- Initial applications target OT monitoring, cybersecurity and energy optimization.
- Local processing is intended to improve data control and reduce some AI costs.
- Chilean pulp and paper company CMPC’s three-year monitoring deployment provides a foundation for future agents.
Even as Elon Musk and investment banks such as Barclays talk up a future of humanoid robots, industrial companies are building the less glamorous foundations of physical AI: the computing, security and control systems that put intelligence to work on the factory floor.
Schneider Electric is working on those foundations with HPE and Nvidia, developing industrial AI agents to tackle three practical barriers to adoption: data sovereignty, cybersecurity and the cost of running AI.
The companies introduced Industrial AI Agents at the Edge on Oct. 21 as an initial demonstration of their collaboration. The first applications will focus on operational technology (OT) monitoring, cybersecurity and energy optimization.
“The next generation of industrial innovation will be driven by intelligence embedded directly into industrial operations,” said Gwenaelle Huet, EVP of industrial automation at Schneider Electric, in a statement. “While much of today's focus is on autonomous machines, we see a significant opportunity to apply Physical AI to the industrial systems that power the world's critical industries.”
The proposed system combines HPE Private Cloud AI, co-engineered with Nvidia, with Schneider’s industrial expertise. It is designed to run AI close to operations while allowing customers to retain control over sensitive data.
Writing more broadly about industrial AI agents in a LinkedIn article in March, Zededa VP Sachin Vasudeva described the shift toward intelligence deployed on site.
“We're moving from rigid, preprogrammed logic to dynamic, learning agents deployed directly on-site – right on a manufacturing robot, embedded in a subsea drilling sensor, or monitoring a remote pipeline,” he wrote.
Schneider says keeping and processing data close to its source can reduce the attack surface. It also says local deployment can reduce some cloud inference, data-transfer and token-related costs, while giving customers greater control over where and how AI workloads run. The announcement provides no quantified savings or comparative security results.
CMPC provides the monitoring foundation
Schneider points to CMPC Chile, a pulp and paper company, as an example of the operational foundation on which the new agents could be built. For the past three years, CMPC has used Schneider’s Operational Technology Monitoring System (OTMS) in production to monitor the health and availability of control systems across its continuous pulp and paper operations, according to Schneider.
OTMS gives CMPC’s teams a unified view of control systems, servers, workstations and switches, helping specialists identify emerging issues, diagnose problems and prioritize maintenance. Schneider describes industrial AI agents as a future addition to that deployment, with people retaining control of critical decisions.
How much control will the agents have?
FNTV has asked Schneider to clarify whether its initial industrial AI agents only recommend actions or can independently change physical operations. We will update this story with Schneider’s response.
Schneider’s announcement describes OT agents that can recommend or initiate corrective actions, cybersecurity agents that can recommend or execute responses, and energy agents that can recommend or implement adjustments. The company also says customers will retain control of critical operations, but does not specify which actions require human approval.
Ben Weaver, a research analyst at ABI Research, told FNTV that industrial AI agents will remain subject to human oversight.
“Industrial AI agents will remain primarily governed by human-in-the-loop principles for reasons varying from manufacturers' trust to regulatory causes,” he said.
“While optimizations in Advanced Process Control and PID Loops will move closer to autonomy, humans will still make final decisions in major operational changes. Deterministic safety will be maintained with rules-based restrictions, guardrails and internal policy and comprehensive digital twins for testing and verification.”
Weaver also noted that Schneider is not the only industrial automation supplier pursuing a similar physical AI stack.
“Honeywell Technologies has been pursuing the development of a similar physical AI layer with its Experion Cognition debut at Borouge International’s Ruwais facility in the United Arab Emirates,” he said.
Honeywell’s June announcement describes a live proof of concept at Ruwais.
FNTV has already reported on Schneider’s industrial software architecture and its competition with ABB and Siemens. The new agents could add another layer to that stack.
Learn more about industrial AI and tech sovereignty on Fierce Network
Schneider's PTC acquisition brings together 'one hell of an industrial stack'
Opinion: The AI factory is leaving the data center
Opinion: Can you turn it off? The new test of tech sovereignty