Real-Time Intelligence with IBM Time Series Models on Confluent
huggingface ·
IBM and Confluent announced early access to IBM Granite time series foundation models on Confluent Cloud, enabling real-time forecasting and anomaly detection directly on streaming data via Flink SQL. The partnership aims to replace costly, bespoke per-series models by letting domain experts apply pretrained models without building a dedicated ML stack. A portfolio of four models (PatchTST-FM, FlowState, TTM, and TSPulse) supports different needs such as probabilistic forecasting, irregular time steps, large-scale CPU inference, and anomaly detection. Native inference in Confluent provides zero configuration, built-in governance and traceability, lower cost, and stronger security because data stays within Confluent Cloud. Value examples include factory-line forecasting and anomaly detection that spot drift before failures, plus grocery demand planning across the entire SKU catalogue. IBM reported that each percentage point of accuracy is worth millions and productivity gains range from 5 to 10 times. The models are also available for on-premises use through Confluent Platform in the future.
huggingface