Why Operational Data and Security Are the Real Bottlenecks for Industrial Automation

That is the useful counterweight to the annual parade of “technology trends”: the constraint is not a shortage of AI, robots, or vendor slides. It is the unglamorous state of operational data and exposed OT networks.

For enterprise IT teams, the 2026 trend list is less a procurement menu than an attack-surface map. Physical AI may be moving toward factory floors, but a robot trained by watching an operator still depends on systems that can be trusted to deliver clean data and resist interference.

The automation gap is also an infrastructure gap

MarketScale identifies physical AI as a growing focus in manufacturing, alongside an 80% automation gap. The phrase sounds like an opportunity. In practice, it describes a large population of sites that have not deployed even the previous generation of automation.

Operations leaders often point to capital constraints. MarketScale’s reporting instead identifies data hygiene and OT cybersecurity as the central bottlenecks for AI and robotics adoption. That distinction matters. Buying a cobot is a capital decision. Giving it access to production data, industrial controllers, and connected workflows is a security decision that persists long after the purchase order closes.

A facility with poorly structured operational data cannot reliably run modern automation. A facility with weakly secured OT cannot safely expand connectivity around machinery. Neither defect is fixed by attaching “AI” to a project name.

Physical AI lowers one barrier, not the others

MarketScale describes physical AI systems that learn tasks by observing a human operator rather than relying on conventional code or teach-pendant programming. That could make robotics more accessible for work that varies too much for rigid automation routines.

It also changes the deployment model. More people may be able to configure a system through demonstration rather than specialized engineering. The technical barrier falls. The governance burden does not.

Every new path from human behavior to machine action deserves scrutiny: where training data is stored, who can alter it, how changes are logged, and whether access to the system can become a route for lateral movement into the OT environment. A simpler interface is not a smaller attack vector. It is often the opposite.

End-of-line deployments deserve the first hard questions

Packing, palletizing, labeling, and quality checks are emerging as near-term automation targets, according to MarketScale. Nearly half of manufacturers cited in its coverage plan to implement end-of-line automation within 24 months. These are attractive projects because they can avoid a deeper reconfiguration of the production line and may offer faster returns.

That convenience can invite negligence. End-of-line systems still touch production data, operator workflows, networked equipment, and often third-party support channels. Flexible cells add another variable: MarketScale notes integrated motion systems designed to let one cobot cover up to 10 meters of horizontal range. Fewer units may improve the floor-space equation. They also concentrate operational dependency in a smaller number of systems.

The practical sequence is bleak but clear. Inventory the OT environment. Clean and structure the data. Define access boundaries. Test what happens when the automation system loses connectivity or receives bad inputs. Then consider the robot.

The trend for 2026 is not simply physical AI. It is the collision between ambitious automation plans and the infrastructure companies neglected while waiting for the next shiny machine.