IBM Granite Time Series Models Now Run on Confluent Cloud

According to the post, Confluent manages model serving, infrastructure, scaling and runtime operations, so users do not need a separate machine learning platform, data warehouse or provider credentials. The models can be invoked from Flink SQL, and access opens on Confluent Cloud on AWS. Confluent Platform support is planned for on-premises and hybrid environments. The post said this removes the months usually spent wiring models into production and avoids a database hit on every call.

The post described the release as part of a broader shift away from custom, project-based analytics. It said traditional decisions relied on "one bespoke model at a time and months of expert work on each," so teams modeled only a few hundred series and covered the rest with safety margins. A time series foundation model, by contrast, is trained once on varied signals and generalizes to series it has never seen, producing forecasts, anomaly scores, similar historical patterns and control conditions from an input window. These functions are meant for demand planners, fraud analysts and process engineers to use on their own streams, with no data science team required.

The post said IBM's Granite time series models have been downloaded more than 44 million times. IBM tested the models in its own products and with design partners in cement, steel, pulp and paper, food and telecommunications. According to the post, every point of accuracy is worth millions, productivity gains run 5 to 10 times, and work that previously waited for specialists now sits with domain experts who own the decision.

As a use case, the post cited a tempering line in a chocolate factory, where temperature, speed and throughput are sampled every few seconds. The model can forecast the line's output through the evening shift, allowing a planner to act before a shortfall occurs. It can score today's run against the line's normal dark-chocolate behavior, find the closest match in plant history, and condition on settings controlled by the crew.

Confluent said its data streaming platform provides the live business context the models need. It continuously streams, connects, governs and processes real-time data, so each model receives recent history, keyed per series, without keeping a separate data store. The resulting capabilities include forecasting, anomaly detection, similarity search, classification, gap-filling and optimization directly where the data lives.

Editor's Summary

IBM and Confluent launched IBM Granite time series foundation models on Confluent Cloud, allowing real-time forecasting and anomaly detection to run in Apache Flink without extra data infrastructure. The service, initially available on Confluent Cloud on AWS, is aimed at domain experts and will be extended to on-premises Confluent Platform deployments.