This dissertation examines the construction of antimicrobial resistance (AMR) as a public health risk. Its focus is on how AMR is co-produced among a network of medical professionals, scientists and science journalists. The research advances three main arguments: first, narratives and definitions of health risk are not absolute or fixed, but constituted in the discourses and practices of global, national and local actors; second, the production of knowledge about the risk of AMR is not a linear process but one in which various definitions, interests, and practices are involved, and influence one another; and third, conceptualizing health risk as discursive co-production provides a more robust and nuanced understanding of how risks are defined and understood by stakeholders, particularly in relation to attributions of responsibility, blame, victimhood, and resource allocation. I argue that this represents a novel way of imagining and conceptualizing risk communication. The research involved the development of a novel methodology, which I call ethnography of risk, that brings together hospital ethnography, in-depth interviews, and qualitative analysis of media coverage and policy documents. The results of this study show that health risks are co-produced through processes of negotiation between different and co-existing types of knowledge, including situational and embodied experience, emotional memory, and expert assessments. Second, it argues that risks are multifaceted and constituted at the intersection of different perspectives, such that AMR is understood and addressed as a personal risk, a professional risk, a global risk, and a political risk. Third, it shows that stakeholders perform boundary work and blame shifting to justify why they preferred certain ways of knowledge over others. Fourth, various stakeholders reified, and co-produced, the deficit model of risk communication through narratives and actions that keep creating the conditions in which the supposed knowledge deficit is circulated. Finally, AMR lacks a compelling narrative and is communicated as abstract, lost in a plethora or other, more urgent risks. These results open up new ways of conceptualizing health risks beyond the biomedical model and emphasize the need for studies in risk communication and health communication that critically examine the production and circulation of risk knowledge.