The mass balance, and every constant it rests on
World bauxite production (World Mining Data 2024) × the gallium content of bauxite (~50 ppm, literature range 30–80) gives the gallium embedded in mined ore. Times the share of bauxite processed to alumina (~85%) gives the gallium passing through alumina refineries — the only place it is economically recoverable. Against that we set actual primary gallium production (WMD ~987 t; other estimates 300–760 t, because China, ~99% of supply, does not report). The gap is what’s thrown away.
Bias bounds (shown, not buried): the ppm and production figures are the uncertain terms; every step below carries its low–high bracket. Gallium production statistics are genuinely contested — that uncertainty is part of the finding, not smoothed over. Trade caveat: gallium exports under HS 811292, shared with germanium (identical unit values) — 811292 is a customs catch-all also nominally covering hafnium/indium/niobium/rhenium/vanadium — so “gallium trade” is a bundle — see below. Inputs: production.json + reconciled trade → gallium.json.
The mass balance — where the gallium goes
From the gallium embedded in world bauxite down to what’s actually recovered. Bars to scale; the shaded bracket is the low–high bound at each step.
Why price can’t fix it
The gallium exists — tens of thousands of tonnes of it flow through alumina refineries every year. What’s missing is recovery capacity: the extraction step is built at only a handful of plants, and adding it means retrofitting a refinery, a multi-year capital project. So when gallium’s price quadrupled after China’s 2023 export controls, world supply could not follow — there was no idle capacity to switch on, and the raw material’s economics are set by aluminium, not gallium. This is the difference the atlas keeps drawing between a metal you can mine more of and one you can only recover more of, eventually. Gallium is the cleanest proof: a supply chain that discards most of the gallium in its feedstock by design, capped upstream by a host it cannot command.
Does published work agree? Yes — and this is a check, not a citation for its own sake. Peer-reviewed gallium material-flow studies put Bayer-liquor losses at roughly 95–96% and find that ~90% of primary gallium comes from that one Bayer route. This atlas reaches ~94% discarded from bauxite tonnage and recovery bounds and lands in the same place. Stated honestly, this is a consistency check, not an independent measurement: our calculation uses the same load-bearing constants the material-flow studies use (gallium ppm in bauxite, Bayer-liquor recovery), so the agreement rules out an arithmetic or bookkeeping error on our side — it does not independently confirm the shared assumption. The atlas draws the same line elsewhere for USGS/WMD (“independent compilations, not independent measurements”); it applies here too. The one constant everything rests on (ppm) is shown with its range. One caveat worth stating plainly: gallium can be recovered at much higher rates from new manufacturing scrap (offcuts from making GaAs wafers — recovery near 27%). That looks like an escape valve, but it isn’t a second mine: new scrap is a fixed fraction of gallium already produced and sold, so it scales with demand, not against a shortage. It cannot open a new primary fountain — the recycling story lives on its own page, not here.
And the trade data lies about it
Who imports gallium (revealed demand, 2024). But note the coding problem this metal exposes:
| Importer | share of gallium imports |
|---|
Why one metal, done deep, matters
A scorecard across 32 materials shows breadth; a single chain followed to the bottom shows the mechanism is real. Gallium, traced physically, turns every abstract axis of the hardest-cases scorecard into a number you can check: can’t scale becomes “94% discarded, capped by refinery retrofits”; concentrated becomes “99% from one country that doesn’t report”; thin market becomes “under 1,000 tonnes a year”; trade illusion becomes “bundled with germanium under one catch-all HS code.” The atlas’s thesis is not a correlation across a spreadsheet — here it is one metal’s physical reality, with the error bars drawn in.