Each "bit" in the architecture is mainly related to how many different memory positions you can address with a CPU opcode (i.e. an assembly command) that needs to access memory or with a CPU register that holds an memory address (so an 8 bit register can only address 256 memory positions, a 16 bit one 65,538 memory positions, a 32 bit one 4,294,967,296 memory positions and so on) - basically every additional bit in the register size of a CPU that addresses memory roughly doubles the memory that CPU can access and thus use (not considering hacks to go around those limits, such as the Spectrum 128K had).

Meanwhile Moore's Law predicts that "the number of transistors in an integrated circuit doubles about every two years" which for integrated circuits roughly means that every two years the amount of memory that can be in the same integrated circuit area doubles.

So if Moore's Law kept working, each architecture bit roughly gives you two more years until the availability of memory catches up- This means that the transition from 32-bit to 64-bit architectures per Moore's Law gave us 64 years until we have to move to an architecture with more bits because memory available is so big that CPUs are unable to use so much memory.

Were we are now, 64 bits is enough to address 8 Exabytes of memory.

That said, extrapolating from the current trend in growth of memory in PCs, we will reach the point when we will start hitting the constraints of that in ... NEVER - thanks to the AI bubble the newer PCs are coming out with the same or less memory than the previous generation, so the trend of available memory is not upwards but downwards.