Why architecture is becoming a competitive advantage
Manufacturing organizations have spent years digitizing individual parts of their operations. Production systems became connected, enterprise platforms became more sophisticated and cloud infrastructure expanded access to data and analytics. Yet many companies are now discovering that the next phase of transformation is not about deploying another technology layer. It is about how all those layers work together.
As operational data increasingly moves between factory-floor systems, edge infrastructure, enterprise applications and cloud environments, architecture has become a strategic issue rather than a purely technical one. Decisions about where data lives, where it is processed and how it is governed are beginning to influence everything from AI adoption and operational resilience to cybersecurity and production continuity. What was once viewed as an IT concern is rapidly becoming a question of industrial competitiveness.
Data without context does not scale
Many digital transformation initiatives begin with the assumption that more connectivity will naturally create better visibility and better decisions. In practice, manufacturers often discover that moving data is easier than creating value from it. Years of investment in digital technologies have created vast quantities of operational information, but that information frequently remains fragmented across different systems, sites and business functions.
“Manufacturers have embraced digital transformation in recent years with investments in core technologies and data analytics,” Ted Combs, Industry Principal, Consumer Products at AVEVA says. “But piecemeal adoption means that factories often grapple with siloed data. Created by different systems and shaped by local definitions, industrial data is rarely consistent across sites.”
The consequences are familiar across the sector. Different facilities develop their own naming conventions, integration methods and data structures. Over time, organizations accumulate layers of workarounds, translators and specialist expertise simply to keep information flowing between systems. The technical debt grows while the ability to scale new initiatives becomes increasingly constrained.
Magnus McCune, CTO at HiveMQ, believes many organizations still approach the challenge from the wrong direction. “The most expensive mistake is treating this as a data-movement problem when it is a data-meaning problem,” he says. “Teams pick a destination, usually a cloud data lake, and optimize everything for getting raw data there. Then they discover that something like 80 percent of the effort goes into wrangling and cleaning data after the fact, and only 20 percent into actually using it.”
The weakness in that approach becomes visible when companies attempt to replicate successful projects elsewhere. Rather than building on previous progress, they often find themselves repeating the same integration exercise because every site has evolved differently.
“The tell is what happens at the second site,” McCune continues. “I have watched a manufacturer spend the better part of a year getting one plant working, then start the next plant and find the data structure is completely different, so the integration starts again from zero. Nothing compounds. Every site is a fresh project.”
Moving information into a central repository is only part of the challenge. Data that lacks context, consistency and clear meaning simply creates larger problems at greater scale. A connected digital backbone helps establish a common operational foundation across sites, systems and functions, making it easier to standardise information, share insight and support new technologies. Without that foundation, analytics, AI and automation initiatives often struggle to move beyond isolated deployments.
The edge versus cloud debate misses the point
Much of the discussion around modern industrial architecture focuses on edge versus cloud, as if organizations must choose between them. Most manufacturers are discovering that both environments are essential. The challenge is understanding which decisions belong where.
For Ed Hutchinson, Principal Ecosystem and Development Manager for the Red Hat Industrial Business, operational autonomy remains non-negotiable. Manufacturing facilities must continue operating even when connectivity is interrupted, particularly in environments where production continuity is critical.
“Most facilities need to be autonomous so that when networks go down, they are still able to produce,” he says. “It is fairly easy to tie these facilities into cloud and on-premise infrastructures, however, as workloads become more advanced there becomes more gravity to the data. The larger the data feeds and repositories get, the harder it is to transport that data for remote computation.”
That challenge is becoming increasingly important as AI workloads consume larger datasets and organizations seek real-time operational responsiveness. The question is no longer where data should be stored, but where decisions should be made.
McCune argues that latency and operational consequence provide the most useful framework. “If a decision closes a control loop or carries a safety or production-continuity consequence, it belongs at the edge, because the round-trip to the cloud is longer than the decision window,” he explains. “If a decision needs cross-site context, large-scale history, or heavy compute, it belongs in the cloud.” His summary is simple but powerful: train in the cloud, act at the edge.
Stephan Pottel, Manufacturing Strategy Director EMEA at Zebra Technologies, sees the same shift occurring through the growing use of AI at the frontline. Organizations are generating enormous volumes of information from devices, inventory, machines and workflows, creating opportunities for faster operational decisions.
“We know a huge volume of data is generated at the edge of organisations from devices, machines, inventory and workflows,” he says. “Some of the processes around quality, compliance and worker safety require high levels of accuracy and immediate responsiveness.”
This is driving increased adoption of AI capabilities directly on devices and operational equipment. Rather than sending every interaction to the cloud, organizations are increasingly deploying AI inference locally where decisions need to be made. “We are seeing this important strategic shift towards on-device AI models,” Pottel says. “With latency removed, things like risks, defects and anomalies are picked up and acted on there and then, downtime is avoided and throughput is protected.”
What emerges is not a battle between edge and cloud but a recognition that each serves a different purpose. The most effective architectures increasingly treat them as complementary parts of the same environment, placing workloads where they create the greatest operational value.
Convergence is becoming governance
For years, discussions around IT and operational technology convergence focused heavily on organizational culture. The assumption was that the primary challenge involved persuading different teams to work together. While those issues remain important, the nature of convergence is changing.
“The number one challenge for any organization looking to incorporate disruptive technology is addressing the organizational barrier itself,” Hutchinson adds. “Although some companies have made efforts to align different groups between IT and OT with a common goal of supporting production, many still keep these teams separate.”
At the same time, the practical mechanics of convergence have evolved. The technology required to move information between environments is increasingly mature. The more difficult questions now revolve around governance, ownership and accountability. “For years convergence was framed as an org-chart and cultural problem,” McCune continues. “The practical reality is that convergence is happening at the data layer first, and the org chart is following, not leading.”
The organizations making the strongest progress are creating shared, governed environments that both IT and OT teams trust. Operational systems contribute real-time information while enterprise systems consume structured and contextualized data. The focus shifts away from moving information and toward agreeing what that information means.
Traditional industrial architectures were built around clearly defined operational boundaries. As manufacturers connect factory-floor systems with cloud platforms and enterprise applications, those boundaries become much harder to maintain. “The classic layered model assumed data mostly stayed within its layer,” McCune adds. “Once operational data is genuinely flowing edge-to-cloud, you need a deliberate, governed crossing point rather than a hundred ad-hoc ones.”
Hutchinson believes security must be embedded into architectural decisions from the outset rather than addressed later. “For any converged IT/OT modernization effort, it is imperative to bring the teams focused on security into the planning sessions to guarantee that all sides can help architect the best path forward around data and execution.”
What emerges is a broader definition of convergence. Success is no longer measured solely by connectivity. It increasingly depends on whether organizations can establish common standards, governance models and security practices across highly distributed environments.
Building for AI not dashboards
AI is changing the purpose of industrial architecture. Visibility remains important, but organizations are increasingly building environments designed to support real-time decision-making, automation and AI-driven workflows rather than simply delivering information to human users. “Five years ago, industrial architecture was optimized to move and store data so a human could look at it later,” McCune says. “The implicit consumer was a person reading a screen. AI changes the consumer.”
As organizations deploy AI models, digital assistants and increasingly autonomous systems, architecture requirements become far more demanding. Context, governance and trust move from desirable features to essential capabilities. “A dashboard tolerates a cryptic tag name because a human knows what it means,” McCune explains. “A model does not. It needs to know what the tag is, which asset it belongs to and which process it serves.”
This helps explain why many organizations are investing heavily in data foundations before scaling AI initiatives. Combs describes a progression that begins with trusted visibility, advances through machine learning and deterministic AI models, and eventually supports generative and agentic AI. Each stage depends on the same prerequisite: reliable, consistent and trusted operational information.
Pottel sees a similar evolution unfolding closer to the frontline. He describes the emergence of ambient intelligence, where sensing technologies, software platforms, edge processing and AI combine to create a real-time operational picture that continuously reflects conditions across the factory. “Ambient intelligence brings together the physical layer, the data layer, an AI analysis layer and the execution layer with an appropriate level of automation,” he says. “It is a living system or reflection of what is happening in that moment.”
The ambition extends beyond visibility. Increasingly, organizations are building environments where human workers and AI agents operate together, sharing information and supporting decisions in real time. “This combination of human workers and agents could be described as augmented collective intelligence,” says Pottel. “It supercharges the frontline worker and delivers autonomous operations which increase productivity and margins.”
For all the attention being paid to AI, cloud computing and automation, the underlying challenge remains surprisingly consistent. Organizations need architectures capable of moving information, context and decisions across increasingly complex operations without sacrificing resilience, security or operational control.
The manufacturers making the strongest progress appear to share a common philosophy. They view architecture not as a collection of technologies but as the foundation on which future capabilities will depend. In that environment, data becomes a compounding asset, AI becomes easier to scale, and every new site, application and operational system strengthens the wider business rather than adding another layer of complexity.
