The useful question is what changes for users, developers or buyers, and whether the announcement stays industry context or becomes something people can actually use.

Until now, machine learning within data clean rooms was restricted to basic SQL queries or single-node Python, limiting scalability and practical use. Snowflake’s ML Jobs now allows data scientists to deploy standard Python ML stacks with distributed training, hyperparameter optimization, custom packages, and GPU compute directly inside multiparty collaborations. Models train on combined data from multiple organizations while raw records remain in their original accounts, and the process runs automatically without manual intervention. This addresses longstanding challenges in industries like advertising, financial services, and healthcare where multi-source data is critical but sharing raw data is restricted.

The technology enables collaborative model development where each party retains control over their data. For example, an advertiser can build audience and measurement models using ad logs from publishers, identity resolution data from providers, and transaction signals from retailers—all within a governed environment. The advertiser’s proprietary model logic remains hidden from data providers, while providers approve explicit workloads for their data. As new data flows in, models score continuously on fresh signals without renegotiating access, making multiparty ML both practical and scalable for ongoing campaigns.

ML Jobs simplifies the development and operational workflow for data scientists. Users can write standard Python code, use any required packages, and work in familiar IDEs or notebooks. A short YAML spec declares the job’s requirements, eliminating the need for Docker images, container registries, or manual infrastructure provisioning. Scaling to multiple nodes or GPUs is handled via parameter changes, and job registration or updates are managed through spec modifications. Once approved, templates run via a single SQL call from any orchestrator, with activity history and container logs available for auditing and debugging.

The technology unlocks advanced use cases such as causal lift measurement, probabilistic identity resolution, and campaign optimization agents. For instance, a brand and retailer can jointly run an uplift model inside the collaboration, with impression logs and transaction data remaining in their respective accounts. Similarly, identity resolution models can train on combined CRM and graph data without raw records leaving their environments. These capabilities enable more accurate attribution, higher match rates, and dynamic audience targeting—all while preserving data governance and privacy.