Critical Materials Atlas
Supply structure · concentration over time

The critical-minerals concentration story runs two ways

The materials everyone worries about have not moved together. Over 1995–2024 the battery metal cobalt concentrated sharply and cleanly — its production concentration more than tripled while more countries came online. The materials now under Chinese export controls did the opposite, diversifying off near-monopoly highs — antimony (−0.34) and graphite (−0.19) on solid country coverage, rare earths (−0.18) on only five reporting countries in the baseline decade, and platinum-group metals easing too. And the “critical minerals are concentrating” figure usually cited from an average is essentially two materials, lithium and cobalt: remove them and it disappears. Built from production data alone — no trade, no consumption model.

0.14→0.45
Cobalt HHI (tripled) while its reporting producers rose 14→20 — the clean concentrator
−0.27
avg change of antimony and graphite — the two export-controlled materials with enough reporters to test; both diversified
→ −0.002
the cited critical mean (+0.034) once lithium & cobalt are removed — the average is two materials
−0.095
median change of the materials on the EU’s first (2011) critical list — frozen pre-window, they fell

There is a real result in this data, and it is not the one usually told. Line up every critical material by how its production concentration changed and they split into two camps pulling in opposite directions. That split — not a tidy class trend — is the finding, and it needs no comparison group to see: the materials are compared only to their own past. The single tidiest claim you can make from a group average, “critical minerals are concentrating,” turns out to rest almost entirely on two of them.

The whole distribution, one dot per material

Each material’s change in production HHI, 1995–2004 → 2015–2024, on one axis (0 = many equal producers, 1 = a single producer). This is the most honest object on the page: it shows the two outliers doing the work instead of hiding them inside an average. The vertical tick is each group’s median.

critical material comparable control (Au/Ag/potash/iron ore) other ordinary mineral (transport-sited)▏ = median (all 23 criticals; the 4 comparable controls)

The concentration average is two materials

The number that gets repeated is a group mean, and it says critical-material production grew more concentrated by +0.034. It is true, and it is misleading: two materials sit at the far edge of the distribution and carry the whole thing. Lithium’s HHI more than tripled (0.21 → 0.71) and cobalt’s tripled too (0.14 → 0.45). Remove those two and the average of the other 21 is essentially zero (−0.002).

Remove the two biggest movers — lithium and cobalt:
Critical materialsMean changeMedian changeShare that rose
All 23+0.034+0.04615 / 23
The other 21 (without Li & Co)−0.002+0.04413 / 21

The median does hold — most of the 23 rose a little — so a weak “typical material drifted up” claim survives for this set. But see the next section: it does not survive freezing the definition, so it is not a durable property of critical materials, it is a property of which materials the lists have come to include.

Freeze the definition, and the broad claim reverses

The lists were rewritten repeatedly between 2011 and 2023 — inside the change window — and materials were added partly because they had concentrated. Lithium, added to the EU list in 2020, is the clearest case. The clean test is to freeze the definition to the first EU critical-raw-materials list (2011), which sits at the end of the baseline decade: a material critical in 2011 cannot have been listed because of concentration seen after 2011. Seven of our materials were on that list. Run the same measure on just them:

The same change, on the pre-window (EU 2011) definition:
Critical materialsMedian changeShare that rose
Full set (23, today’s lists)+0.04615 / 23
Frozen to EU 2011 (7) antimony, cobalt, fluorspar, graphite, PGMs, rare earths, tungsten−0.0953 / 7

The materials critical before the window mostly diversified — antimony, graphite and rare earths came off near-monopoly highs, and PGMs eased from a lower base. So “critical minerals are concentrating” as a class statement is substantially an artifact of list expansion. What is left once selection is controlled for is not a class trend at all — it is the divergence: one material on that 2011 list, cobalt, concentrated hard, and it is the cleanest signal in the dataset.

The clean case, and the caveated one

Cobalt is the finding you can lean on. It was defined critical in 2011, before any of this; its production HHI tripled (0.14 → 0.45); and its reporting producers rose from 14 to 20 — so the concentration is real, not an artifact of fewer countries reporting. The mechanism is no mystery either: the baseline decade coincides with the collapse of Congo’s state producer during the 1996–2003 wars, and the rise since is Congo’s return and expansion to roughly 70% of world mine supply. Naming the cause does not soften the measurement — it explains it. Lithium is the one to be careful with. Its HHI rose more still (0.21 → 0.71), but it was listed only in 2020 (so selection bites), and its reporting producers fell from 10 to 7 — the one big mover where thinner reporting inflates the index. It belongs in the picture, labelled, not leaned on.

Every critical material, ranked production HHI change, 1995–2004 → 2015–2024

grew more concentrated diversified† thin coverage (<8 producers)

Does a control group help? A little, and it shows its own limit

The divergence above needs no control — each material is compared to its own past. But it is fair to ask whether ordinary commodities behaved differently. Most don’t make a clean comparison: bulky, low-value minerals (gypsum, salt, aggregates, talc) deconcentrate because they are produced next to where they are used — transport economics, not a counterfactual. The genuinely comparable non-critical commodities are the globally-traded, geology-sited ones. We first drew that set as three — gold, silver, potash — and an adversarial pass added a fourth we had missed: iron ore. Four is a check, not a control group, so each is shown by name:

And they split. Iron ore concentrated (+0.08 on broad coverage — the Australia–Brazil consolidation of the seaborne market); silver barely moved; gold and potash spread out. So the comparable non-criticals show no class trend either — which is this page’s point. “Critical vs ordinary” is not the axis that explains concentration; the mechanism inside each market is.

The coverage question, stated correctly. Concentration indices can rise simply because fewer countries report. Reporter counts actually rose in most of these series — but a group average of reporter counts does not clear a nonlinear, per-material index, so it is checked per material. The one place it bites hardest is lithium (10 → 7), which is why lithium is shown as an individual dot and not leaned on. Applying that test systematically — a sum over fewer than eight reporters cannot be a world census however plausible the number looks — two of the materials on this page fail it: lithium and rare earths (five reporters in the baseline decade). Both are shown; neither carries an argument on its own. Cobalt — the case the page rests on — goes the clean way: 14 → 20 producers and a tripling HHI.
What this rests on, and what it doesn’t. Concentration is measured on mine/primary production tonnages from the BGS World Mineral Statistics panel — one consistent source across every material, computed only for years with at least five reporting producers. Like every BGS figure on this site, these are shares of the countries that report — which is why reporter counts are checked per material, thin series carry a †, and the one material whose count fell while its index rose (lithium) is shown, not leaned on. It uses no trade data, no apparent-consumption estimate, and no model, so it is independent of everything still open in the atlas’s consumption layer. It measures where material is produced, not refined or consumed — a material can have diversified mines and a concentrated midstream — graphite and rare earths are the live cases: their mining diversified in this data while their processing stayed concentrated in China (that is the chokepoint question, and a different measurement). The level of concentration (criticals sit higher than ordinary minerals) is partly definitional and is deliberately not the claim here; the change is.