IBM and Confluent are bringing machine learning models designed for time-series prediction directly into enterprise data streaming environments. Under an early access program launched on Confluent Cloud, users can deploy IBM's Granite Time Series foundation models to perform real-time forecasting, anomaly detection, and optimization on streaming data without extracting it to separate machine learning platforms. The partnership aims to democratize predictive capabilities that traditionally required specialized data science teams. The collaboration addresses a longstanding enterprise AI challenge: most machine learning models operate on static datasets pulled from production systems, introducing latency that undercuts decision-making value. The new capability lets business users—demand planners, fraud analysts, process engineers—run forecasting directly on live data streams using Apache Flink. IBM reported that pilot deployments across cement, steel, food production, and telecommunications generated 5-10x productivity improvements in tasks like predictive maintenance and inventory optimization. The models arrive pre-trained on diverse industrial datasets and require no additional configuration to integrate with Confluent's platform. The company handles model serving, infrastructure scaling, and governance internally. Confluent Cloud users on AWS can access the models immediately, with support for on-premises Confluent Platform deployments coming later.