build_energy_footprint.py → energy_footprint.json. Sources per row (IMA/Pidgeon LCA, Fraunhofer ISE, IAI, ICSG, worldsteel, IFA, USGS).Bar = total primary MWh/t · number in bar = t CO₂/t · elec = electricity MWh/t · right = stage-matched top share
Three mechanisms, not one
1 — Power-sited (aluminium, polysilicon, silicon metal). These run on electricity: their feedstock (alumina, quartz) is cheap and shippable, so the price of power dominates and the furnace goes where electricity is cheapest and most reliable. This is the clean “energy chokepoint wearing a metal’s name” case — an ore embargo can’t touch it, a power-price shock can, and diversifying means building at cheap reliable power elsewhere, not opening a mine.
2 — Fuel/reductant-sited (magnesium, ferrochrome, steel, ammonia). Here the energy is mostly not electricity. Magnesium’s Pidgeon retorts are coal-fired batch units (~1–3 MWh/t of electricity; the ~78 MWh/t is primary energy including embodied ferrosilicon) sited on dolomite, ferrosilicon and cheap coal — not on a cheap grid. Steel and ammonia site on coking coal and natural gas as reductant and feedstock. Cheap energy still concentrates them, but as fuel plus a co-located reductant or ore, so the fix is different: it’s about coal, gas and material logistics, not just electricity price.
3 — The control: copper smelting. Low energy (~2.5 MWh/t) — and yet not diffuse: China holds ~40–45% of refined copper. A low-energy stage that concentrated anyway, on concentrate logistics, sulphuric-acid handling and industrial policy. It is the proof that energy intensity is a driver of the map but not the whole of it.
The through-line survives, sharpened: decarbonising and diversifying are often the same project for the power-sited stages — the only way to move polysilicon or aluminium off China’s coal grid and cut ~16–36 t CO2/t is to rebuild them at cheap clean power. For the fuel-sited stages the lever is coal/gas and reductant supply; for copper it is smelter policy. The mechanism map →