gesponsertIndustrial Edge AI Industrial Edge AI Also Needs to Understand Windows XP

Gesponsert von

IEI Integration Corp.

The Taiwanese company IEI Integration Corp. does not see Edge AI primarily as a question of computing power. What matters more is how modern edge platforms can work with legacy software while addressing data sovereignty, cybersecurity and long product lifecycles across diverse deployment environments – a conversation.

Anyone talking about Edge AI quickly ends up talking about computing power. TOPS, GPUs, CPUs, NPUs, accelerators, platforms. For many manufacturers, the message still seems to be “bigger, faster, more powerful”. So there is a lot of talk, and a lot of promises. That is understandable, but in the end, it is only half the story. In industry, the success of an Edge AI application rarely depends on raw computing performance alone. It depends far more on whether that system fits into an environment that already exists.

And much of that environment may already be decades old.

In many factories, machines have been running for 20, 30 or 40 years, especially in Europe. The systems are paid for, productive and good enough for the job they do. What is often old is not the machine itself, but the digital environment around it. That includes operating systems, interfaces, certified software and operator interfaces. Anyone who wants to introduce Edge AI must understand why operating systems such as Windows XP or Windows 7 can remain in industrial environments long after mainstream vendor support has ended.

A Central Challenge for Digitalization

At Computex 2026 in Taipei, Y.T. Lee, Vice President Networking & Edge Computing at IEI, described exactly this point as a central challenge for industrial digitalization. With nearly three decades of experience in embedded and advanced computing, IEI has developed a diverse portfolio spanning Edge AI, networking and medical solutions. Today, its offerings range from compact embedded products to mid- to high-performance platforms, with extensive customization capabilities for different application and deployment requirements. The portfolio extends beyond individual boards and box PCs to integrated platforms that combine computing performance, software and management capabilities to address industry-specific requirements.

This is also where customer expectations are changing. In many projects, it is no longer enough for an AI application to work in a pilot setup. Customers are looking for platforms that can be rolled out, maintained, updated and operated over many years. The proof of concept becomes a long-term operational challenge: How does the system remain available? How is it secured? How is it integrated into existing machines and plants? And who supports the customer when a demo setup turns into a productive installation?

Many providers are making promises in this direction. It becomes more interesting when Lee is not talking about future visions, but about the legacy systems that are part of industrial reality. In industry, legacy is not automatically a problem to be eliminated. Very often, it is exactly what earns a company its money.

Old Software Does Not Disappear Just Because New Hardware Arrives

Many industrial applications have been developed, adapted, validated and in some cases certified over many years. They run reliably, are deeply embedded in production processes or machine controls, and were not designed to move to a new platform every few years. That is where industry differs strongly from consumer products. A smartphone is replaced when it gets old. A machine is not.

Lee puts it very plainly in the conversation: “If you want to modify or redesign this kind of software, it is a big problem.” In the English original, he calls it a “big headache”. He is referring to applications that cannot simply be rewritten because costs, know-how, certifications and risks are attached to them.

For IEI, the virtualization of industrial software workloads and legacy operating systems is therefore an important building block. The idea is that new hardware does not replace old industrial software. Instead, existing workloads can continue to run on modern platforms. Lee describes it like this: A company can deploy a modern high-performance platform from Intel or AMD while continuing to run validated legacy applications within a virtualized environment. That may not sound particularly glamorous, but the point is industrial relevance.

Many companies are not building a completely new AI factory. They are looking for answers to questions such as: How do I bring more computing power into an existing installation? How do I keep an old application alive? How do I reduce the number of individual computers without rewriting the software? And how do I do that without turning maintenance into a computer science degree?

According to Lee, IEI addresses these requirements through virtualization and centralized management, enabling legacy and modern workloads to coexist on scalable edge platforms while keeping system resources visible and manageable. This is particularly interesting for small and medium-sized companies that do not have large IT departments. Edge AI should not become another specialist topic on top of everything else. It has to become an operable system.

Edge AI Is Not Just About Latency

Edge AI is often justified by short response times. That is true, but it does not go far enough. Machine vision, quality control or machine monitoring all benefit when decisions are made close to the machine. But for industry and medicine, there is another aspect: data cannot always go to the cloud.

Lee refers to medical applications where genetic data may not leave the country in certain markets. That is an extreme example, but it makes the basic conflict visible. Cloud AI is not automatically the right answer just because it is technically possible. Especially with sensitive industrial, medical or corporate data, the questions are control, data sovereignty and security. Edge systems can help because they process data locally and pass on only results, metadata or selected information.

For European companies, this point is especially relevant. Data protection, cybersecurity, intellectual property and critical infrastructure are not side issues. Anyone analyzing data must clarify where it is processed, who has access to it and how it is protected. Local edge processing therefore becomes a practical way to retain greater control over sensitive data and determine what information is transmitted beyond the site. That is not spectacular, but it is close to practice.

From Fixed Rules to Trainable Systems

For many industrial AI applications, the starting point was machine vision. Quality control is an obvious use case: point a camera at something, detect deviations, reduce scrap. But expectations are changing there as well. Lee says that customers used to rely mainly on rule-based systems. For optical inspection, features, tolerances and defect patterns had to be defined very precisely. Today, the focus is shifting toward training systems with good and bad examples. For selected applications, this can reduce the effort required to define every inspection rule manually.

One example from the conversation is a customer in Denmark that produces canned fish. In the past, Lee says, the company mainly looked for the cheapest hardware that could perform the specific task. With AI, this changes. The platform has to be more powerful and better suited to the application so that AI frameworks, models and analysis can work together.

This highlights a point that is often overlooked in AI discussions: Edge AI needs more than hardware. It needs an environment in which customers can actually get their application running. The platform concept is decisive here. For IEI, an Edge AI solution is not just a processor, an accelerator and a housing. It is a combination of industrial-grade hardware, virtualization, networking and security functions, framework support, central management and validation. Only this combination determines whether an application can still be operated in everyday use after the pilot project.

This is where providers such as IEI become interesting for industrial customers: not as AI magicians, but as interpreters between computing platforms and production reality. Depending on the application, the platform looks different. In manufacturing, the focus is on machine vision, quality control and machine monitoring. In medicine, local data processing and regulation move to the foreground. In transportation, energy supply, industrial networks or 5G infrastructures, availability, secure communication, network throughput and orchestration become more important. For customers, this means that adaptation, validation and integration become relevant alongside hardware selection.

From AI Vision to Real-Time Safety Response

At Computex 2026, IEI demonstrated how this approach can extend beyond visual recognition. In an AMR application, a TANK-XM813 edge system combined mobile cobot management with AI-based intrusion detection through IEI’s iVEC platform. Using Intel® OpenVINO™, the system analyzed camera images locally and initiated a stop response when a person entered the operating area.

The interesting point is less the individual hardware than the connection between perception and action. AI inference and mobile cobot management were consolidated on one edge platform instead of being handled by separate systems. The demonstration therefore illustrates how Edge AI can move from identifying an event to becoming part of the operational response.

This example also highlights a point that is often overlooked in AI discussions: Edge AI needs more than hardware. It needs an environment in which customers can actually get their application running. For IEI, an Edge AI solution is not just a processor, an accelerator and a housing. It combines scalable edge computing, virtualization, networking, cybersecurity, AI framework support, centralized management and system validation, with platform options that can be tailored to different deployment requirements. Only this combination determines whether an application can still be operated in everyday use after the pilot project.

This is where providers such as IEI become interesting for industrial customers: not as AI magicians, but as interpreters between computing platforms and production reality. Depending on the application, the platform looks different. In manufacturing and Edge AI deployments, the focus is on machine vision, quality inspection, AI inference and machine monitoring. In medical applications, local data processing, reliability and regulation move to the foreground. In networking, transportation, energy and 5G infrastructure, availability, secure communication, network throughput and orchestration become more important. For customers, this means that platform flexibility, customization, validation and integration become as important as hardware selection itself.

A Challenging Market

The conversation makes clear that Europe sets a high bar for deployment readiness. European customers demand stability, long-term availability, documentation, security and, increasingly, evidence of alignment with regulatory requirements. Lee mentions the Cyber Resilience Act and IEC 62443, among others. This increases the effort for manufacturers, but it can also become a competitive advantage.

Such requirements shift the benchmark. Providers that compete mainly on price come under pressure when customers ask for evidence, processes and long-term security. This is especially true in critical infrastructure, rail, marine, energy supply and industrial networks. There, it is not enough for a system to work today. It has to remain operable for years and be securely updateable.

Networking is one of IEI’s core solution areas, complementing its Edge AI and medical portfolios. Its networking and edge platforms support applications that require secure data transmission, local processing, high availability and centralized management. For Lee, Edge AI and networking are increasingly interconnected: data may need to be processed locally while systems remain securely connected, manageable and protected throughout the deployment lifecycle. Especially in Europe, he sees noticeable demand for secure network environments and local processing.

When Edge AI Has to Work Without a Large IT Team

Many companies want to digitalize, but they do not have large internal IT teams. In Germany, this does not only apply to small businesses. Many medium-sized machine builders, EMS providers and suppliers operate complex systems without having specialists for AI, IT, security and virtualization available for every edge project.

This is why Lee describes central management functions as part of IEI’s strategy. Systems should not only provide computing power, but also report their status, make resources visible and assign workloads appropriately. If Edge AI is to work on a broader scale, it must not be manageable only for companies with large IT departments.

This is exactly where many digitalization projects fail: at the transition from pilot installation to stable operation. Who maintains the system? Who updates it? Who detects failures? Who ensures that the application runs on the right hardware? IEI is trying to reduce this gap through pre-integrated systems, virtualization, central management and local support.

Supply Chain Resilience Is No Longer Just a Purchasing Issue

Industrial Edge AI depends on processors, memory, SSDs, network technology, housings, cooling and energy consumption – and therefore on the same supply chains that have been causing trouble for the electronics industry for years. Lee speaks openly about shortages in memory, SSDs, flash and now also CPUs. IEI addresses these issues through close collaboration with strategic technology suppliers, together with the ability to qualify alternative components and system configurations through its own engineering and validation capabilities when necessary.

Lee also puts the old lean logic into perspective. In the past, low inventories were considered optimal. Today, he says, companies need sufficient stock again “to get through the winter”. It is a simple image, but an effective one. For customers, the decisive point is not whether a provider can absorb every price wave. What matters is whether the provider can deliver at all, and whether changes are validated before they enter an industrial system.

Energy and Temperature Become Part of the Edge Equation

One point comes up almost in passing, but it fits well with the current industrial situation: edge systems have to become more powerful, while also dealing with higher ambient temperatures and rising energy costs. Lee says that devices used to be designed for 40 or 45 °C. Today, in some cases, manufacturers have to think more in the direction of 50 or 55 °C.

Rising ambient temperatures, more densely packed control cabinets and higher computing power are increasing the demands on cooling and energy efficiency. IEI also sees this as a reason to consolidate several functions on more powerful platforms. If five individual computers can be replaced by one system, it is not only the space requirement that can fall. Maintenance, complexity and, in the best case, energy consumption can also be reduced. This is not a general rule, but it is an important starting point for industrial edge projects.

Local Presence Still Matters

For all the technology involved, one simple point remains: European customers want contacts. Not just data sheets, delivery times or a support ticket in another time zone. Especially in more complex edge and networking projects, proximity matters. IEI therefore maintains a local office in Willich, near Düsseldorf. Originally, it was focused more on logistics, but it now also handles support, training and local communication. According to Lee, product marketing and engineering teams regularly come to Europe to support internal teams, distributors and customers. Lee also talks about plans for further European markets.

This is relevant because deploying Edge AI, networking or medical platforms involves more than selecting a piece of hardware. It requires platform selection, integration, customization, validation, lifecycle management and local technical support. For European customers, local presence can therefore be important, especially when geopolitical uncertainty, supply chain problems and security requirements already add enough complexity to the project.

Edge AI Has to Fit Into Existing Industry

In the end, the conversation with Y.T. Lee leads to a simple but important conclusion: Industrial Edge AI will not succeed because it sounds modern. It will succeed when it solves existing problems. Computing power is part of that, of course. But not only that. Edge AI has to carry legacy software forward, process data locally, meet regulatory requirements and remain operable over long lifecycles – even in warm control cabinets, with scarce components and limited IT resources.

Or, to put it a little more casually: Industrial Edge AI also needs to understand Windows XP.

That is not a step backward. It is the reality in many factories. Those who ignore it may build impressive demos. Those who take it seriously have a chance to turn Edge AI into a tool that actually arrives in everyday industrial use. (sb)

(ID:50950785)