Cyberinfrastructure is an essential part of how science is done today, but sustaining the projects that provide these services is a challenge. We studied 11 long-term data infrastructure projects, most focused on the Earth Sciences, to understand the characteristics that contributed to their sustainability. Among our sample group, we noted the existence of three types of project groupings: Database, Framework, and Middleware. Database projects aim to bring together data and data resources for use. Middleware projects seek to develop software and technology. Framework projects focus on developing best practices. The conflicting expectations, limitations, and needs of academia and cyberinfrastructure development presented challenges for all project types. Although none of the studied projects began with a formal governance model, each project adopted a model over time. Most of the efforts started as funded research projects, and nearly all became organizations in order to become sustainable. Projects were often funded for short time scales, but had the long-term burden of sustaining and supporting open science, interoperability, and community building–activities that are difficult to fund directly. This transition from “project” to “organization” was challenging for most efforts, and specifically in regards to leadership change and funding issues. Some common approaches to sustainability were identified within each project grouping. Framework and Database projects both rely heavily on the commitment to, and contribution from, a disciplinary community. Framework projects often used bottom-up governance approaches to maintain the active participation and interest of their community. Database projects succeeded when they were able to position themselves as part of the core workflow for disciplinary-specific scientific research. Middleware projects borrowed heavily from sustainability models used by software companies, while maintaining strong scientific partnerships. Cyberinfrastructure for science requires considerable resources to develop and sustain itself, and much of these resources are provided through in-kind support from academics, researchers, and their institutes. More work is need to find appropriate models that help sustain key data infrastructure for Earth Science over the long-term.