ai-waw.com Search

Forecasts · Copper · AIWAW-2026-R02

Copper semi-finished products output in the EU 2026–2031 – summary

Output of copper and copper-alloy semis in the EU-27 (IWCC classification, excluding wire rod) stagnates: the median of the forecast – an equal-weight ensemble of five models (the hybrid demand, trade and plant model and four statistical models) – is 3,946 kt in 2031 (2025: 3,886 kt; +0.3% a year, 80% interval −2.7% to +3.2%). The hybrid model sets the split by country and product, the intervals and the scenarios. Copper semis – 2,670 kt, alloy semis – 1,248 kt; wire rod, the intermediate for wire counted separately – 2,342 kt (2025: 2,282 kt). The 2026 nowcast (STS index NACE 24.44, January–August 2026) – 3,908 kt. 2031 scenarios (medians): accelerated grid build-out 4,054 kt; extra-EU import surge 3,934 kt; closures under high energy costs 3,820 kt; construction recession 3,737 kt – differences from the base of −209 to +108 kt, several times smaller than the base 80% interval (3,293–4,706 kt). The forecast does not constitute investment advice.

approved version 2.0 Base year 2025 (latest annual data) · nowcast 2026 · forecast 2027–2031 ai-waw.com Method

At a glance

  • EU-27: copper and copper-alloy semis output (excluding wire rod) – compound annual growth 2025–2031 0.3 %/yr p10–p90: −2.7–3.2
  • EU-27: copper wire rod output – compound annual growth 2025–2031 0.4 %/yr p10–p90: −2.0–3.0
  • EU-27: copper and copper-alloy semis output (excluding wire rod) in 2026 (nowcast) 3,907.5 kt product p10–p90: 3,659.1–4,169
  • EU-27: copper and copper-alloy semis output (excluding wire rod) in 2031 3,946.4 kt product p10–p90: 3,293–4,706.5
  • EU-27: copper wire rod output in 2031 2,341.5 kt product p10–p90: 2,019.3–2,717.8
  • EU-27: copper and copper-alloy semis output (excluding wire rod) – Cu content in 2031 3,512.7 kt Cu p10–p90: 2,938.8–4,176.1

kt = thousand tonnes; kt Cu – copper content, kt product – product mass (including alloys).

Accuracy: In the backtest (4 origins 2015–2021, horizon 1–5 years) the model ensemble is not more accurate than the random walk: mean MAPE 8.9% vs 8.5% (v1.3 ensemble on the same data – 9.0%). The hybrid model alone has a MAPE of 9.1%, and 8.3% with sector indices known in advance: most of the error comes from annual use and trade shocks, not from the business-cycle forecast. The forecast is therefore mainly about the level, structure and range of uncertainty and the effects of scenarios, rather than an edge over simple extrapolation. Details →

Forecast limitations (13)
  • The 2027–2031 EU-27 level is set by an equal-weight ensemble of five models; the hybrid model (weight 1/5) gives the structure, intervals and scenarios. Plant events (projects, closures) act mainly on the split between countries.
  • Hybrid model factors explain about 37% of the variance of the 2031 semis total (wire rod – about 85%); the rest is annual product shocks, country shares, the model ensemble and interval calibration – scenarios move the median less than the interval width.
  • The demand elasticity to sectors is identified mainly by 2020–2021 (β = 0.9; without those years 0.3); β and η uncertainty comes from a block bootstrap of years and is carried into the draws, but the grid and recession scenario results depend on β (variant in the note).
  • The copper end-use structure (IWCC End Use Summary) is global and is applied to the EU unchanged; the allocation of end uses to 4 sectors and the sector × product matrix are assumptions (class C) with sampled uncertainty. The construction share in tubes and brass in the EU is probably higher than in the global structure.
  • Capacity is documented for only part of the plants; the rest – a lower bound from 2021–2025 output and an assumed utilisation. The 'about 270 kt' figure for the Orsk rod mill is an output level in the company report, not capacity. Wire drawn at cable makers has no independent capacity measure.
  • Including electrodeposited foil (CN 7410) in the IWCC product 'Copper Plate, Sheet & Strip (Inc. Foil)' is an assumption (class C) – IWCC lists heading 7410 only in a general code list; foil is a separate sub-form in the plant module, and its share of the product is taken from PRODCOM 2025 (constant apart from projects). Without EU foil projects, total semis in 2031 would be lower by about 26 kt (about 0.65%).
  • Project probabilities and delays are assumptions by status; many projects in company documents do not state a capacity increase and have no quantitative term (e.g. the new Contirod and Upcast II lines at the KGHM plant in Orsk).
  • Trade: the CN 7407–7411 → IWCC product key is explicit, and the IWCC–Comext reconciliation residual is modelled statistically; exports to the USA after the 50% tariff are a permanent level reduction from 2026 data, without diversion of other exporters' supplies to the EU market.
  • The copper balance is a check (output not adjusted): the R01 distribution is drawn independently of R02, and the ceilings on cathode imports and direct-melt scrap are assumptions. The p10 metal headroom is negative from 2030 and the p5 headroom from 2028 (upper output tail with low refined copper output).
  • The 'other EU countries and unallocated output' item is about 47% of 2025 semis: mainly wire drawn at cable makers, wire rod in Belgium, France and Greece and Spanish wire rod drawn into wire within the same company (La Farga, Cunext – outside PRODCOM sold output), countries without data.
  • GDP and investment forecast uncertainty is taken from historical volatility (class C), as IMF and EC forecast error statistics are not in the forecast data; the recession scenario parameter comes from 2008–2010 (construction output F).
  • The backtest covers the simplified hybrid model (without capacity, events and the sampled matrix); a backtest of the full hybrid and the sensitivity of conformal multipliers to the origin set – in version 2.1. No archived vintages of IWCC, Comext and IMF/EC forecasts. The intervals of total semis in 2027–2028 and of alloy tubes and alloy wire are wider than the calibration target (up to 1.9 times), because the common group and total components exceed the target of products with a small conformal multiplier.
  • The copper price, conversion margins, aluminium substitution (cables, windings, heat exchangers), copper intensity of electric vehicles and data-centre demand are not separate model variables (they act only through historical trends).

The forecast is analytical and informational; it does not constitute investment advice or a financial recommendation. Values are uncertain as described by the intervals and in the method section. Terms of use →

The summary is open to text and data mining, including model training, with attribution; the full forecast data remain reserved.

Scenarios

  • Base (Monte Carlo) – Distribution of Monte Carlo draws (hybrid model, 2026 nowcast, model ensemble, calibration of the final distribution after capacity constraints). Named scenarios are shifts of the same draws under changed parameters – their intervals show uncertainty within the scenario. (2031: p50 3,946.4 kt product, p10–p90 3,293–4,706.5)
  • Accelerated grid build-out – Energy and grid demand grows over the whole horizon at the top-tercile rate of 2011–2025 (Eurostat production indicators C271, C2732, F42) – a historical parameter. The direction is consistent with the EU Action Plan for Grids COM(2023) 757 (EUR 584 billion of grid investment this decade); the plan's scale was not converted into semis demand. (2031: p50 4,054.2 kt product, p10–p90 3,380.2–4,816.2)
  • Extra-EU import surge – New capacity just outside the EU: the Hailiang plant in Tangier (tubes, rods, foil; the company document points to the European market) supplies part of its output to the EU, and the new CRW Metal rod mill in Türkiye (per the company – for the Turkish market) displaces part of EU wire rod exports to Morocco, Tunisia and Serbia. The share χ is an upper bound – an assumption. (2031: p50 3,934.2 kt product, p10–p90 3,285–4,693.2)
  • Closures under high energy costs – Stress test: in countries with an industrial electricity price above the 2025 EU-27 average (Eurostat nrg_pc_205), 10% of form output closes from 2027 and 15% in total from 2029, weighted by the form's energy intensity; part of the output is taken over by other EU plants, the rest by imports. (2031: p50 3,819.6 kt product, p10–p90 3,181.7–4,545.1)
  • Construction recession – In 2027–2029 construction output repeats the worst three-year period of 2001–2025 (Eurostat, construction F: 2008–2010), with transmission to industry (fittings, heat pumps, air conditioning – C28); from 2030 – the base path. (2031: p50 3,737.2 kt product, p10–p90 3,104–4,487.9)

Method

The 2027–2031 EU-27 level is set by an equal-weight ensemble of five models; the hybrid model (scheme: demand → trade ∥ plants → output) gives the structure, intervals and scenarios. (1) Semis demand in four sectors – construction, energy and grids, industry, transport: IWCC end uses × Eurostat production indicators, forecast from GDP and investment (IMF, EC) with uncertainty, sector × product matrix, common elasticity from a block bootstrap of years. (2a) Trade: imports as use penetration, exports from partner GDP weighted by the Comext structure and with the US export level after the tariff, CN 7407–7411 → IWCC product key reconciled with history. (2b) Plants: capacity from documents or a lower bound from output, foil separately, ICSG, projects and closures with probability and delay. (3) Output = demand − imports + exports, country split with capacity dynamics, projection onto capacity and company lower bounds, copper balance check against R01, cathodes and scrap; 10,000 draws, 2026 nowcast (STS NACE 24.44), conformal calibration of the final distribution (c 0.53–1.99); four named scenarios on the same draws.

AI note

Claude agents (Anthropic) were used to retrieve and read documents (IWCC sheets, Eurostat databases, company reports, registers), to build the plant register, to verify company figures in primary documents, to check data consistency and to draft code. All forecast numbers are computed by deterministic Python code (fixed random seeds) – the language model does not generate values. Confidential-data fills and assumptions are classed C and explicit in data files. Independent reviews (data and sources, mathematics, industry, reproducibility, text – A1–A5) were carried out by separate Claude agent instances; the result is approved in the portal panel.

How to cite

ai-waw.com, Copper semi-finished products output in the EU 2026–2031, AIWAW-2026-R02 v2.0, 2026-10-17 – https://www.ai-waw.com/en/forecasts/eu-copper-semis-production-2026-2031

Full forecast →