But why is that?

Because in a world with a perfectly linear sensor response one stop more light means twice as many photons on the sensor and electrons in the well, thus one more binary digit to represent that doubling. That assumes no other interference, like noise in the lower digits caused by the readout electronics.

Once you apply a gamma curve the whole relation vanishes and any number of bits can represent arbitrary dynamic range, although at the cost of tonality - you may experience banding.

Thank you!

I learn something every day