The magnets behind the models
The silicon story follows a pattern: the raw material is not the problem. The problem is what has to be done to it before it is useful. And the capabilities required to do that are concentrated in a small number of countries through decades of industrial investment.
The same pattern applies — even more sharply — to rare earth elements. Rare earths get a lot of press as a geopolitical pressure. But the real danger is not where the ore deposits are. The situation, again, is more interesting and more complicated than the headline version suggests.
The rare earth story, as most people have heard it, goes like this: a handful of exotic elements exist in only a few places on Earth, China controls most of those places, and therefore China has a stranglehold on the materials needed to build AI.
That version is partly right, mostly misleading, and wrong about the most important part.
Rare earths are not rare #
Rare earth elements — a group of 17 metals including neodymium, praseodymium, dysprosium and terbium — are relatively abundant in the Earth's crust. They are not rare in the way platinum or iridium is rare. They are "rare" in the sense that useful concentrations in economically mineable deposits are less common than iron ore or aluminium, but they exist in meaningful quantities on every populated continent.
Significant deposits are found in China, Australia, the United States, Brazil, India, Vietnam, Canada, Russia, Greenland, Tanzania and South Africa. There is no scarcity problem here.
The dangerous concentration is the industrial system that converts the ground into usable components.
Separation. Refining. Metal and alloy production. Permanent-magnet manufacturing. These steps are where China's dominance is nearly complete, and where it is genuinely hard to replace.
The full chain #
The rare-earth supply chain is long, and each step is harder than the last:
Rare-earth-bearing ore
↓
Mining and crushing
↓
Physical concentration
↓
Mixed rare-earth chemical product
↓
Separation into individual elements
↓
Rare-earth oxides
↓
Metals and specialised alloys
↓
NdFeB permanent magnets
↓
Motors, pumps, generators, precision actuators
↓
Cooling fans, liquid pumps, power infrastructure
↓
AI data centre
The separation step is where the genuine difficulty begins. Rare-earth elements have almost identical chemical properties. Pulling them apart requires long sequences of solvent-extraction stages, careful chemical handling, and management of waste that may contain low levels of thorium or uranium. The process is not something you stand up quickly in a new country.
China's position at each stage:
Mining share: high
Separation share: extremely high
Magnet manufacturing share: near-monopoly
This matters because China's real advantage is beyond deposits. It is:
- decades of accumulated separation expertise
- large-scale chemical processing infrastructure
- trained engineers and established process knowledge
- low-cost industrial clusters with downstream customers nearby
- the ability to process feedstock imported from other countries
China does not need to control every ore deposit to control the processing chain. Ore from Australia, Myanmar and other countries flows into Chinese processing facilities. The chokepoint is the refinery and the magnet factory.
The four rare earths #
Not all seventeen rare earths are equally relevant to data centres and AI systems.
Neodymium (Nd) and praseodymium (Pr) are the principal magnetic ingredients in neodymium-iron-boron (NdFeB) permanent magnets. These are the strongest permanent magnets commercially available. NdFeB technology underpins the motors and generators in most modern industrial and consumer equipment.
Dysprosium (Dy) and terbium (Tb) are added in smaller amounts to help NdFeB magnets retain their strength at elevated temperatures — essential for motors and generators that run hot. These are the heavy rare earths, and they are both more concentrated geographically and harder to process than the light rare earths.
If you hear that China has tightened export controls on dysprosium or terbium specifically, that is a serious signal. The alternatives to Chinese heavy rare earth processing are currently limited.
Rare earths and AI systems #
Here is an interesting fact: rare earths are largely not inside the chips themselves.
An advanced AI processor is built primarily from silicon, copper, tungsten, cobalt, tantalum and specialised insulating oxides. Most of these are not rare earths.
Rare earths enter the AI system through everything that surrounds and supports the chips:
- Cooling fan motors — the motors in the thousands of fans in every data centre use NdFeB magnets
- Liquid cooling pump motors — increasingly important as rack density rises
- Compressors for cooling systems
- Precision motors and actuators in semiconductor manufacturing equipment
- Power generators and wind turbines supplying grid electricity
- Hard-disk drive actuators — still used for bulk storage in data centres
- Industrial robots used in chip manufacturing and logistics
- The motors inside ASML machines, Lam Research tools and other fab equipment
The distinction:
Silicon disruption
→ directly threatens chip production
Rare-earth disruption
→ threatens motors, cooling systems,
manufacturing equipment and grid infrastructure
A rare-earth shock would not immediately kill every running GPU. The existing chips would keep working. But it would progressively stop manufacturers from building new cooling equipment, expanding server capacity, maintaining semiconductor manufacturing tools and constructing new data-centre infrastructure. The constraint on growth would be severe.
The countries that matter #
China — the full chain #
China is the largest rare-earth miner. But, it is also the dominant operator of the complete vertical chain from ore to finished magnet.
Even ore mined elsewhere often ends up processed in China, because Chinese separation and refining capacity is so much larger and more developed than anywhere else. A new mine in Australia or Canada may produce mixed concentrate that gets shipped to a Chinese facility for the difficult separation and downstream steps.
The correct way to think about it:
China's rare-earth industrial ecosystem — treated as a single integrated node — is currently irreplaceable at global scale. No individual company within it is as singular as ASML is in EUV lithography. But the system as a whole has no equivalent elsewhere.
Australia — the most important alternative #
Australia has both significant rare-earth resources and the most strategically important integrated producer outside China: Lynas Rare Earths.
Lynas operates the Mt Weld mine in Western Australia — one of the richest known rare-earth deposits. It does initial processing at Kalgoorlie, also in Western Australia. The separated rare-earth products are produced at a large facility in Kuantan, Malaysia.
Lynas is strategically critical precisely because it operates a mine-to-separated-product chain that does not pass through China. That makes it one of a very small number of companies in the world that can deliver refined rare-earth materials from outside the Chinese system.
However, Lynas alone cannot replace Chinese scale. It can supply a meaningful portion of non-Chinese demand for some light rare earths. It is not a complete substitute.
United States — the emerging domestic chain #
The principal US rare-earth operation is MP Materials, which runs the Mountain Pass mine in California — the largest rare-earth deposit in North America — and has expanded into magnet manufacturing in Fort Worth, Texas.
MP Materials represents the most serious US attempt at building an end-to-end domestic chain from mine to magnet. It is strategically important. Its heavy rare-earth separation capabilities are still developing, and its magnet production is ramping toward meaningful scale.
The CHIPS and Science Act and related policy initiatives have pushed significant funding toward domestic rare-earth and critical mineral processing. The ambition is clear. The timeline to full self-sufficiency is not.
Myanmar — the hidden heavy rare-earth source #
Myanmar has become a significant source of heavy rare-earth feedstock, particularly from ion-adsorption clay deposits that are rich in dysprosium and terbium. These deposits are easier to mine than hard-rock deposits.
The complication is that most of this material flows into Chinese processing chains. Myanmar's geopolitical situation also makes it an unreliable long-term supply source for countries trying to reduce China dependence.
New mine is not enough #
When a country announces a new rare-earth deposit — and politicians often treat such announcements as solving the supply problem — what has actually been demonstrated is step one of a nine-step process:
- Geological resource confirmed ✓
- Economic feasibility study
- Environmental approval and permitting
- Mining infrastructure built
- Ore concentration facilities built
- Chemical separation technology developed and qualified
- Metal and alloy production established
- Magnet manufacturing qualified to commercial standards
- Customer qualification and supply agreements
Steps 2 through 9 typically take between five and fifteen years. For heavy rare earth separation specifically, the chemistry and process knowledge required for step 6 is not widely available outside China. And step 8 — qualifying a new magnet for use in motors, generators and semiconductor equipment — requires extensive testing and certification by the manufacturers who will use them.
The difficulty of replacement by stage:
| Stage | How hard to replace |
|---|---|
| Finding a deposit | Medium |
| Building a mine | High |
| Separating light rare earths | Very high |
| Separating heavy rare earths | Extremely high |
| Producing qualified NdFeB magnets | Extremely high |
| Building motors and components at industrial scale | High |
A terminology correction #
A wide range of critical materials gets lumped under "rare earths" in political discussions. This creates confusion about where the actual risks lie. These materials are strategically important for semiconductors and AI — but they are not rare-earth elements:
- gallium and germanium (semiconductor compounds)
- lithium (batteries)
- cobalt (battery cathodes, some semiconductor uses)
- tungsten (chip interconnects)
- tantalum (capacitors and chip layers)
- indium (displays and compound semiconductors)
- hafnium (gate dielectrics in advanced chips)
- platinum-group metals (various industrial uses)
China also has significant control over several of these non-rare-earth critical materials — especially gallium and germanium, where it has applied export controls in recent years.
The correct way to think about AI-critical materials is as three separate strategic groups:
Group 1 — Semiconductor foundation materials
Silicon, gallium, germanium, hafnium, tungsten
→ directly in chips and transistors
Group 2 — Electrical and construction materials
Copper, aluminium, steel, lithium, nickel
→ in cables, cooling, batteries, structure
Group 3 — Rare-earth functional materials
Neodymium, praseodymium, dysprosium, terbium
→ in motors, magnets, cooling, power equipment
Each group has different geography, different supply risks, and different timelines for any potential disruption. Treating them as one problem leads to both over-reaction in some areas and under-reaction in others.
The corrected rare-earth conclusion #
Rare earths are not geographically locked to China. The ore is not the chokepoint.
What China controls is the industrial system — separation chemistry, refining capacity, alloy production, magnet manufacturing — that converts mixed ore concentrate into the finished components that everything else depends on.
That system took decades to build. It benefits from scale, infrastructure, engineering knowledge and downstream customer relationships that cannot be recreated with a single investment announcement or a government subsidy program.
Lynas in Australia and MP Materials in the US are the most serious attempts to build alternatives. Both matter enormously. Neither currently replaces what China provides at scale, particularly for heavy rare earths and high-performance magnets.
The practical upside — compared to the ASML situation — is that rare-earth disruption would affect AI systems more slowly and indirectly than a disruption to EUV lithography or TSMC fabrication. The magnets and motors already installed would keep working. New capacity would be constrained. The pressure would build over months and years rather than stopping production overnight.