An important challenge in the field of three-dimensional ultrafast laser processing is to achieve permanent modifications in the bulk of silicon and narrow-gap materials. Recent attempts by increasing the energy of infrared ultrashort pulses have simply failed. Here, we establish that it is because focusing with a maximum numerical aperture of about 1.5 with conventional schemes does not allow overcoming strong nonlinear and plasma effects in the pre-focal region. We circumvent this limitation by exploiting solid-immersion focusing, in analogy to techniques applied in advanced microscopy and lithography. By creating the conditions for an interaction with an extreme numerical aperture near 3 in a perfect spherical sample, repeatable femtosecond optical breakdown and controllable refractive index modifications are achieved inside silicon. This opens the door to the direct writing of three-dimensional monolithic devices for silicon photonics. It also provides perspectives for new strong-field physics and warm-dense-matter plasma experiments.