Beyond the model: data infrastructure as the foundation for autonomous virtual laboratories

Artificial intelligence (AI) and machine learning are reshaping biotechnology, but their impact is still constrained by inadequate data infrastructure. Although recent advances have focused on model architectures, data quality, standardization, and interoperability remain the main bottlenecks. A data-centric approach that strengthens data practices across the design-build-test-learn cycle is needed to advance AI-enabled biomanufacturing. In this review, we argue that curated repositories, consistent metadata, real-time validation, and efficient learning strategies are essential prerequisites. We highlight the manual test-to-learn ingestion step as a critical source of latency that can be mitigated through semantic standardization and structured data collection. By addressing these data-centric limitations, the field can move toward reliable, scalable, and ultimately autonomous virtual laboratories that adhere to the Findable, Accessible, Interoperable, Reusable principles.