Artificial intelligence is transforming how patent data can be exploited within the scientific value creation cycle. Patents contain rich technical specifications, emerging trends and domain-specific knowledge that are often difficult to access with traditional search and analysis methods. In this talk, I discuss how AI – including NLP, deep learning, knowledge graphs and large language models – enables large-scale indexing, linking and exploration of patent information together with scientific literature, domain-specific knowledge, and research data. I will highlight opportunities for researchers to discover relevant knowledge (about solutions, experiments, indicators, etc.) from linked patent knowledge in order to empower future innovations, and supporting cross-disciplinary scientific exploration for technical and scientific knowledge. At the same time, I will address briefly key risks and open questions: opacity and bias in AI models, the consequences of substituting expert search practices with automated pipelines, and the need for explainable, trustworthy systems with humans firmly in the loop.