Take or Pay
A take-or-pay capacity agreement is a foundry contract in which a customer commits to pay for a guaranteed minimum volume of wafer capacity over a set period, whether or not it actually orders that many wafers, in exchange for a locked-in reservation on capacity that would otherwise go to the highest bidder.
The floor is the entire point of the contract. Below the committed volume the customer's bill does not change, which transfers demand risk from the foundry to the customer; above it, additional wafers are usually billed per order the normal way, so the agreement behaves like a fixed floor with variable volume stacked on top rather than a flat price for everything.
This is how foundries justify building capacity before demand exists. A gigafab-scale module can cost billions of dollars and takes years to qualify, so a foundry will not commit that capital on a customer's forecast alone; a signed take-or-pay commitment, often backed by a cash deposit, is what lets a leading-edge line get built on a specific timeline rather than waiting for orders to prove the demand first.
| Deal type | Who carries demand risk | Typical use |
|---|---|---|
| Take-or-pay | customer pays the floor regardless of use | securing scarce leading-edge capacity |
| Spot / open market | foundry, in the form of idle capacity | mature-node, price-sensitive orders |
| Prepayment / deposit | customer, cash committed up front | new node ramp, e.g. Apple-TSMC 3nm deals |
| Joint venture capacity | shared between both parties | Sony-TSMC Kumamoto, memory JV fabs |
Apple's deals with TSMC are the most visible public example. Reporting on TSMC's 3 nm and 2 nm ramps has repeatedly described Apple prepaying for guaranteed capacity allocation to secure first access to a new node, a structure that functions like a take-or-pay commitment even where the exact contract terms are not public; Sony's joint investment in TSMC's Kumamoto fab is a related but distinct structure, where the customer takes an equity stake in the capacity itself rather than only a purchase commitment.
The customer is betting that guaranteed access is worth the downside. If end demand falls short, the customer still pays the floor and effectively buys wafers it does not need; the trade only makes sense when losing access to leading-edge capacity to a competitor would cost more than that worst-case overpayment, which is exactly the position a company shipping a flagship product on a brand-new node is usually in.
Read a take-or-pay deal through a *risk-transfer* lens rather than a *price* lens: the headline wafer price is almost beside the point, because the actual asset being sold is certainty — a guaranteed slot on a gigafab line that someone else cannot take from you — and that certainty is precisely what justifies a foundry breaking ground on a multi-billion-dollar module before a single customer wafer has been ordered.