Essentially all stars form in giant molecular clouds ( GMCs ) . However , inside GMCs , most of the gas does not participate in star formation ; rather , denser gas accumulates in clumps in the GMC , with the bulk of the stars in a given GMC forming in a few of the most massive clumps . In the Milky Way , these clumps have masses M _ { cl } \lesssim 5 \times 10 ^ { -2 } of the GMC , radii r _ { cl } \sim 1 { pc } , and free-fall times \tau _ { cl } \sim 2 \times 10 ^ { 5 } { yr } . We show that clumps inside giant molecular clouds should accrete at a modified Bondi accretion rate , which depends on clump mass as \dot { M } _ { cl } \sim M _ { cl } ^ { 5 / 4 } . This rate is initially rather slow , usually slower than the initial star formation rate inside the clump ( we adopt the common assumption that inside the clump , \dot { M } _ { * } = \epsilon _ { ff } M _ { cl } / \tau _ { cl } , with \epsilon _ { ff } \approx 0.017 ) . However , after \sim 2 GMC free-fall times \tau _ { GMC } , the clump accretion rate accelerates rapidly ; formally , the clump can accrete the entire GMC in \sim 3 \tau _ { GMC } . At the same time , the star formation rate accelerates , tracking the Bondi accretion rate . If the GMC is disrupted by feedback from the largest clump , half the stars in that clump form in the final \tau _ { GMC } before the GMC is disrupted . The theory predicts that the distribution of effective star formation rates , measured per GMC free-fall time , is broad , ranging from \sim 0.001 up to 0.1 or larger and that the mass spectrum of star clusters is flatter than that of clumps , consistent with observations .