A 64-bit ripple-carry adder is 64 gate delays deep and sits on the critical path of nearly every instruction. What does carry-lookahead buy, and what does it cost?

A. Delay growing like log n instead of n, at the cost of many more gates

The carry chain is the bottleneck, and lookahead breaks it by computing each carry directly from the inputs rather than from its predecessor, logarithmic depth for a lot more silicon. It is the area-versus-speed trade in its purest form, which is why real adders are hierarchical compromises rather than either extreme.