CoreWeave partners with Harell Data for AI training on proprietary biotech datasets

CoreWeave is betting that the next frontier for its GPU cloud business isn’t just training chatbots. It’s training AI models on biotech data that’s too valuable, and too sensitive, for anyone to let leave the building. The Nasdaq-listed cloud infrastructure company announced a collaboration with Harell Data on September 23 to deploy CoreWeave’s GPU resources inside Harell’s secure AI training platform. The setup lets biotech companies monetize proprietary scientific datasets without ever handing over the raw data itself. How the deal actually works CoreWeave will supply NVIDIA A100 and Hopper series GPUs to handle training, fine-tuning, and inference workloads. The critical wrinkle: all of that compute happens within Harell’s environment, meaning the underlying datasets never get extracted or transferred to a third party. The business model is built around what Harell calls a Models-as-a-Service marketplace. Data owners host their proprietary datasets on the platform. AI developers pay metered compute fees to train models against that data. The data owners then receive a revenue share based on how their datasets contribute to model performance. The platform has already launched ear...