Infrastructure makes it easier, faster or possible for investigators to study objects of research. It does so by making available consistent and stable resources and services such as data, collaboration tools, sites of sample collection, or calibrated instruments. This paper offers the concept of the kernel of a research infrastructure as a new unit of analysis for investigating the enabling capacities of infrastructure. The kernel is the core resources and services an infrastructure makes available (called the cache), as well as the work, techniques and technologies that go into sustaining that availability (called addressing). By inspecting and comparing the kernel of two long-term scientific enterprises, this paper demonstrates how focusing on the kernel can help explain key qualities of research infrastructure such as flexibility and persistence in the face of dramatic changes to the objects, methods and practice of science.