IBM and Confluent expand the integration of Granite time-series models in Confluent Cloud with semantic intelligence
IBM's Granite Time Series models have, according to IBM, already surpassed 50 million downloads, and now in Confluent Cloud also offer semantic intelligence alongside the existing forecasting and anomaly detection, accessible via SQL in Apache Flink without a GPU.
IBM states that its Granite time-series foundation models have reached more than 50 million downloads (previously reported as 44 million), and together with Confluent it is expanding their deployment within the Confluent platform with a new capability called semantic intelligence, which complements the existing forecasting and anomaly detection. The models run on CPU without requiring a GPU and are accessible via SQL in Apache Flink, so they can also be used by people without a specialized data science background. The companies cite manufacturing, retail, banking, and aviation as suitable deployment areas.
The integration itself was announced earlier: the models are natively built into Confluent Cloud (the fully managed variant on AWS) as well as into Confluent Platform for on-premises and hybrid deployments. According to the companies, they enable real-time forecasts, anomaly detection, similarity search, classification, missing-data imputation, and optimization directly on streamed data, without the need to export time-series data to a separate ML platform or data warehouse.
According to the companies, there is no need to operate dedicated infrastructure for model serving, including GPUs; inference results are written to Kafka topics and are available to other systems (alerting, dashboards, lakehouse, AI agents) while maintaining the same governance, lineage, and access-control rules as for other data on the platform. The source articles mention case scenarios (e.g., controlling an ice cream manufacturing process based on SCADA telemetry); details on individual use-case scenarios and further technical specifics are not fully covered in the available text.
Why it matters
For companies running Confluent, this is an opportunity to deploy forecasting and detection models directly into the streaming layer without building their own ML team and GPU infrastructure, which the companies say shortens deployment time and reduces operating costs. For individual analysts, operations engineers, or planners, it means they can call ready-made forecasting and anomaly detection functions directly from SQL queries in Flink, without having to train or maintain a separate model for each time series.
What was added since the original report
Verified updates
-
The models have 50+ million downloads (not 44M+); the models run on CPU without a GPU; new capability: semantic intelligence (in addition to forecasting and anomaly detection); accessible via SQL and Apache Flink; suitable for manufacturing, retail, banking, and aviation
- The models have 50+ million downloads (not 44M+)
- The models run on CPU without a GPU
- New capability: semantic intelligence (in addition to forecasting and anomaly detection)
- Accessible via SQL and Apache Flink
- Suitable for manufacturing, retail, banking, and aviation
Two audiences, two different impacts
What this means
For individuals
Analysts, operations engineers, or planners can now call forecasting, anomaly detection, and semantic intelligence directly from SQL queries in Flink on streamed data, without needing to train their own model or involve a data science team.
For a business
Companies using Confluent can deploy real-time forecasting and anomaly detection without dedicated GPU infrastructure, with native inference available in both Confluent Cloud and the on-premises variant, which companies say reduces costs and speeds up adoption in operations such as manufacturing, retail, banking, and aviation.
ProcessesCheck the original
Event sources
clearly official source · 1 publisher, 0 independent. We count feeds from the same owner only once.