In the age of AI, speed makes the difference, not scale

A lot of this has to do with speed. Many AI applications have a decision-making process that takes place instantly. Think of robots used on production lines or self-driving cars: they collect data, process it immediately, and base their actions on it in real time. Data processing takes place locally—at the edge—because otherwise there would be too much delay in executing the decision. Keeping this latency as low as possible is essential.

If the computing power is located centrally in the cloud, the data first has to be sent there via the network, after which the decision is returned to the application via the network. “In itself, fiber offers a quick solution to that,” says Bijnens. “But, in the age of AI, things need to move even faster. In any case, this central model has its limitations." To avoid sending data back and forth, it makes sense to keep both data and computing power at the edge.

The data explosion associated with AI further fuels this trend. “Take healthcare,” says Bijnens. “Medical monitoring generates a constant stream of large volumes of data for a patient. But do you really need to consolidate it all in a central cloud? Much of this data is only needed very briefly to make a real-time decision."

Why would you set up a massive central storage system for that and burden the network with heavy data traffic? “We need to think differently: what data do we store, where do we store it, and for how long? If we do that at the edge—as close as possible to the computing power and only for as long as necessary—we’ll be able to handle data much more efficiently than we do today."

That wouldn't be bad news for the network either. Depending on the source, data traffic is expected to increase threefold, fivefold or even more in the coming years. One reason for this is that people are increasingly discovering fixed wireless access, where they swap their landline connection for a 5G hotspot. With so much more network traffic, saturation and congestion don't seem far off.

“An additional challenge is that today's networks aren’t designed to handle the advent of AI,” explains Bijnens. “They’re designed for download traffic: for video streaming, for example. But AI processes data and provides feedback. In other words, it requires a large amount of upload capacity as well." Fiber may be the key factor here, because the network allows symmetry: both downloading and uploading. “This definitely confirms that our investment in fiber was the right choice.” At the same time, though, keeping the distance between data and computing power as short as possible—that is, at the edge—remains the best option for certain applications.

The fact that there is now a shift toward the edge should not really come as a surprise. After a period of centralization, a shift toward decentralization inevitably follows. “We see this in business organizations, in politics, and so in IT too.” Until twenty years ago, companies operated on a decentralized basis by definition, using their own servers. There was then a move toward the cloud in reaction to this. But the cloud has now become too heavy. “That leads to vulnerability. If you do everything in just one cloud, it becomes a single point of failure.”

Moreover, if that cloud is provided by a US vendor, it raises questions about sovereignty in times of geopolitical tension. That’s why those who really rely on the cloud for everything are now also exploring alternative options. “In this context, the importance of portability is increasing significantly,” Bijnens continues. “You want to be able to move your data and applications from that cloud to another platform easily.”

And once you start weighing the pros and cons of the cloud versus the edge, other considerations also come into play. “This therefore involves issues such as latency, compliance, and data location. In the future, there will be a greater need for flexibility in that area. You have to start preparing for that today." Of course, cost is also a factor in here. “In this new context—with increased data traffic and data storage—the costs of cloud services are rising rapidly. Retrieving data from the cloud—known as data repatriation—is often so expensive that it effectively leads to a lock-in."

So an alternative is welcome. “IT comes at a price,” says Bijnens. “The trick is to optimize your IT costs from an economic standpoint while still ensuring that all your needs in terms of functionality and compliance are met.” AI can act as an orchestrator in this process, automatically calculating the desired balance between every aspect. “But first, data portability has to be feasible.”

This new perspective on infrastructure is, by extension, changing our whole approach to IT. “The role of the application developers is also changing. They have to think more in conceptual terms, and then generative AI will handle the code." Once again, though, it is essential for the underlying infrastructure to be designed with this new way of working in mind—and for IT as a whole to start thinking more out of the box.

“For example, we’re increasingly moving toward a token-based IT environment,” says Bijnens. “Not only in the context of AI, as we’re already seeing today, but for the network too. But, of course, the network has to be ready for that. In short, the changes we make to the network architecture today will shape the years ahead."

The shift toward the edge clearly aligns well with concepts such as ʽNetwork as a Service' and platform-driven models. “They complement each other—it's true. Once again, the conclusion here is that the network can no longer be a dumb pipe. At the same time, it won’t be the IT professional who determines what connectivity, computing power, and storage are needed for a particular application. The expectation is that we’re moving toward a model in which AI agents drive these decisions on their own and pay for them themselves via APIs.”

This is a totally different model from that of the current network, which is essentially based on unlimited data volumes at a fixed rate. “Pricing will be handled completely differently in the future. Data volume won’t be the only factor: the desired speed and location will also come into it, for example."

Edge computing, by definition, has a significant local component. This also raises a practical question: how do you balance that local character with international data traffic? “You immediately realize here how important interoperability is,” says Bijnens. “Compare it with roaming in telecom. When you cross a national border, your cellphone automatically switches to a different network."

The same thing will happen with the edge. “A self-driving car that crosses the border will switch to another provider’s solution in a similar way. This will create an edge continuum." Several European operators are now working on this. Proximus, too, is in talks about joining a consortium like this.

In any case, the demand for edge computing poses a major challenge. “Not just from a technical standpoint—as you have to approach your network differently—but also from a business perspective. It’s really forcing us to think about this like a startup." Proximus already has a head start thanks to its experience as a historical operator. Over the past few decades, it has developed a great deal of decentralized infrastructure to manage the traditional copper network. There are multiple buildings and street boxes throughout the country—in every city and town.

“Now that we’re decommissioning the copper network, a lot of space will be freed up. We can use this in the context of edge computing, for example to provide local computing power." This gives Proximus a unique position in the IT landscape. “The hyperscalers have always focused solely on scale—and thus on centralization. But, with our physical footprint, we have a significant advantage at the edge."

Perhaps the biggest challenge is that a company, like Proximus, has to take decisions today about investments in the edge, even though most of the applications that will use it do not yet exist. “We don’t yet know what the killer app of the future will be,” concludes Bijnens. But there are suspicions, of course. “Personally, I think it’s robotics. To provide computing power for robots you need chips and batteries, which makes them expensive."

For this reason, robots may soon look completely different. If you remove the chips and batteries, you are left with a dumb—but affordable—robot. It gets everything it needs, such as computing power, from the edge. “Electric car manufacturers have a lot of excess capacity. If they use it to produce low-cost house, garden, and kitchen robots, that could lead to a massive acceleration in innovation—with the edge playing a key role."

Cloud expert Thierry Van Nuffelen breaks down what matters most.