IBM and Confluent are bringing time-series foundation models closer to the data as it flows, with the models now running natively inside Confluent as built-in functions.
In Confluent Cloud the models are fully managed, with nothing to deploy. On premises, they run within a customer's own infrastructure using Confluent Platform for Apache Flink. The models run on CPUs, without requiring GPUs.
The family of four models has passed 50 million downloads. Like foundation models for text and images, time-series foundation models learn general patterns from diverse historical data, then apply that knowledge to new series and tasks — giving teams a starting point without training a separate model for every series. Depending on the task, teams can also fine-tune a model or combine its outputs with other business logic.
With Apache Flink, teams can apply forecasting, anomaly detection and semantic intelligence to streaming data using SQL, making model outputs available to the applications and people responsible for the next action.
The approach covers mission-critical use cases across industries. In production, a pretrained model can forecast the share of tubs expected to be in spec four hours ahead and help operators evaluate setpoints before changing them. In retail, it can forecast demand across products and stores, producing a 28-day forecast refreshed as new sales data arrives. In banking, it can flag card payments that fall outside a card's expected spending range, and on trading desks it can forecast volume to pace large orders.
The same building blocks support semantic search: time-series embeddings can retrieve historical patterns across equipment, so an engineer investigating an unusual sensor reading can search for similar shapes, such as a spike and recovery or a gradual rise, and inspect the associated maintenance records.




