For an ideal band-limited additive white Gaussian noise channel with bandwidth $B$ and received signal-to-noise ratio $\mathrm{SNR}$, the maximum reliable information rate is

$$C=B\log_2(1+\mathrm{SNR})$$

bits per second.

This is the Shannon-Hartley theorem.

Capacity increases with both bandwidth and SNR, but neither resource gives unlimited linear improvement. Because SNR appears inside a logarithm, repeatedly multiplying signal power produces diminishing increases in capacity at fixed bandwidth.

The theorem also exposes a power-bandwidth trade-off. A system can sometimes compensate for lower SNR by using more bandwidth, or conserve bandwidth by operating at higher SNR.

$C$ is not the bit rate of a particular modulation. It is an asymptotic information-theoretic limit for the stated channel model. Practical modulation and coding schemes operate below this boundary and differ in how closely they approach it for finite complexity and delay.