AIoT

The Cloud Is Not Enough: AI Execution in the OT Layer for Industry 4.0

Inkwelldata

05 May 2026 • 4 min read

Artificial Intelligence is changing the enterprise, particularly in domains where data is centralised, such as software development, documentation, and enterprise knowledge systems.

In these environments, data is structured and easily accessible. It can be aggregated, processed asynchronously, and acted upon without immediate operational consequences. Cloud-based architectures align well with these characteristics, providing the scale and flexibility needed for model development and deployment.

Industry 4.0 systems operate on a different foundation.

They consist of distributed assets such as sensors, meters, controllers, and networks. Each asset reflects a continuously changing physical condition. The data generated represents the current operating state of the system, rather than information that can be treated purely as historical input.

In this context, insight alone is insufficient. Decisions must be applied within the same timing and operating conditions in which the data is produced. When data is transferred to a remote system for evaluation, decision logic becomes separated from the process it affects. Even small delays can introduce risk in tightly coupled systems.

AI therefore requires a distributed execution model. Cloud systems support training and coordination. Execution takes place within the environment where the system operates.

A Distributed Architecture for AIoT

AIoT systems are structured across three coordinated layers, each with a distinct role.

Cloud and IT provide model training, optimisation, and long-term analysis across fleets of assets.

Devices and control systems such as PLCs and SCADA platforms manage deterministic behaviour and enforce safety constraints.

Operational infrastructure provides the coordination layer, managing system state, processing events, and controlling how data and decisions propagate.

This intermediate layer connects data, logic, and control. It ensures that decisions can be made in context while maintaining structured interaction with enterprise systems.

Without it, architectures tend towards either isolated device logic or centralised cloud processing. Both approaches limit scalability and control in distributed environments.

A Sovereign and Governed Data Layer

The operational layer establishes where authority resides within the system.

In industrial environments, authority must remain with the infrastructure operator. This includes control over data, execution, and system evolution.

A sovereign data layer ensures that data is evaluated and acted upon at source. Decisions about whether data remains local or is shared externally are made within the operational environment, based on defined policies.

Governance is enforced at execution time. Each event is assessed in context, and the outcome determines whether it is resolved locally or propagated.

This model supports:

Regulatory alignment, including NIS2 and the EU Data Act

Operational continuity, independent of connectivity conditions

Data portability, avoiding dependency on specific vendors or platforms

As AI systems become more autonomous, maintaining this control within the operational boundary becomes essential.

Altior as Data Operating Infrastructure

Altior implements this model as a data operating infrastructure within the OT environment.

It provides a persistent execution layer between physical assets and enterprise systems. Its role is to structure how data is interpreted, how logic is applied, and how information is shared.

Core capabilities include:

Integration of heterogeneous devices and communication networks

Transformation of raw telemetry into structured, consistent data

Execution of logic in response to events

Control of data propagation beyond the operational boundary

Integration with enterprise and cloud systems through governed interfaces

This layer is independent of cloud providers, transport technologies, and application frameworks. It remains stable as upstream systems evolve, supporting long-term flexibility and avoiding vendor lock-in.

Local Execution Through Digital Twins

Execution is managed through digital twins.

Each asset is represented as a stateful runtime entity that maintains its configuration, current condition, and associated logic. This logic may include rules, state transitions, or AI inference models.

When an event is received, it is normalised and mapped to the corresponding twin. The twin evaluates the event in context, considering its current state and defined conditions.

Local resolution

If no action is required, the event is resolved locally.

Conditional execution

If conditions are met, logic is executed and actions are applied within the system.

This approach ensures that decisions are made in alignment with the current system state, without reliance on external processing.

Deterministic control systems continue to manage core automation. The digital twin layer introduces context-aware processing that operates alongside these systems.

Controlled Data Propagation

Not all data needs to be shared with external systems.

In this architecture, data propagation is selective and policy-driven. Events are transmitted only when they represent a meaningful change, exception, or outcome.

Routine telemetry remains within the operational environment.

Because routing decisions are made within the digital twin, the data that is shared is already contextualised and structured. Enterprise systems receive relevant information rather than raw data streams.

This reduces unnecessary data movement and ensures that governance is applied at the point of origin.

Network Abstraction and Security

Industrial environments include a wide range of communication technologies and protocols.

Altior abstracts this complexity through a network virtualisation layer. Connectivity is managed independently from execution logic, allowing devices to be integrated without affecting higher-level processing.

Security is embedded within the same layer.

Zero trust model

A zero trust model is applied across devices, users, and services. Authentication, authorisation, and encryption are enforced at the level of interactions. Key management remains within the operational environment, ensuring that control over data and access is retained by the infrastructure owner.

AI Execution in Context

AI models are incorporated into the system as part of the execution logic.

Models developed in cloud environments are deployed into the operational layer and triggered by real-time events. Because they operate within the digital twin, they are evaluated against the current system state.

Outputs can be applied directly within the system or shared externally according to policy.

This integrates AI into operational processes, allowing it to influence behaviour in real time while remaining under local control.

From Cloud Models to Governed Execution

Industry 4.0 systems require architectures that align decision-making with system state and control with system ownership.

This requires an infrastructure layer that executes logic where state is observed, governs how data is shared, and remains independent of evolving platforms.

The cloud remains essential for model development, coordination, and human oversight. It is not sufficient for execution in distributed operational systems.

Altior provides this capability as a sovereign data layer within the OT environment. It enables AI to be applied as a governed operational function across distributed systems, while maintaining control with the infrastructure owner.