Materials Databases: Foundations of Modern Digital Materials

, we map the ecosystem of computational and experimental databases, and argue that database architecture, which covers ingestion, curation, metadata, provenance, and access interfaces, strongly influences the performance and trustworthiness of modern AI models. We classify computational repositories into bulk-property and surface/interface resources, and summarize representative experimental databases spanning crystal structures, catalysis, energy storage, and characterization. Beyond single-modality repositories, we highlight integrated platforms that connect computed descriptors with context-rich experimental evidence and tool interfaces, enabling iterative hypothesis testing and closed-loop validation. Building on these examples, we propose a database-model-experiment roadmap for training and deploying graph neural networks, machine learning interatomic potentials, and large language model-based AI Agents. Finally, we outline key bottlenecks that must be addressed for reliable autonomous discovery, including FAIR (Findable, Accessible, Interoperable, Reusable)-aligned standardization, bias and missing negative results, and cross-code reproducibility.