A sunk cost is a cost already incurred and cannot be recovered by any future action. Money spent on R&D for a product that won’t ship; legal fees on a deal that fell through; the down-payment on a machine that turned out to be wrong for the job. The defining property is irreversibility: nothing you do now changes that the money is gone.
The decision rule that follows: sunk costs are irrelevant to future decisions. When evaluating two paths forward, the only costs that matter are the costs you’d still incur (or could avoid) by choosing each path. Costs that are the same under both options — including all sunk costs — wash out of the comparison.
This is hard for humans. Many people will throw good money after bad because they “have so much invested already.” The classic example: the company has spent $8M on a product, needs another $2M to finish, but the finished product is forecast to bring in only $1M of revenue. The right decision is to abandon — the project is worth going forward — even though that means accepting the $8M loss. The wrong decision (very common in practice) is to spend the $2M anyway to “salvage” the $8M, which just makes the loss $9M.
In replacement decisions specifically, the original purchase price of the defender (the existing asset you might replace) is a sunk cost. What matters is its current market value (the opportunity cost of not selling it now) and the future operating costs of keeping it vs. replacing it. Book value, original cost, and historical depreciation are all irrelevant to the keep-or-replace decision — only the forward-looking cash flows matter.
The discipline of recognising and ignoring sunk costs is one of the most important habits an engineer-economist can develop. See Opportunity cost for the complementary concept (the cost of not taking an alternative) and Replacement decision for the keep-or-replace application.