The most promising method for measuring primordial non-Gaussianity in the post-Planck era is to detect large-scale, scale-dependent galaxy bias. Considering the information in the galaxy power spectrum, we here derive the properties of a galaxy clustering survey that would optimize constraints on primordial nonGaussianity using this technique. Specifically, we ask the question of what survey design is needed to reach a precision σðf loc NL Þ ≈ 1. To answer this question, we calculate the sensitivity to f loc NL as a function of galaxy number density, redshift accuracy and sky coverage. We include the multitracer technique, which helps minimize cosmic variance noise, by considering the possibility of dividing the galaxy sample into stellar mass bins. We show that the ideal survey for f loc NL looks very different than most galaxy redshift surveys scheduled for the near future. Since those are more or less optimized for measuring the baryon acoustic oscillation scale, they typically require spectroscopic redshifts. On the contrary, to optimize the f loc NL measurement, a deep, wide, multiband imaging survey is preferred. An uncertainty σðf loc NL Þ ¼ 1 can be reached with a full-sky survey that is complete to an i-band AB magnitude i ≈ 23 and has a number density ∼8 arcmin −2 . Requirements on the multiband photometry are set by a modest photo-z accuracy σðzÞ= ð1 þ zÞ < 0.1 and the ability to measure stellar mass with a precision ∼0.2 dex or better (or another proxy for halo mass with equivalent scatter). Finally, we estimate that for the idealized case of a survey measuring all halos down to a mass 10 10 h −1 M ⊙ on the full sky out to high redshift, in principle a precision of order σðf NL Þ ∼ 0.1 can be achieved.