Martin, a Munich-based company that builds waste-to-energy plants, has deployed the open-source Stackable data platform on German cloud provider Ionos – a move it said keeps its plant telemetry beyond the reach of US hyperscalers and the US Cloud Act.
The firm is a fourth-generation family business founded in 1925 by engineer Josef Martin. It designs, builds and services thermal waste treatment plants. It hands each plant to its client – often a municipality or a public-private partnership – and then supports it for years after. Around 250 people work at its Munich headquarters and more than 1,000 across the group, which has built more than 1,100 incineration lines in its century of operation.
Each line carries hundreds of sensors that track combustion, emissions, plant condition and energy recovery. Together they produce about 10GB of telemetry per line each day. Martin previously analysed that data in a decentralised way between different people, tools and manual exports.
“Data sovereignty was very important to us,” said Max Schönsteiner, head of research and development at Martin. “Not because our facilities handle highly sensitive data, but because, as part of our waste management remit, we are part of critical infrastructure and did not want to become dependent on a large hyperscaler.”
That data sovereignty concern drove the choice of Stackable, an open-source, Kubernetes-native big data platform. It packages components such as Apache Spark, Kafka, NiFi, Trino and Superset as Kubernetes operators, so customers assemble only the tools they need and can move the whole data layer between clouds or back on-premises without rebuilding it.
At Martin, the pipeline streams sensor readings from an edge device via MQTT into a NiFi broker, writes them to object storage, then runs Spark jobs scheduled by Airflow into Iceberg tables, queried through Trino and visualised in Superset and Grafana. The only managed service is the Kubernetes cluster itself, supplied by Ionos, and the infrastructure is deployed as code with OpenTofu.
Schönsteiner said the stack is entirely open source, with no licence fees and no proprietary software: “Thanks to Stackable’s open-source capabilities and its universal Kubernetes technology base, we can switch seamlessly to an on-premise solution at any time if we wish.”
That exit option was central to the decision. Ionos – a subsidiary of German telecoms group United Internet, which runs Martin’s compute in Frankfurt – is a German company under European law. Schönsteiner said the US Cloud Act effectively gave US authorities “an entry gate” into data held by US companies, and that he wanted a provider with no US parent.
Martin frames predictive maintenance as the destination but has not yet shipped a machine-learning product. “We are currently working on different AI ideas, but we do not yet have a full product,” Schönsteiner said.
The near-term work is data quality and curation, then a semantic layer so staff can “talk to the plant data” through a large language model. He said predictive maintenance on core waste processing line components – such as boilers and furnaces – is hard, because availability is the plant’s most important metric and the cost of a wrong call is high.
The immediate gain is harmonisation, he said, adding: “We now combine physical-thermodynamic models with measurement data and further operational information – or, to put it another way: experience meets technology.”
The roll-out began in January 2023 and reached full production within around two years. It is now live across Martin’s sites, with 35 lines connected so far. The company plans to equip as many of its lines as possible, then package the analytics as a service for plant operators.
