The Semantic Web is potentially the massive global open knowledge base that Artificial Intelligence has been looking for since its origins. After two decades, Linked Open Data is the closest existing exemplar of such a resource. Nevertheless, LOD and similar, maybe bigger, private knowledge graphs have unlocked just a little beyond encyclopaedic question-answering. Many expected that the Semantic Web would encode common sense knowledge, i.e. what AI is mostly missing. Even machine learning approaches, which have pushed forward many machine intelligent tasks, show their limits when generalisation capabilities, such as generating or using common sense knowledge, are required. The focus of this talk is on the role of the Semantic Web as a source of global knowledge for AI at large, traversing both lessons learned and open problems.