BigInteger would only be used if you know it will not be a decimal and there is a possibility of the long data type not being large enough. BigInteger has no cap on its max size (as large as the RAM on the computer can hold).
From here.
It is implemented using an int[]:
110 /**
111 * The magnitude of this BigInteger, in <i>big-endian</i> order: the
112 * zeroth element of this array is the most-significant int of the
113 * magnitude. The magnitude must be "minimal" in that the most-significant
114 * int ({@code mag[0]}) must be non-zero. This is necessary to
115 * ensure that there is exactly one representation for each BigInteger
116 * value. Note that this implies that the BigInteger zero has a
117 * zero-length mag array.
118 */
119 final int[] mag;
From the source
From the Wikipedia article Arbitrary-precision arithmetic:
Several modern programming languages have built-in support for bignums, and others have libraries available for arbitrary-precision integer and floating-point math. Rather than store values as a fixed number of binary bits related to the size of the processor register, these implementations typically use variable-length arrays of digits.