First-generation R&D-based endogenous growth models have been criticized because they predict strong scale effects (growth rate proportional to the size of the population) and because they make the knife-edge assumption of exactly constant returns to scale to man-made inputs. These two drawbacks are absent in semi-endogenous growth models, which nevertheless require an exogenous source of growth (usually population growth). More recently, R&D-based growth models have successfully address the problem of strong scale effects, by considering vertical and horizontal innovation, but are still conditioned by alternative knife-edge assumptions. We revisit these criticisms introducing an exhaustible natural resource as an essential input which makes increasing returns to scale to man-made inputs compatible with non-explosive sustained growth. The instability problem usually associated with increasing returns is overcome thanks to the existence of imperfect markets and knowledge externalities in a decentralized economy with two separate sectors: final output and R&D. Both sectors, the former directly and the latter indirectly, depend on the natural resource. Without the requirement of any knife-edge assumption, we prove that growth is fully endogenous, stable and void of strong scale effects for a general R&D-based growth model.