. "Interconnection of Large Loads to the Interstate Transmission System, Docket No. RM26-4-000" — Federal Energy Regulatory Commission — https://www.ferc.gov/rm26-4 — 2026-07-22 — A
10. "FACT SHEET: FERC Directs Nation's Largest Grid Operator to Create New Rules to Embrace Innovation and Protect Consumers" — Federal Energy Regulatory Commission — https://www.ferc.gov/news-events/news/fact-sheet-ferc-directs-nations-largest-grid-operator-create-new-rules-embrace — 2025-12-18 — A
11. "One Year After Executive Orders, U.S. Nuclear Energy Renaissance Is in Full Swing" — U.S. Department of Energy, Office of Nuclear Energy — https://www.energy.gov/ne/articles/one-year-after-executive-orders-us-nuclear-energy-renaissance-full-swing — 2026-05-23 — A
12. "Energy price cap will rise by 4% from October 2026" — Ofgem — https://www.ofgem.gov.uk/press-release/energy-price-cap-will-rise-4-october-2026 — 2026-08-26 — A
13. "Petroleum markets responded to disruptions in the Middle East in the second quarter" — U.S. Energy Information Administration — https://www.eia.gov/todayinenergy/detail.php?id=67865 — 2026-07-15 — A
14. "United States on track for record natural gas production in 2026" — U.S. Energy Information Administration — https://www.eia.gov/todayinenergy/detail.php?id=67944 — 2026-08-12 — A
15. "Retirement delays of U.S. electric generating capacity may continue in 2026" — U.S. Energy Information Administration — https://www.eia.gov/todayinenergy/detail.php?id=67206 — 2026-02-23 — A
16. "U.S. coal-fired generating capacity retired in 2025 was the least in 15 years" — U.S. Energy Information Administration — https://www.eia.gov/todayinenergy/detail.php?id=67427 — 2026-04-13 — A
17. "Battery storage capacity averaged 70% growth over the last three years" — U.S. Energy Information Administration — https://www.eia.gov/todayinenergy/detail.php?id=67925 — 2026-08-07 — A
18. "Electricity 2026 — Executive summary" — International Energy Agency — https://www.iea.org/reports/electricity-2026/executive-summary — 2026-03-26 — A
19. "ENTSO-E Publishes Expert Panel Final Report on 28 April 2025 Blackout in Spain and Portugal" — ENTSO-E — https://www.entsoe.eu/news/2026/03/20/entso-e-publishes-expert-panel-final-report-on-28-april-2025-blackout-in-spain-and-portugal/ — 2026-03-20 — A
20. "Fervo Energy Secures $421 Million in Non-Recourse Project Financing for Cape Station" — Fervo Energy — https://fervoenergy.com/fervo-energy-secures-421-million-in-non-recourse-project-financing-for-cape-station/ — 2026-03-19 — A
21. "Data center load made up 9% of PJM wholesale costs so far in 2026: market monitor" — Utility Dive, reporting Monitoring Analytics — https://www.utilitydive.com/news/data-center-load-pjm-wholesale-market/828917/ — 2026-08-27 — B
22. "PJM capacity prices hit price cap, reserve shortfall grows" — Utility Dive — https://www.utilitydive.com/news/pjm-capacity-auction-price-cap-reserve-shortfall/825282/ — 2026-07-15 — B
23. "NERC forecasts peak demand to rise 24% on new data center loads" — Utility Dive — https://www.utilitydive.com/news/nerc-10-year-peak-demand-forecast-jumps-24-on-new-data-center-loads/810955/ — 2026-01-30 — B
24. "PJM trims near-term load forecast on stricter data center vetting, economic outlook" — Utility Dive — https://www.utilitydive.com/news/pjm-interconnection-load-forecast-data-centers/809717/ — 2026-01-15 — B
25. "Sudden data center load losses prompt NERC alert, recommendations" — Utility Dive — https://www.utilitydive.com/news/data-center-load-disruptions-nerc-alert-recommendations/818036/ — 2026-04-21 — B
26. "Court rejects DOE 'emergency' order delaying coal plant retirement as overstep" — Utility Dive — https://www.utilitydive.com/news/appeals-court-vacates-doe-emergency-order-michigan-power-plant/830189/ — 2026-09-11 — B
27. "DOE emergency orders are incurring additional costs. What are the benefits?" — Utility Dive — https://www.utilitydive.com/news/doe-emergency-power-plants-reliability-benefits-costs/822934/ — 2026-06-25 — B
28. "MISO capacity prices fall as new supply outpaces demand growth" — Utility Dive — https://www.utilitydive.com/news/miso-capacity-auction-nerc/818797/ — 2026-04-29 — B
29. "Commercial electricity use will likely surpass residential in 2027: EIA" — Utility Dive, reporting EIA STEO — https://www.utilitydive.com/news/data-centers-electricity-residential-rates-eia/820342/ — 2026-05-15 — B
30. "What data center developers need to know about FERC's large load directives" — Utility Dive — https://www.utilitydive.com/news/data-center-interconnection-ferc-large-load-show-cause/824501/ — 2026-06-19 — B
31. "Commonwealth Fusion Systems aims to complete demonstration reactor with $1B funding round" — Utility Dive — https://www.utilitydive.com/news/what-a-billion-dollar-funding-means-for-commonwealth-fusion-systems-and-its/828515/ — 2026-08-24 — B
32. "Transformers in 2026: Shortage, Scramble, or Self-Inflicted Crisis?" — POWER Magazine — https://www.powermag.com/transformers-in-2026-shortage-scramble-or-self-inflicted-crisis/ — 2026-01-02 — B
33. "Phantom Data Centers Didn't Break the Power Grid — They Proved it Was Already Broken" — POWER Magazine — https://www.powermag.com/phantom-data-centers-didnt-break-the-power-grid-they-proved-it-was-already-broken/ — 2026-05-15 — B
34. "Regulator Approves AEP Ohio's Landmark Data Center Tariff" — POWER Magazine — https://www.powermag.com/regulator-approves-aep-ohios-landmark-data-center-tariff/ — 2025-07-10 — B
35. "DOE Unveils Initiative to Add 5 GW of Nuclear Capacity Through Uprates and Restarts" — POWER Magazine — https://www.powermag.com/doe-unveils-initiative-to-add-5-gw-of-nuclear-capacity-through-uprates-and-restarts/ — 2026-03-12 — B
36. "Texas Pauses Data Center Interconnections Pending Statewide Audit" — Akin Gump — https://www.akingump.com/en/insights/alerts/texas-pauses-data-center-interconnections-pending-statewide-audit — 2026-08-10 — B
37. "PUCT affirms curtailment authority over co-located data centers in first net metering case under Senate Bill 6" — White & Case LLP — https://www.whitecase.com/insight-alert/puct-affirms-curtailment-authority-over-co-located-data-centers-first-net-metering — 2026-08-05 — B
38. "Review of NERC's 2026 Summer Reliability Assessment" — Grid Strategies LLC (Adria E. Brooks) — https://gridstrategiesllc.com/review-of-nercs-2026-sra/ — 2026-07-06 — B
39. "China wasted enough wind and solar to cover all new power demand in H1 2026" — Global Energy Monitor — https://globalenergymonitor.org/article/china-wasted-enough-wind-and-solar-cover-all-new-power-demand-h1-2026 — 2026-08-01 — B
40. "China's installed solar power generating capacity surpasses coal power for the 1st time: National Energy Administration" — Global Times, citing China NEA — https://www.globaltimes.cn/page/202609/1369526.shtml — 2026-09-01 — B
41. "EEI Data: Electric Companies to Invest $1.4T to Support Customers, Power Growth" — Edison Electric Institute / Electric Perspectives — https://www.electricperspectives.com/capital-expenditures-grid-investment/ — 2026-05-27 — B
42. "Solar and wind try to navigate Trump's obstacle course for tax credits" — Canary Media — https://www.canarymedia.com/articles/solar/solar-wind-trump-obstacles-tax-credits — 2026-06-15 — B
43. "BloombergNEF's New Energy Outlook 2026" — BloombergNEF — https://about.bnef.com/insights/clean-energy/bloombergnefs-new-energy-outlook-2026-transition-to-newer-technologies-expanded-electrification-to-strengthen-nations-energy-security/ — 2026-05-19 — B
44. "Enhanced geothermal notches another win as Google buys 400 MW from Fervo" — TechCrunch — https://techcrunch.com/2026/09/02/enhanced-geothermal-notches-another-win-as-google-buys-400-mw-from-fervo/ — 2026-09-02 — B
45. "US partners with Westinghouse, Cameco and Brookfield on $80B nuclear deployment" — Utility Dive — https://www.utilitydive.com/news/westinghouse-cameco-brookfield-nuclear/803999/ — 2025-10-28 — B
46. "US offshore wind stop-work orders reversed: Courts push back across all projects under construction" — Aegir Insights — https://www.aegirinsights.com/us-offshore-wind-stop-work-orders-reversed-courts-push-back-across-all-projects-under-construction — 2026-02-02 — B
47. "Existing US grid can handle 'significant' new flexible load: report" — Utility Dive, reporting Duke Nicholas Institute — https://www.utilitydive.com/news/us-grid-headroom-flexible-load-data-center-ai-ev-duke-report/739767/ — 2025-02-13 — B
48. "In 2026, virtual power plants must scale or risk being left behind" — Utility Dive — https://www.utilitydive.com/news/in-2026-virtual-power-plants-must-scale-or-risk-being-left-behind/810321/ — 2026-01-27 — B
49. "Solar and storage to lead 86 GW capacity surge in 2026" — pv magazine USA, reporting EIA — https://pv-magazine-usa.com/2026/04/28/solar-and-storage-to-lead-86-gw-capacity-surge-in-2026/ — 2026-04-28 — B
50. "Manufacturers say AEP Ohio still inflating data center demand after halving forecast" — Utility Dive — https://www.utilitydive.com/news/aep-ohio-data-center-load-tariff-oma-manufacturers/811583/ — 2026-02-10 — B
51. "GE Vernova's Gas Turbine Backlog Hits 116 GW as Power Orders More Than Double" — Turbomachinery Magazine — https://www.turbomachinerymag.com/view/ge-vernova-gas-turbine-backlog-hits-116-gw-as-power-orders-more-than-double — 2026-07-23 — B
52. "U.S. transformer market faces severe supply constraints as lead times extend to four years" — pv magazine USA — https://pv-magazine-usa.com/2026/05/11/u-s-transformer-market-faces-severe-supply-constraints-as-lead-times-extend-to-four-years/ — 2026-05-11 — B
53. "Bloom Energy Delivered Power to an Oracle Data Center in 55 Days" — The Motley Fool — https://www.fool.com/investing/2026/09/09/bloom-energy-delivered-power-to-an-oracle-data-cen/ — 2026-09-09 — C (single-source, company-originated claim; flagged in data gaps)
54. "Natural gas turbine manufacturers see order backlog grow as worldwide demand soars" — EUCI — https://www.euci.com/natural-gas-turbine-manufacturers-see-order-backlog-grow-as-worldwide-demand-soars/ — 2026-01 — C (undated at article level; used only for Siemens/Mitsubishi corroboration, never as sole support)
55. "U.S. electricity prices continue steady increase" — U.S. Energy Information Administration — https://www.eia.gov/todayinenergy/detail.php?id=65284 — 2025-05-14 — A (2025 publication; used for the pre-2026 price trend baseline only)

Macro context sources (from the shared brief, not re-verified here): FOMC minutes 2026-07-29 (https://www.federalreserve.gov/monetarypolicy/fomcminutes20260729.htm); Crunchbase and KPMG Venture Pulse H1 2026; Wilson Sonsini on the January 2026 semiconductor tariff and export-control changes.

More research in this sector · Explore trend records