A capacity built from sectors
The familiar high-density 3½-inch disk has a definite layout behind its label. Microsoft’s archived capacity explanation lists 80 tracks on each of two sides, 18 sectors per track and 512 bytes per sector. Multiply those quantities and the result is 1,474,560 bytes. IBM’s diskette-driver documentation describes the same double-sided, 80-cylinder, 18-sector format and specifies 512-byte sectors.
Those are counts of places where data can be stored. The capacity is not an arbitrary decimal chosen for the packet; it follows from the geometry of this particular disk format. Other floppy formats can use different layouts, so the calculation describes the common 1.44 MB format rather than every object called a floppy disk.
The label changes units halfway
Dividing 1,474,560 by 1,024 produces 1,440. In binary units that is 1,440 KiB. Divide that number by 1,000 and the familiar 1.44 appears. The name combines a binary first step with a decimal second step, which is why treating it as a consistent modern unit causes confusion.
Using decimal megabytes throughout gives 1.47456 MB, because a decimal megabyte is one million bytes. Using binary mebibytes throughout gives 1.40625 MiB, because a mebibyte is 1,048,576 bytes. These are different descriptions of the same sector total. Changing the unit does not alter how much the disk holds.
Files need bookkeeping space
The full sector count is not all available for a document. Microsoft’s example reserves one 512-byte boot sector, 18 sectors for two file allocation tables and 14 sectors for the root directory. Those structures use 16,896 bytes altogether. Subtract them from the sector capacity and 1,457,664 bytes remain for file data in that layout.
The allocation tables record how storage is assigned, while the directory needs room for its entries. Even an otherwise empty disk therefore has work to do before its owner copies anything onto it.
A capacity label, a total byte count and free space answer different questions. The 1.44 MB name survived as a convenient identifier for the format, while the arithmetic explains why software could report a different-looking number without any missing physical storage.