Schneider Electric is developing industrial artificial intelligence agents with NVIDIA and Hewlett Packard Enterprise for deployment close to process equipment, combining local AI infrastructure with its established automation and operational technology monitoring software. The collaboration, introduced at NVIDIA GTC in Berlin on 21 October, targets plant monitoring, cybersecurity and energy optimisation.
The companies call the proposed arrangement Industrial AI Agents at the Edge. It uses HPE Private Cloud AI, co-engineered with NVIDIA, alongside Schneider Electric’s industrial software and engineering expertise. The aim is to analyse operational information and support action within the production environment while retaining more control over sensitive data and the computing resources used to process it.
Process plants operate differently from discrete robotic workcells. Chemical, paper, energy and other continuous manufacturing facilities rely on control systems that maintain equipment conditions and production parameters over extended operating periods. Adjustments to one machine can influence upstream and downstream processes, making apparently local decisions dependent on a broader understanding of the plant.
Controllers, servers, operator workstations, engineering systems, instrumentation and industrial networks generate large amounts of operational information. Some data describes the physical process, while other records identify equipment configuration, alarms, maintenance events or network activity. Interpretation requires those sources to be connected with reliable asset information and engineering context.
Schneider Electric’s first proposed group of agents concerns operational technology monitoring. These applications will combine equipment configurations, process topology, alarms, maintenance history and operating information to identify developing problems and assess their likely effect on production. The company envisages agents recommending or initiating corrective action and checking subsequent equipment behaviour.
A control system alarm may indicate a genuine fault, an expected response to changing process conditions or an issue elsewhere in the installation. A monitoring application with access to maintenance records and process relationships could provide a more informative diagnosis than an isolated alert, provided that the records are accurate and the model interprets them correctly.
The foundation for this work is Schneider Electric’s existing Operational Technology Monitoring System, or OTMS. CMPC Chile, a pulp, paper and packaging producer, has used OTMS for approximately three years to monitor the health and availability of control systems, servers, workstations and switches. The system helps maintenance and operations teams investigate problems and prioritise work.
That established deployment demonstrates the use of the monitoring system, not a completed roll-out of the newly introduced AI agents. CMPC and Schneider Electric describe future potential for faster investigation and decisions, while retaining human control of critical action. No independently verified results from the proposed agents have been published for the site.
Cybersecurity is the second application. The agents are intended to correlate network traffic, control system behaviour, asset configurations and operational events when assessing abnormal activity. In an industrial environment, an unexpected configuration change can have different implications depending on whether it results from approved maintenance, an operator intervention or unauthorised activity.
Combining security information with process context may support investigation, but detection still depends on the data available and the quality of the models used. An automated response could also have consequences if it interrupts a controller or connection essential to safe operation. The partners have not published detailed authorisation rules for taking such action, so the capability should not be described as unrestricted autonomous control.
The third group focuses on energy use. Schneider Electric plans agents that examine consumption patterns, equipment performance, production requirements and operating conditions to identify possible adjustments. Electrical and thermal loads within industrial processes are linked to throughput and product specifications, limiting the extent to which energy consumption can be optimised without affecting production.
A change to a pump speed or heating schedule, for example, may alter electricity demand while also affecting flow, temperature or process stability. A useful recommendation must account for the permitted operating envelope and be evaluated against measured results. The companies have not disclosed independently validated energy savings or productivity improvements from the new applications.
HPE Private Cloud AI provides the proposed local computing platform. Running models nearer operational systems can reduce some requirements for transferring sensitive plant data to external services, while giving the organisation greater control over infrastructure and the way workloads are operated. It does not eliminate cybersecurity exposure or data management responsibilities merely by locating equipment on site.
Schneider Electric intends to use its EcoStruxure Industrial Insights applications to deploy, manage and govern multiple agents. Common oversight is necessary where several tools consume overlapping data or propose changes affecting related equipment. Permissions, model updates, data access and the authority to act must be controlled alongside the normal industrial change management process.
HPE also announced broader additions to its AI infrastructure portfolio at GTC. Those products have different availability arrangements and should not be taken as evidence that every element of Schneider Electric’s industrial agent programme is already commercially deployed. The collaboration announcement provides no sitewide commercial launch timetable or pricing for the proposed applications.
In established process control systems, responsibility for safety and production continues to depend on suitable engineering design, validated equipment and authorised operational decisions. The planned agents add a potential analytical layer to that environment, but their wider adoption will depend on measured detection performance, the reliability of recommendations and evidence that the tools can operate without destabilising the processes they are intended to support.
