Breakthrough technological inventions create the foundation for many innovative opportunities. Through novel scientific fundamentals and unique combinations of knowledge, they form a new basis for technology development and innovation, shifting their industry's mindset about what is feasible and valuable. Prior research has, to date, characteristically taken an organizational‐level perspective on technology breakthroughs, seeking to explain the occurrence of breakthroughs as a function of firm‐level characteristics and processes. This tells us less, however, about the underlying knowledge structures of the technologies themselves. This research is based on the assumption that an examination of the knowledge foundations of high‐potential inventions can enrich our understanding about the underlying features of innovations that transform industries and advance societies. In this manner, we can clarify how certain technologies are advanced and extended, providing the basis for future discoveries. Our analysis focuses on high‐potential patents: those having the highest number of forward citations in a given class. We conduct this analysis on a sample of 298 breakthrough patents and two comparison nonbreakthrough groups in drug and semiconductor classes. Our results show that breakthrough technologies, compared with nonbreakthroughs, are more likely to build on: (1) the past technological developments of others, by backward citing and embodying knowledge from prior discoveries (earlier patents), but not one's own previous developments; (2) the latest technologies, by backward citing recent patents; (3) geographic proximity, evidenced by the fact that the focal patents backward cite patents from one's own country; and (4) greater technological breadth, evidenced by the fact that knowledge captured in the patents encompasses a larger number of technical fields. Based on these results, we frame the occurrence of breakthroughs as involving a process where these significant developments happen early in the progress of a technology, but after some relevant knowledge has accumulated. This reinforces the concept that a learning effect needs to occur before breakthroughs can happen. In addition, our findings suggest that the knowledge underlying breakthroughs is likely to come from other organizations or individuals, rather than the developing entity. They also incorporate broader insights from technological diversity but do not exhibit geographic diversity. Instead, they are associated with geographic proximity, which may better enable knowledge sharing and integration given the reliance on other entities and diverse knowledge.