From the source
- IBM and Confluent are putting IBM's time-series foundation models natively inside Confluent, so teams can run forecasting and anomaly detection on streaming data with SQL through Apache Flink.
- The models have more than 50 million downloads, and the work answers a timing problem: a forecast that arrives after a production run, a stockout or a payment authorization is accurate but too late to change the outcome.
- In Confluent Cloud the models are fully managed with nothing to deploy, while on premises they run inside a customer's own infrastructure using Confluent Platform for Apache Flink.
- They run on CPUs without GPUs, which the companies say can reduce the need for dedicated GPU infrastructure and simplify deployment.
- IBM and Confluent describe uses across manufacturing, retail, banking and airlines, including keeping production lines in spec, detecting unusual store activity, assessing card payments and forecasting trading volume.
- One illustrated scenario has a pretrained TTM model read SCADA telemetry and QA measurements every 10 minutes and forecast the share of ice cream tubs expected to be in spec four hours ahead.






