Aims : Establishing precisely how stars and interstellar medium distribute within the central 100 pc area around an AGN, down to the pc scale, is key to understanding the late stages of transfer of matter onto the accretion disc. Methods : Using adaptive optics-assisted (SPHERE-VLT) near-IR images in the H band, Ks band, and several narrow bands of the Seyfert 2 galaxy NGC 1068, we analysed the radial distribution of brightness in the central r <  100 pc area down to the pc scale. The median-averaged radial profiles are fitted by a cusp (power law) plus a central point source. A simple radiative transfer model is also used to interpret the data. Results : We find that the fit of radial brightness profiles beyond 10 pc is done quite precisely at Ks band by a cusp of exponent −2.0 plus a central point source and by a cusp of exponent −1.2 at H. The difference of exponents between H band and Ks band can be explained by differential extinction, provided that the distribution of dust is itself cuspy, with an exponent −1.0. However, the derived stellar density is found to follow a r−4 cusp, which is much steeper than any other cusp, either theoretically predicted around a massive black hole, or observed in the centre of early- and late-type galaxies or in mergers. Introducing a segregation in the stellar population with a central excess of giant stars leads to a somewhat less steep exponent; however, the de-reddened luminosity of the stellar cusp, as well as the mass of dust and gas all appear much too high to be realistic. An alternative scenario, where the Ks-band profile is well fitted by a combination of radiation from a stellar cusp identical to the H-band profile and thermal emission of warm/hot dust heated by the central engine appears much more satisfactory. NGC 1068 is shown to satisfy a relationship between half-light radius, cusp luminosity, and exponent that we established using a sample of luminous infrared galaxies (LIRGs) and ultraluminous infrared galaxies ULIRGs. This suggests that the cusp is the remnant of a recent starburst. We identify the central point-like source with the very hot dust at the internal wall of the putative torus and derive an intrinsic luminosity that requires a central extinction AK ≈ 8, a value consistent with predictions by several torus models. Conclusions : The overall picture revealed by this study is closely consistent with the scheme of a central rather steep stellar cusp, embedded in a diluted medium of warm dust, while a compact and dense structure identified with the putative torus is required to interpret a highly reddened point-like central source of very hot dust.