First, white light consists of a mixture of EM waves of many different frequencies as well as different phases (and, in the unpolarized case, polarizations, although white light can also be polarized). You can't have a superposition of waves of different frequencies be out of phase everywhere - if they're out of phase somewhere, then they'll be in phase somewhere else because they'll have gone through a different number of cycles.
Second, even if you only considered components of a single frequency (i.e. color), white light is an incoherent mixture of random phases and polarizations, not ones that are precisely uniformly distributed around the circle (either in real space for polarizations or the circle of phases). And $N$ randomly chosen (i.e. independent and uniformly distributed) phases don't perfectly cancel out when added; instead, they only partially cancel out and the magnitude of the sum is $\sqrt{N}$ times the magnitude of each term by the central limit theorem. It's actually very difficult to arrange it so that the phases precisely cancel each other out, as that requires strong correlation between the constituent phases.
So the overall amplitude still grows as you add in more contributions - just more slowly than if they were added coherently, as with a laser. This is also why having more people sing in unison (but with randomly offset phases) makes the total sound louder rather than quieter.