Let's say you want pi to a million digits. Using a lazy list populated by the Machin's formula will take you several hours of calculation. But a hard-coded decimal representation of a million digits will take less than a megabyte of memory/disk space, even if inefficiently stored as one digit per byte and can be retrieved virtually instantly.

In practice, there's little reason to need that sort of precision. A 64-bit float has about 20 decimal digits of precision, which is enough to calculate the circumference of the Earth to less than a hair's breadth. Apparently 61 digits of pi is enough to calculate the circumference of the observable universe to within one Planck length.

In practice, floats are simply good enough for most purposes.