Chart 1 from the The peak lithium experiment: supply, reserves and price to 2045 analysisMOLUSCCAOpen Analysis
Forecasting experiment · Lithium · Real USGS and IEA data

The "peak lithium"
experiment

Lithium is the metal the energy transition runs on. Its price rose 13x and then fell 85% in four years. We took the real production, reserve and adoption data and asked two questions: is the world actually running out, and what happens to the price over the next twenty years.

The short answer: there is no peak lithium. Reserves nearly tripled while production grew tenfold, and on our own demand forecast the world uses only about 30% of today's known reserves by 2046. The scarcity is not geological. It is a timing problem, and it shows up as violent boom and bust cycles rather than exhaustion.
The data

What actually happened

Four years of the most dramatic price action in any major commodity.

2010 price
$5.2k
per tonne carbonate
2022 peak
$68.1k
annual average, 13x rise
2025 low
$10.5k
an 85% crash
Production
10.3x
28kt to 290kt since 2010
Chart 2 from the The peak lithium experiment: supply, reserves and price to 2045 analysis
Battery-grade lithium carbonate, annual average, USGS and Benchmark Mineral Intelligence assessments. The 2022 spike and the collapse that followed are the defining features of this market.
Question one

Is there actually a peak?

The phrase "peak lithium" assumes a fixed stock being drawn down. The data says otherwise, and this is the clearest finding in the whole study.

Chart 3 from the The peak lithium experiment: supply, reserves and price to 2045 analysis
Reserves against annual production. Between 2010 and 2025, production grew 10.3 times while reserves grew from 13 to 37 million tonnes. We mined far more and ended up with far more.
YearReservesProductionYears of supply
201013 Mt28,100 t463 yrs
201514 Mt36,000 t389 yrs
202021 Mt82,500 t255 yrs
202226 Mt146,000 t178 yrs
202430 Mt221,000 t136 yrs
202537 Mt290,000 t128 yrs
Why reserves grow while you mine them

Reserves are not a measure of what exists. They measure what is economically extractable at current prices with current technology. When the price rose, deposits that were previously uneconomic became reserves overnight, and exploration budgets expanded. High prices create reserves. That is why the years-of-supply figure falls slowly even as production explodes.

Our own forecast reinforces this. Cumulative demand from 2026 to 2046 comes to about 11.0 million tonnes of lithium content, which is 30% of today's reserves and roughly 10% of known resources. Geology is not the binding constraint this century.

Question two

Where the demand comes from

Lithium demand is mostly a bet on electric vehicle adoption. So the forecast is only as good as the adoption curve underneath it.

Chart 4 from the The peak lithium experiment: supply, reserves and price to 2045 analysis
EV share of new car sales with a logistic curve fitted to real IEA data. R squared of 0.959, against 0.918 for an exponential. The fitted inflection point is 2025, meaning adoption is passing its steepest phase now.
A callback worth making

In an earlier experiment we tested an S-curve on exchange rates and it failed badly, because a currency has no ceiling to saturate against and the fitting routine invented an absurd one. Here the same curve is the correct choice, because EV adoption genuinely saturates: it cannot exceed 100% of new car sales.

Same mathematics, opposite verdict. That is the entire lesson about model choice. You are not picking the most sophisticated tool, you are matching the shape of the method to the shape of the problem.

Where this fit is weak, stated plainly

The fitted ceiling sits at the lower edge of the range we allowed, which is a warning sign. You cannot reliably estimate a saturation ceiling from data that has not saturated yet. Seven observations, all from the early and middle part of the curve, cannot pin down where it flattens. We carry that uncertainty through the simulation rather than pretending it away.

The honesty test

Could any model have seen the spike coming?

Before forecasting forward, we tested whether a reasonable model trained on the pre-boom era could have predicted what followed. We fitted a trend on 2010 to 2020 and let it predict blind.

Chart 5 from the The peak lithium experiment: supply, reserves and price to 2045 analysis
Predicted against actual. The model missed the 2022 spike by 79% and then missed the 2025 crash by the same margin in the opposite direction. MAPE of 44%.
YearPredictedActualError
2021$13,332$17,000-22%
2022$14,534$68,100-79%
2023$15,844$32,000-50%
2024$17,272$13,500+28%
2025$18,829$10,542+79%
2026$20,527$19,000+8%
What this tells you about the forecast below

A smooth trend fitted to a decade of calm was worse than useless across a boom and bust. It was confidently wrong in both directions. Any twenty-year lithium forecast, including ours, should be read with that failure in mind. We are not publishing a number we expect to be right. We are publishing a range and the mechanism that generates it.

The model

Why we did not use a price model

Fitting a curve to price history assumes the past pattern continues. For a commodity that is the wrong shape entirely, so we modelled the physical market instead.

demand = EV sales x lithium intensity + grid storage + industrial supply(t) = supply(t-1) x growth( price(t-5) / marginal cost ) price = marginal cost x (demand / supply) ^ 2.4

The critical term is the five-year lag. A new mine takes four to seven years from investment decision to production. So when prices spike, capital floods in and the supply arrives half a decade later, by which time demand has often moved on. That single delay is what generates boom and bust cycles in every mined commodity, and it is why lithium behaves nothing like a currency or an equity.

Chart 6 from the The peak lithium experiment: supply, reserves and price to 2045 analysis
Median demand against median supply capacity. The supply response lags demand growth, and the gap is where price pressure comes from.
The forecast

Twenty years, twenty thousand simulations

Chart 7 from the The peak lithium experiment: supply, reserves and price to 2045 analysis
Historical price and the simulated forward distribution. The band is wide on purpose. Note how the median sits near the marginal cost of production for years before demand growth reasserts itself.
YearMedian price80% bandDemandSupply
2030$6,162$5,500 to $12,433404 kt472 kt
2035$6,269$5,500 to $12,873462 kt540 kt
2040$8,807$5,500 to $22,248576 kt590 kt
2046$16,640$5,500 to $53,092883 kt716 kt
P(another crash below $8k)
98%
at some point before 2046
P(another spike above $40k)
21%
at some point before 2046
Expected boom-bust cycles
1.0
per simulated path
Reserves used by 2046
30%
of 2025 known reserves
The finding that matters

A 98% chance of another crash and a 21% chance of another spike is not a contradiction. It is the same mechanism seen twice. The lagged supply response guarantees that the market overshoots in both directions. The risk in lithium is not that it runs out. It is that anyone who invests at the top of a cycle loses their capital before the mine is finished.

This is precisely the kind of risk that is hard to insure and hard to finance, because the exposure is to timing rather than to any single event.

Try it yourself

Move the assumptions

The two numbers that decide everything are how high EV adoption climbs and how quickly supply responds. Drag them and watch the balance change.

Demand vs supply in 2040
balanced
demand 576 kt against supply 580 kt

Simplified version of the model above, running live. Deficit means prices rise, surplus means they fall.

Scenarios

Five futures

Scenario2030 median2040 medianP(spike)
Base case$6,106$8,73321%
Fast adoption (K=0.95, r+50%)$21,534$5,50054%
Slow adoption (K=0.60)$8,018$7,95219%
Sodium-ion breakthrough (intensity -8%/yr)$5,500$10,48626%
High-cost supply (cost $14k)$13,913$15,15845%
Reading the table

Faster adoption does not simply mean higher prices. In the fast-adoption case the 2030 price is far higher, but by 2040 it is lower than the base case, because the early price signal pulls forward an enormous supply response that then overshoots. High-cost supply is the scenario that keeps prices structurally elevated, since it raises the floor rather than the peak.

Conclusion

The question behind the question

"When does lithium run out" is the wrong question, and asking it leads to the wrong decisions.

The world is not short of lithium. It is short of lithium at the right time and the right price. Reserves grew nearly threefold while production grew tenfold, and our own twenty-year demand path consumes less than a third of what is already known to be economically extractable.

The real risk is cyclical, not geological. A five-year lag between investment and production means the market cannot help but overshoot. Our simulations put a 98% probability on another crash below $8,000 and a 21% probability on another spike above $40,000, with about one full cycle per path.

And every long-run lithium forecast is secretly a forecast about two things: where EV adoption saturates and how fast capital responds to a price signal. Neither is knowable to within the precision that most published forecasts imply.

Theme