Single-sided and double-sided bandwidth needs to be specified explicitly to be clear. The double-sided bandwidth of a modulated signal at baseband (including the negative frequencies) corresponds directly to the occupied bandwidth at RF (passband) in either the positive or negative frequencies: for example a baseband signal that extends from -1 MHz to +1 MHz in spectral occupancy when frequency translated to a real carrier at 100 MHz, will occupy the spectrum from 99 MHz to 101 MHz as well as the spectrum from -101 MHz to -99 MHz.
See the graphic below showing the related spectrums for complex baseband and real passband modulated signals. An IQ Mixer translates the baseband signal to passband.
It is when the baseband signal is completely real that the positive and negative frequency depiction becomes redundant since the the spectrum will be complex conjugate symmetric (the positive and negative frequencies have the same magnitude and opposite phase). For that reason we can describe all the information using a single-sided bandwidth (positive frequencies only). It only makes sense to use single-sided bandwidth for real signals. However the relationship between this real spectrum at baseband considering both positive and negative frequencies and the spectrum in the passband is the same as described above. That said, if a real baseband spectrum has a single-sided bandwidth of 1 MHz (for example), then it's double-sided bandwidth at baseband including both the positive and symmetric negative frequencies will be 2 MHz, and the passband bandwidth will be 2 MHz as well.
For comparison to the prior graphic, below shows the case for spectrums for real baseband and real passband modulated signals. A single real multiplier (mixer) and local oscillator can be used in this case to translate the baseband signal to passband. In this graphic, the double-side bandwidth at baseband is $B$ and the single-sided bandwidth would be $B/2$.
When we use “negative frequencies” we are referring to frequency components of the complex form $e^{j\omega t}$ in contrast to real sinusoidal components such as $\cos(\omega t)$. These other existing posts provide more details on the frequency translation process and how the baseband and passband signals are related as well as intuition for "positive" and "negative" frequencies:
Frequency shifting of a quadrature mixed signal
About the process to convert basedband signal into passband
