It has recently been proposed in Refs. [1,2,3] that the dark energy could be attributed to the cosmological properties of a scalar field with a non-standard dispersion relation that decreases exponentially at wave-numbers larger than Planck scale (k phys > M Pl ). In this scenario, the energy density stored in the modes of trans-Planckian wave-numbers but sub-Hubble frequencies produced by amplification of the vacuum quantum fluctuations would account naturally for the dark energy. The present article examines this model in detail and shows step by step that it does not work. In particular, we show that this model cannot make definite predictions since there is no welldefined vacuum state in the region of wave-numbers considered, hence the initial data cannot be specified unambiguously. We also show that for most choices of initial data this scenario implies the production of a large amount of energy density (of order ∼ M 4 Pl ) for modes with momenta ∼ M Pl , far in excess of the background energy density. We evaluate the amount of fine-tuning in the initial data necessary to avoid this back-reaction problem and find it is of order H/M Pl . We also argue that the equation of state of the trans-Planckian modes is not vacuum-like. Therefore this model does not provide a suitable explanation for the dark energy.