Helmholtz Munich researcher Fabian Theis, one of the central figures behind the virtual cell concept, argues that the field should move beyond predicting how cells respond to change and toward controlling biological outcomes. The shift comes as large funders commit serious money to the data that goal requires. In a Latent Space podcast interview published November 6, Theis discussed where virtual cell research stands after a difficult stretch in which simple baselines often matched sophisticated models. The backdrop is $1.8 billion in commitments announced for the Biohub initiative, with backing from Biohub, the US National Institutes of Health, the Department of Energy, Google and Meta. The initiative centres on new imaging and measurement tools that generate large, co-registered datasets intended for AI training. Interest in virtual cells cooled after the 2025 Arc Virtual Cell Challenge, where several papers found that a constant baseline often beat models at perturbation prediction, meaning forecasting how a cell's state changes after gene editing, drugs or environmental shifts. Some researchers noted that common evaluation metrics can reward uninformative average predictions because most genes barely change. With better-calibrated metrics, deep learning models can outperform that baseline. Theis's position is that perturbation prediction on existing datasets is not a hard enough problem to justify complex algorithms. He argues the real challenge is predicting how biology behaves across scales, from molecules to cells, niches, tissues, organs and whole organisms, and how each level influences the others. Researchers lack a clear understanding of these scale interfaces and the data needed to study them, and the Biohub effort is aimed at that data gap. Theis's work also includes lab-in-the-loop experimentation, where models propose experiments and results feed back into them, and precision medicine research at Helmholtz's Computational Health Center, which combines AI with molecular, imaging, genomic and clinical data. For Australian businesses and professionals, this is mainly a research and funding story with little direct effect today, since nothing discussed is a ready-to-use tool for most firms. Organisations in pharmaceutical, biotech and health research that handle genomic, clinical or imaging data should note that Australia's Privacy Act treats health information as sensitive information, so any data sharing or AI training needs to be checked against those obligations. Everyone else can reasonably set this aside for now, though it is worth watching as public datasets and tools mature.