The parts list runs to thousands of items, and until recently almost nobody near the top of the supply chain read it for chemistry. Now a team at Taiwan Semiconductor Manufacturing Co. is working through the equipment stack looking for dysprosium, terbium and neodymium, the magnetic ingredients that help a modern fab run and that also sit inside one of the world’s most contested export regimes.
TSMC has begun assessing whether it can replace rare-earth components in its production tools with alternatives that use none of the metals, according to three people familiar with the matter. The review is aimed at reducing the risk that export controls interrupt output, the people said. Reuters reported the effort on Oct. 22.
The work is in an early phase. Any substitution sends a component back through supplier qualification and, in most cases, a fresh round of testing before it can be used at volume, the people said. A TSMC spokesman declined to comment on the review.
The company said separately that operations were normal after a typhoon passed over the island, with no major impact.
For chipmakers, rare earths have long been treated as someone else’s problem. The elements appear in only trace amounts in the silicon itself. The exposure sits in the machinery: the permanent magnets that drive linear motors on lithography stages, the sensors that hold a chamber at a fixed distance, the pumps and robots that shuttle a wafer between process steps.
A single advanced tool can carry dozens of such assemblies, and each one is qualified as part of the tool. Swap a magnet and you are not changing a consumable. You are reopening a qualification file that took years to close.
The trigger is policy, not geology. Beijing has spent two years tightening control over rare-earth exports, first with licensing requirements on seven medium and heavy elements in April 2025 and then with a broader set of rules later that year that reached magnets embedded in finished goods. Chinese processors account for roughly 90% of global refining capacity, according to industry estimates, a concentration that has no near-term substitute.
Those rules were written to be applied selectively. Exporters must apply for licenses, and the review process gives Beijing a dial it can turn without banning anything outright. For a fab operator, that is worse than a tariff. A tariff has a price. A licence has a queue.
“The lesson of the past two years is that the risk migrates,” said one supply-chain consultant who advises chipmakers on materials sourcing. “You solve the wafer, and then you find out the problem is in the motor.”
Equipment vendors are running their own reviews, according to two people at suppliers who asked not to be named discussing customer work. The large toolmakers have an advantage TSMC does not: they know which sub-assemblies use sintered magnets and which use bonded ones, and they can redesign a stage around a different magnet grade if a customer is willing to pay for the engineering.
Willingness is the constraint. Ceramic and ferrite alternatives exist for some low-force applications, but they are bulkier and less stable at temperature. A magnet that loses strength inside a vacuum chamber is not a drop-in replacement for one that does not.
Outside China, magnet capacity is thin and already spoken for. Producers in the U.S., Japan and Germany have announced expansions, and much of that output is committed to defence and automotive customers on multi-year contracts. A chipmaker arriving in 2026 with a request for qualified, traceable, non-Chinese high-grade magnets joins a queue of its own.
That leaves time as the binding constraint. People involved in the review describe a horizon of years rather than quarters, with the first wins likely to come in support equipment, the pumps and chillers and handling systems where a slightly larger motor costs little. The precision stages at the centre of a fab are the hardest case and will be the last to move.
There is also the question of whether the effort changes anything if the controls never bite. TSMC’s customers, which include Apple and Nvidia, have begun asking suppliers for documentation on material origin, and documentation is cheaper to produce than a redesign. Some of the work now under way may end as a paper trail rather than a new parts list.
Analysts said the review is best read as insurance rather than a rupture. A foundry that can show a path to rare-earth-free tools has more room to negotiate, both with suppliers and with governments deciding whether to grant licences. The alternative, holding no fallback at all, leaves a company exposed to a decision made elsewhere.
What the review will not do is move the risk off the island. Taiwan imports nearly all of its rare earths, and the tools that use them are built in Europe, Japan and the U.S. Stripping the metals out of the equipment shifts the dependency from a mine in Inner Mongolia to a magnet plant in Vietnam or South Carolina, and from a customs form to a supplier’s capacity plan.
For now the deliverable is a map: which tools matter, which components can change, and how long the change would take. TSMC has not said when the assessment will conclude, and people familiar with it said no decision has been made on whether to fund the substitutions at scale.
The next signal will be quieter than an announcement. It will be a supplier receiving a specification sheet with a line item that says, simply, no rare earth.
