Grid transformers rose with all electrical equipment, and their steel moved to China

1. What we set out to test, and why this is a note

Before the pandemic a large power transformer could be ordered with a lead time of under a year; by 2024 "36-month lead times" were "commonly quoted", with a maximum of 60 months^[1]. Average lead times for large power transformers almost doubled from 2021–22 to 2023–24, and power-transformer prices rose by around 75% in real terms since 2019, according to the International Energy Agency's industry survey^[5]. The US producer price index for power and distribution transformers averaged 195 in 2019, 313 in 2022 and 343 in 2024^[7].

Those are national figures and surveys. We asked what world trade shows, and whether the rise in its value came through prices or through volumes. In demand-and-supply terms, prices and quantities moving together indicate demand shifting along a supply curve^[11]; we hoped to measure how steep that curve was for traded grid equipment. The design was filed before any estimate, after two independent language models reviewed a draft as referees and a research agent assembled the literature under a rule that every source be opened before use.

Two things turned the test into a description. The filed pre-trend rule failed: transformer unit values were drifting relative to the controls long before 2021, so the before-and-after comparisons cannot be read as effects. And the referees' objection that the pattern may be a boom in electrical equipment generally turned out to be right: against motors, pumps and compressors, the transformer gap is no longer clearly different from zero. That second point is itself the first finding: the rise belongs to electrical equipment broadly. What follows reports what the trade record shows, labels what could not be separated, and says what would settle it.

2. Background

A utility's lead times for large transformers went from 12–18 months to 18–36 months^[3]. With materials in hand, manufacturing takes 12 to 16 weeks, but lead times for the inputs lengthened too^[2]. Manufacturers reported record order backlogs in 2024^[5].

Grain-oriented electrical steel (GOES) and the copper conductor "each account for roughly 25 percent of final LPT production costs"^[1], and labour on average 36 percent of manufacturing cost^[1]. GOES prices rose 70% in 2022 compared with 2020^[4] and "doubled between 2021 and mid-2023"^[5]. Russia accounted for almost 10% of global GOES production capacity in 2020 and its exports were sanctioned^[4]. The United States has one GOES producer, which can meet 12–20% of domestic demand^[1].

Against a pure equipment story: US large-transformer plants raised capacity utilisation from 40% to 78% between 2011 and 2023, as reported by a national laboratory citing the US International Trade Commission^[6]; and "the primary cause of delays in transmission projects remains permitting, particularly in advanced economies"^[5].

3. Data and method

Coverage. The world record is CEPII BACI, which ends in 2024: its newest release (V202601) has no 2025, and the atlas watches for the next one. Section 4.8 carries the world picture one year further with importers' own monthly reports to UN Comtrade, descriptively. The EU record (Eurostat Comext, section 4.6) and the US record (US Census Bureau, section 4.7) are monthly and run to July 2026. Sections 4.1 to 4.5 are the world record and the filed test; the two later sections extend it with the same design.

Trade flows are CEPII BACI^[8] in the HS 2002 nomenclature for every year, so no code changes under the series. A flow is one exporter–importer pair in one six-digit line; a flow-year enters at USD 100,000 or more with a positive tonnage, and unit values more than ten times above or below the line's median that year are dropped. The transformer lines are liquid-dielectric transformers 8504.21 (up to 650 kVA), 8504.22 (over 650 up to 10,000 kVA) and 8504.23 (over 10,000 kVA). The filed comparison goods are lifts, cranes, crushers, concrete mixers and electric welding machines. Welding machines are themselves electrical equipment, so a construction-only version without them is reported beside the filed one. Electric motors (8501.52/53), pumps (8413.70) and compressors (8414.80) were filed as a check on shared electrification demand.

Each estimate compares a flow with itself over time (flow and year effects), which removes some of the product-mix change that makes aggregate unit values unreliable^[9][10] but not mix within a flow. The gaps below are averages over 2023–24 against 2012–2020. Because the shock is to product markets, inference is a wild bootstrap clustered by customs line; with 9 lines, a p-value near 0.01 is suggestive, not conclusive, and the smallest 2023–24 gap the main comparison could reliably detect is 0.22 log points.

4. What the trade record shows

4.1 Transformers rose: first in unit value, late in tonnes

Source: CEPII BACI, release V202601, HS 2002. Computed values: out/buildout_study.json.

Relative to the comparison machinery, transformer unit values fell from +0.15 in 2012 to zero in 2019, were already +0.14 in 2022, and reached +0.24 in 2023 and +0.26 in 2024 (interval +0.04 to +0.48). Relative tonnes rose from -0.10 in 2017 to +0.01 in 2018, stayed roughly level through 2023 (+0.02), and reached +0.17 in 2024 (+0.01 to +0.33). The relative rise appears in unit values first and in tonnes only in the last year.

4.2 Not only transformers

Source: CEPII BACI, release V202601, HS 2002. Computed values: out/buildout_study.json, buildout-study/exploratory_electrical_boom.json.

exploratory rows were not in the filing; they were run after review to answer the referees' question directly and are labelled wherever they appear.

Against construction machinery alone, without the welding machines, the transformer gaps are +22% in unit value and +15% in tonnes, close to the filed comparison. Motors, pumps and compressors also rose relative to the filed comparison machinery: +9% in unit value (p = 0.011) and +8% in tonnes (p = 0.004). Transformers rose about twice as much as those goods taken together. Measured directly against them, the transformer gap is +10% in unit value (p = 0.055) and +9% in tonnes (p = 0.144), with intervals that include zero; that comparison could reliably detect only a gap of about 0.14 log points, larger than the estimate. The honest reading is that goods sharing the demand of electrification gained on other heavy machinery after 2020, transformers more than the others, and that nine customs lines cannot establish how much of the transformer excess is specific to transformers. The three transformer size classes point the same way (+18%, +18% and +23%) but none is distinguishable from zero on its own (p = 0.277, 0.263 and 0.333).

4.3 Steel and copper: unsettled

Source: CEPII BACI, release V202601, HS 2002 · World Bank commodity prices (Pink Sheet), copper. Computed values: out/buildout_study.json.

The filed adjustment nets a quarter of the real change in the GOES unit value and a quarter of the real change in the copper price from transformers only. The referees showed why that is fragile: the comparison machinery also contains copper and ordinary steel, and because both input prices peaked in 2021–22, netting them from one side alone pushes the earlier gap down (-7%) and so widens the rise from 2021–22 to 2023–24 almost mechanically. Netting only electrical steel, which most of the comparison goods do not use, leaves -2% in 2021–22 and +15% in 2023–24. Neither version accounts for labour, about 36% of manufacturing cost for large units, or for the long lag between order and delivery, which means 2023–24 shipments may have been priced on earlier inputs. How much of the transformer rise is materials is not settled by these data.

4.4 World exports of electrical steel concentrated

Source: CEPII BACI, release V202601, HS 2002 · Eurostat Comext, monthly bulk files (CN8) · US Census Bureau, international trade API, imports by HS10. Computed values: out/buildout_study.json, out/buildout_eu.json, out/buildout_us.json. World exports are available for three years only; 2026 is January to July (open points).

Source: CEPII BACI, release V202601, HS 2002. Computed values: out/buildout_study.json.

This is the clearest fact in the trade record, and it does not depend on unit values or on any comparison group; the shares are shares of export value. Between 2019 and 2024 China's share of world GOES exports rose from 12% to 32%, Russia's fell from 14% to 6% (its GOES exports have been under sanctions^[4]), and Japan's stayed near a quarter (23% and 25%). The number of exporters with more than 1% of the market fell from 14 to 11 and the three largest went from 48% to 67%. The price of GOES in trade, by contrast, cannot be read: in a placebo period, 2015–16, its unit value moved by +0.22 (p = 0.024) relative to other alloy steel, so its post-2021 movements are within its normal swings.

4.5 Where the transformers went

Source: CEPII BACI, release V202601, HS 2002. Computed values: out/buildout_study.json.

The intervals for single importing regions are wide; the US tonnage estimate in particular runs from +5% to +485%. They are shown as heterogeneity, not as findings. One exporter result is also worth recording: relative to the comparison machinery, transformer tonnes from China were -19% in 2021–22 (p = 0.038) and +21% in 2023–24 (p = 0.231).

4.6 The EU record to July 2026

The world data stop in 2024. The EU's customs records run monthly to July 2026, at the eight-digit level, and record the number of transformers as well as their weight. A second filing, made before those records were downloaded, applied the same method to the EU's trade with countries outside the EU, in both directions, with the United Kingdom left out of every year so that "outside the EU" means the same partners throughout: 175 months, 86,342 flow-years. Gaps in the table are relative to 2012–2020; 2026 is January to July.

Source: Eurostat Comext, monthly bulk files (CN8). Computed values: out/buildout_eu.json.

Source: Eurostat Comext, monthly bulk files (CN8). Computed values: out/buildout_eu.json.

Against motors, pumps and compressors: still not distinguishable from electrical goods generally. The transformer gap in value per tonne was positive in every year from 2023 — relative to 2019: 2023 +17% (p 0.095), 2024 +16% (p 0.075), 2025 +21% (p 0.072), 2026 +15% (p 0.135) — but no period's interval excludes zero, and the smallest gap the comparison could reliably detect is 0.24–0.29 log points, more than it estimates. The pre-trend rule passes for value per tonne here, unlike in the world data, though with seven lines passing it is weak evidence. Two filed sensitivities do exclude zero — the same comparison at eight digits (4 transformer and 31 comparison lines) in 2023–24 and 2025, and trade between EU members in 2023–24 and 2025 (its 2026 value rests on July records that were incomplete when downloaded) — but the filing made the six-digit comparison the headline, and it stays the headline.

No sign that transformers got heavier. Because the EU counts transformers as well as weighing them, the records can test whether the rise per tonne came from heavier units. It did not: within the same trade flows, transformers' weight per unit was -2% in 2025 and +6% in 2026 relative to 2019, and within 10% of 2019 in every year since (point values from an exploratory check, without intervals; unit counts are usable for 89%–94% of transformer flow-years). A first reading of these records said the opposite; it came from comparing transformers with motors, pumps and compressors, which became lighter per unit (-22% in 2025), and it is withdrawn. Other kinds of mix, such as higher efficiency at a similar weight, are not ruled out.

Source: Eurostat Comext, monthly bulk files (CN8). Computed values: out/buildout_eu.json.

In the value of the EU's imports of grain-oriented electrical steel from outside the EU, China went from 9% in 2019 to 57% in 2025, and Russia from 19% to none; China was second in 2021 and 2023, third in 2022, and first from 2024. The three largest partners' share — whoever they were that year; in 2019 they were Japan, Russia and the United States — rose from 71% to 94%. This is the EU's view of its own imports by value, not world exports and not EU consumption.

The finished transformers moved the same way. The note's US section reports where US transformer imports come from; the same question of the EU record, added here descriptively, gives a sharper answer. EU imports of transformers from outside the EU grew from €334m in 2019 to €2.1bn in 2025, and China's share of them went from 4% to 46%, passing Türkiye (36% in 2025) to become the largest supplier:

Source: Eurostat Comext, monthly bulk files (CN8) · US Census Bureau, international trade API, imports by HS10. Computed values: out/buildout_origins.json.

Source: Eurostat Comext, monthly bulk files (CN8) · US Census Bureau, international trade API, imports by HS10. Computed values: out/buildout_origins.json.

Month by month, the twelve months to 202607 brought €2.2bn of transformers into the EU from outside it, against €0.3bn in the twelve months to December 2019; the same measure for US imports runs from $1.5bn to $7.7bn in the twelve months to 202607. Both records are in current prices, and neither says what was installed.

4.6b Imports are not supply: what the EU actually uses

Every share above is a share of imports, and the EU makes much of both goods itself. A fifth filing, made before these production figures were read, put the obvious question to Eurostat's production statistics: how much of what the EU uses comes from China? Apparent supply is EU sold production plus imports from outside the EU minus exports to outside it, and the answer changes the size of the finding without changing its direction.

Source: Eurostat Prodcom, sold production (DS-059358) · Eurostat Comext, monthly bulk files (CN8). Computed values: out/buildout_supply.json.

In 2024, China supplied 34% of the EU's transformer imports but 9.1% of its supply, because EU makers sold €6.2bn of transformers that year; for electrical steel the two numbers are 52% and 24%. The filing set the rule in advance: where the supply share is less than half the import share, the import share is not read as dependence, and the supply share is the number to use. That is the case for both.

What the supply share does show is a steep climb from a very low base: China went from 0.5% of EU transformer supply in 2019 to 9.1% in 2024, and from 3.3% to 24% of electrical-steel supply. Production is valued at the factory gate and imports include freight, so the two sides of the sum sit on different price bases; sold production also excludes what a maker uses itself, and Eurostat has published no 2025 production yet, which is why these tables stop a year before the trade tables.

4.6c The rest of the grid: five more components

Transformers and their steel are two parts of a grid. A separate study, filed before its data were read and reviewed like this note, applied the same measure to seven components. It aimed at lines that are high-voltage or grid-scale by their own description, so that phone chargers and household fuses are not counted; two of the seven fit that only loosely and are starred: inverters, where the line is a size class rather than a use, and electricity meters, which include household smart meters.

Source: Eurostat Prodcom, sold production (DS-059358) · Eurostat Comext, monthly bulk files (CN8). Computed values: out/eu_grid_supply.json.

On this measure, for six of the seven the import share overstates dependence, by a factor of two to seven. The filing's rule: where China's share of supply is under half its share of imports, the import share is not read as dependence; from 0.5 to 0.8 it partly overstates it; above 0.8 it is a fair guide. Supply shares are banded as low (under 10%), material (10–25%), high (25–50%) and critical (over 50%). Grain-oriented electrical steel, at a ratio of 0.47, is the borderline case. EU makers account for most of the apparent supply of transformers, high-voltage cable, switchgear and switchboards; China's share of their supply is 3% to 9%. Electrical steel and meters (24%) sit at about a quarter.

Large inverters are the exception, and they are not shown to be grid equipment. China supplied 82% of EU imports and 73% of EU supply in 2024 (ratio 0.89: the import share is a fair guide), but the customs line holds every inverter above 7.5 kVA: commercial and utility solar, battery storage and industrial drives as well as grid converters. From 2026 the EU splits inverters by function rather than size. In January–July 2026, at any size, 64% of the inverters Europe bought from China were solar (51% of those it bought from all origins), and China supplied 97% of EU solar-inverter imports and 57% of the rest; the split does not separate sizes, so it does not say how much of the large-inverter line is solar. The supply share is also unsteady: China's share of EU inverter imports is 67%–88% in every year from 2019, but its share of supply swings from 20% to 73%, because EU production enters as round figures (€2.0bn, €3.0bn, €6.0bn) that look like Eurostat estimates, and because exports are large beside it (90% of production in 2024), which leaves supply as a small remainder. Read the inverter row as a dependence in a broad class of large inverters, of imprecise size, that is not shown to be a grid one.

These are value shares on the filed apparent-supply measure: imports are gross, so some of what arrives from China leaves the EU again, and factory-gate production sits beside imports valued with freight. The production and customs lines are matched on their descriptions, most loosely for inverters. Components are not summed: transformers contain the steel. The first run of that study missed a customs code that changed in 2023 and printed zero inverter trade for 2019–2022; a fact-check caught it, and the corrected series is the one shown (its deviation 1).

4.7 US imports to July 2026

A third filing, made before the US data were pulled, applied the method to US imports from the Census Bureau, January 2012 to July 2026, at the ten-digit level: 4,332 origin–line flow-years. The comparison goods are recorded only as counts of units, so the US test is value per unit. A count weighs a small distribution transformer and a large power transformer alike, so value per unit is mostly a measure of what is shipped, not of price; the pre-trend rule fails for every US comparison; and the smallest gap the headline comparison could reliably detect is 0.31 to 0.96 log points.

Source: US Census Bureau, international trade API, imports by HS10. Computed values: out/buildout_us.json.

No headline interval excludes zero, but with this little power that is not evidence of absence: a gap the size of the one estimated in the EU would go undetected. The ten-digit version of the same comparison does exclude zero in 2025 (+49%, p = 0.010), on a comparison that also fails the pre-trend rule. The US evidence on transformers is unstable rather than null. Within flows, transformers' own value per unit rose mainly in 2025 (+23% relative to 2019) and the comparison goods' mainly in 2024 (+32%); the two did not move in the same years. The filed weight test could not be run, because the Census records the kilograms of transformer imports only from 2026.

Source: US Census Bureau, international trade API, imports by HS10. Computed values: out/buildout_us.json.

Almost none of the grain-oriented electrical steel the United States imports as steel comes from China — at most 1.2% of the value in any year since 2021 — and most of it comes from Japan and South Korea. But direct imports are a small part of what the US uses: about 25 thousand tonnes in 2020, against US consumption the IEA puts at 0.15 million tonnes that year^[4], of which the one US producer meets 12–20%^[1]. The arithmetic leaves most of US use arriving inside imported cores and transformers. Those imports grew from $1.5bn in 2019 to $7.5bn in 2025:

Source: US Census Bureau, international trade API, imports by HS10. Computed values: out/buildout_us.json.

Source: US Census Bureau, international trade API, imports by HS10. Computed values: out/buildout_us.json.

The steel inside them is not observed. So the concentration of world GOES exports in China does not appear in the origins of direct US GOES imports, but these data cannot say whether it reaches the US inside transformers made in Mexico, South Korea or elsewhere. Nor do they say why China's direct share is so small; tariffs, trade remedies and supply relationships are all candidates, and none is tested.

4.8 World trade in 2025, from importers' own monthly reports

BACI ends in 2024, but countries file monthly reports to UN Comtrade, and those run into 2026. A fourth filing, made before this pull, asked two descriptive questions of them. The panel is the 83 importers that filed every month of 2024 and of 2025, which held 86% of 2024 world imports of the two electrical-steel lines and 83% of the transformer lines. China, India and Taiwan do not file monthly and are therefore absent as buyers; as suppliers they are counted, because the panel's members report where their goods came from. Importers value imports including freight, BACI does not, so levels are not compared with the rest of this note — only each series with itself.

The concentration continued into 2025. China supplied 36% of the panel's imports of grain-oriented electrical steel in 2024 and 39% in 2025; the three largest suppliers went from 70% to 72% (Japan 26%, Russia 5.0% in 2025). On the thinner panel that filed every month of both half-years (52 importers, 57% of 2024 world imports), China's share was 29% in January–June 2025 and 28% in January–June 2026, with Japan first. That is recorded, not read as a turn: the half-year panel is half the size of the annual one and its months are first releases.

Transformers rose faster than both comparison groups in 2025, on the same panel:

Source: UN Comtrade, monthly imports as reported by importers. Computed values: out/buildout_comtrade.json.

This is one year against one year: no flow-level design, no interval, no pre-period. It is consistent with the gap in section 4.1 continuing into 2025, and it cannot establish that. Data to 2026-06.

5. What failed, and what the checks say

• Unit values are not prices. The annual change in the unit value of US transformer imports correlates with the change in the BLS producer price index at only 0.43 (source: BLS WPU117409). Unit values carry product mix; the direction of that bias is not known, so the unit-value gaps are neither upper nor lower bounds on price change.
• Copper wire (7408.11, wire over 6 mm across, largely rod rather than transformer winding wire) showed no gap relative to copper cathode after 2022; it says little about transformers either way.
• Where the gap sits. Split at the median unit value, the 2023–24 transformer gap is +14% in the lower half (p = 0.039) and +1% in the upper half (p = 0.815): it sits in the lower-value flows.
• Other checks filed in advance — a placebo period for transformers, alternative unit-value bands, leaving out each line, and splits by exporter — are in the results file. All of them use the filed comparison machinery; none compares transformers with other electrical goods, so none bears on the transformer-specific question.

6. What this cannot say

• Nothing about AI or data centres specifically: every buyer of electrical equipment is in these flows.
• Nothing about domestic production: a country that builds its own transformers never appears in trade.
• Not whether supply was constrained, or by what — factories, electrical steel, labour or permitting.
• Not causal.
• Nothing about chip-making equipment, which was dropped from the design because its value per tonne measures which machines were shipped rather than their price.

7. What would settle it

The EU records in section 4.6 show that counting units, not only weighing them, changes the reading; the same records for the United States and for large importers elsewhere would extend it. Beyond that: producer price series by rating from more than one country; and order and capacity data from manufacturers, which trade data cannot see. The concentration of electrical-steel exports, by contrast, is already visible and needs no further identification: it is a fact about who exported the core material of transformers in 2024, though not about who could produce it.

Review record

Design: reviewed before filing by two independent language models as journal referees, which changed the question, the controls, the inference and the claims. Literature: assembled by a research agent under a rule that every source be opened and quoted. Result: an earlier version was reviewed by the same two referees and by a separate fact-checking agent, which traced every number to the results file and every quotation to the verified literature. Their reports turned the paper into this note: the event-study intervals were recomputed with the headline procedure, the electrical-steel figures were widened to both customs lines, a material-netting variant was added, and the comparison with other electrical goods was moved into the main text. That version was checked again by all three, and the EU extension in section 4.6 was reviewed by all three before this version; their review withdrew its first reading that transformers had become heavier. Eleven dated deviations are logged in the filing. The EU extension was filed separately, before its data were downloaded, and records eight further dated entries; the US extension was filed before its data were pulled and records two, including a weight test that turned out not to be runnable. The seven-component supply study in section 4.6c was filed on its own, reviewed by the same two referees and a fact-check before its result was written, and records three dated deviations.

References

1. U.S. Department of Energy (2024). Large Power Transformer Resilience: Report to Congress. July 2024. www.energy.gov
2. U.S. Department of Energy (2022). Electric Grid Supply Chain Review: Large Power Transformers and High Voltage Direct Current Systems. Supply Chain Deep Dive Assessment. February 2022. www.energy.gov
3. U.S. Government Accountability Office (2023). Electricity Grid: DOE Could Better Support Industry Efforts to Ensure Adequate Transformer Reserves. GAO-23-106180. www.gao.gov
4. International Energy Agency (2023). Electricity Grids and Secure Energy Transitions. Paris, October 2023. www.iea.org
5. International Energy Agency (2025). Building the Future Transmission Grid: Strategies to Navigate Supply Chain Challenges. Paris, February 2025. www.iea.org
6. Ramasamy, V., Cooperman, A., Jayswal, R. and Seward, M. (2026). Large Power Transformer Supply Chain Gap Analysis and Domestic Content Strategies for Hydropower Rehabilitation: Supplemental Report. NLR/TP-5700-96742, National Laboratory of the Rockies. docs.nlr.gov
7. U.S. Bureau of Labor Statistics. PPI by Commodity: Machinery and Equipment: Power and Distribution Transformers, Except Parts (WPU117409), annual average of monthly values. fred.stlouisfed.org
8. Gaulier, G. and Zignago, S. (2010). "BACI: International Trade Database at the Product-Level. The 1994-2007 Version." CEPII Working Paper 2010-23. www.cepii.fr
9. Schott, P. K. (2004). "Across-Product Versus Within-Product Specialization in International Trade." Quarterly Journal of Economics 119(2): 647-678. doi.org
10. Silver, M. (2007). "Do Unit Value Export, Import, and Terms of Trade Indices Represent or Misrepresent Price Indices?" IMF Working Paper 07/121. doi.org
11. Shapiro, A. H. (2022). "Decomposing Supply and Demand Driven Inflation." Federal Reserve Bank of San Francisco Working Paper 2022-18. doi.org

Data: CEPII BACI (Etalab Open Licence 2.0), World Bank Pink Sheet and CPI (CC BY 4.0), BLS (public domain). Built by build_grid_trade_paper.py from out/buildout_study.json. Cite as: Critical Materials Atlas (2026). Zenodo. https://doi.org/10.5281/zenodo.21948855