Nuclear power has violently returned to favour, driven by decarbonisation and the electricity appetite of AI data centres. We took the real resource, production and requirement data and asked the only two questions that matter: is there enough uranium, and can it be mined in time.
This is the chart that matters, and it is the one most coverage of the uranium story leaves out.
| Year | Mine production (ktU) | Requirements (ktU) | Gap | Coverage |
|---|---|---|---|---|
| 2010 | 54.0 | 63.0 | +9.0 | 86% |
| 2012 | 58.4 | 61.6 | +3.2 | 95% |
| 2014 | 56.2 | 56.0 | -0.2 | 100% |
| 2016 | 62.1 | 62.8 | +0.7 | 99% |
| 2018 | 53.5 | 59.2 | +5.7 | 90% |
| 2020 | 47.7 | 62.1 | +14.4 | 77% |
| 2021 | 48.3 | 62.5 | +14.2 | 77% |
| 2022 | 49.4 | 63.5 | +14.1 | 78% |
| 2023 | 54.8 | 65.1 | +10.3 | 84% |
| 2024 | 60.2 | 65.7 | +5.4 | 92% |
| 2025 | 63.0 | 68.9 | +5.9 | 91% |
When we ran the same exercise on lithium, reserves had nearly tripled while production grew tenfold. There was no shortage, only a timing problem, and the market was in surplus.
Uranium is the mirror image. Resources are ample, but the market has run a primary production deficit for over a decade, plugged by drawing down stockpiles that were built in a different era. Secondary supply is the only reason the lights stayed on, and it is finite by definition. Two commodities, both essential to the energy transition, with completely opposite supply structures. The lesson is that "critical mineral" is not a risk category. Each one has to be modelled on its own mechanism.
The Red Book puts identified recoverable resources at 7.9 million tonnes of uranium below USD 130 per kgU. Whether that is a lot depends entirely on a number nobody knows.
Reactor requirements grew at 0.68% a year between 2010 and 2025. The World Nuclear Association's reference scenario for 2040 implies 5.32% a year from here. That is roughly an eightfold acceleration in the growth rate, and it has not happened yet.
Every bullish uranium forecast, including the optimistic half of ours, is really a bet that this acceleration arrives. Be suspicious of anyone who states it as fact rather than as the assumption doing all the work.
Before forecasting forward, we checked whether a reasonable trend model fitted to the pre-boom era could have anticipated the renaissance.
| Year | Predicted | Actual | Error |
|---|---|---|---|
| 2021 | $32 | $33 | -2% |
| 2022 | $32 | $41 | -22% |
| 2023 | $32 | $49 | -36% |
| 2024 | $31 | $70 | -55% |
| 2025 | $31 | $59 | -47% |
| 2026 | $31 | $78 | -61% |
A model calibrated on a decade of depressed prices was confidently, consistently wrong once policy and sentiment turned. It could not see a regime change, because regime changes do not appear in the history you fit to. Read our numbers knowing exactly what happened to the last person who extrapolated this market.
Fitting a curve to price history assumes the past pattern continues. For a commodity with a decade-long development cycle that is the wrong shape entirely, so we modelled the physical market.
The critical term is the ten-year lag. A uranium mine takes roughly a decade from discovery to production, and longer where permitting is contested. So the supply response to today's price does not arrive until the late 2030s. That single delay is what turns a demand surge into a sustained price event rather than a brief spike.
Secondary supply is modelled as a declining contribution, because stockpiles, weapons downblending and enrichment tails are drawdowns, not renewable sources. The tighter the market gets, the less of a cushion remains.
| Year | Median price | 80% band | Demand | Gap | P(deficit) |
|---|---|---|---|---|---|
| 2030 | $89 | $60 to $132 | 81 kt | +4.6 kt | 72% |
| 2035 | $123 | $71 to $214 | 96 kt | +19.7 kt | 88% |
| 2040 | $160 | $85 to $260 | 113 kt | +35.2 kt | 95% |
| 2045 | $138 | $82 to $236 | 134 kt | +33.5 kt | 95% |
| 2050 | $83 | $48 to $142 | 158 kt | +2.2 kt | 53% |
The asymmetry is the finding. A 90% chance of exceeding $150 against a 10% chance of falling below $40 is not a forecast of a particular price. It is a statement that the downside is largely closed off, because a market already producing less than it consumes cannot easily flood itself.
Note also that the median falls back after 2040. That is not the deficit resolving painlessly. It is a decade of high prices finally dragging enough supply into existence, which is exactly how mining cycles end.
Reactor demand growth and mine development lead time. Everything else is detail.
Simplified version of the model above, running live. Deficit means the market must draw down stockpiles or bid the price up.
| Scenario | 2035 median | 2050 median | 2040 gap | P(above $150) |
|---|---|---|---|---|
| Base case | $124 | $82 | +36 kt | 90% |
| AI datacentre boom (demand 7.5%/yr) | $254 | $106 | +115 kt | 100% |
| Slow build (demand 2.5%/yr) | $105 | $83 | +24 kt | 89% |
| Faster mine response (6yr lead) | $123 | $75 | +21 kt | 82% |
| Secondary supply collapses (-15%/yr) | $139 | $64 | +39 kt | 91% |
| Higher incentive price ($110/lb) | $260 | $95 | +58 kt | 100% |
Only one lever meaningfully closes the gap, and it is not price. Faster mine development, cutting the lead time from ten years to six, does more to soften the 2040 shortfall than any demand assumption. This is a permitting and capital-allocation problem wearing the costume of a geology problem.
Note the AI data centre scenario. Demand at 7.5% a year produces a severe 2035 price but a lower 2050 price than you might expect, because an extreme early signal pulls forward an enormous supply response. Shocks in mined commodities tend to overcorrect.
Three things we are confident about, and one we are not.
The resource is not the constraint. Identified recoverable resources of 7.9 Mt cover our entire modelled demand to 2050 roughly three times over. Cumulative demand from 2027 to 2050 comes to 2.79 Mt, around 35% of what is already identified.
The deficit is real and it is structural. Ten of the last eleven years produced less than the reactor fleet consumed. That gap was filled by stockpiles built for a world with more weapons and fewer reactors. It is a one-time inheritance being spent.
The lag is the whole story. A decade between investment and production means today's price cannot fix this decade's problem. It can only fix the 2040s.
What we are not confident about: the demand path. The entire bullish case rests on reactor growth accelerating roughly eightfold from its fifteen-year trend. It may well happen, and AI power demand is a genuine new driver. But it has not happened yet, and a forecast that treats it as settled is selling you a conclusion rather than an analysis.