AI infrastructure is becoming decisive for how organisations build modern IT solutions. We had a chat with Olav, CTO of nLogic Group.

In a short space of time, artificial intelligence has become one of the most talked about topics in the technology industry. At the same time, it can be hard to distinguish between hype and real needs. For many organisations, AI is not primarily about models or algorithms – but about the infrastructure that makes the technology possible.

When you talk to technologists about artificial intelligence, the conversation is often more down to earth than the headlines suggest.

Machine learning, regression analysis and pattern recognition have been used in IT systems for many years. Many of the principles behind AI infrastructure are therefore far from new.

At the same time, recent years have introduced something that genuinely stands apart from previous generations of AI. Large language models and generative systems have made the technology far more visible – and opened up entirely new areas of use.

This does not necessarily mean the technology is mature enough for every type of task. In several environments, organisations are still experimenting with how such models can be used safely in critical services.

Nevertheless, there is no doubt that the scale of development has changed dramatically.

When AI infrastructure starts to eat the network

When large models are trained, thousands of GPUs work in parallel. They must continuously exchange data with each other, and this creates enormous volumes of traffic inside data centres.

This makes the network far more important than it has been before.

Where the network could previously be a relatively stable part of the infrastructure, in these environments it can become the bottleneck itself. That is why vendors are now developing network equipment and data centre architecture with extremely high capacity.

For some environments, this is absolutely essential. When large systems work in parallel, even small improvements in performance can make a big difference.

But this type of technology has largely been developed for the very largest technology environments in the world.

Technology developed for the largest environments

Much of the infrastructure being developed now is designed for organisations building AI platforms on a scale that most organisations will never come close to.

For these environments, maximum performance is decisive. That is why heavy investment is going into technology that can reduce latency, increase capacity and handle enormous volumes of data.

At the same time, something interesting happens when such solutions are developed.

Over time, the technology also becomes available at a smaller scale. Capacity that was previously only available to the largest data centres can gradually be adopted by far more organisations.

Innovation in the largest environments therefore helps raise the level of the technology for the entire market.

When technology choices are also about economics

For organisations considering AI related investments, the question is therefore often more practical than technological:

How much capacity do we really need – and what is the right level for us?

This is not just a technological question. It is an investment decision.

Choose too little, and you can quickly run into limits in performance and scalability. Choose too much, and you can tie up significant resources in capacity that is never used.

Vendors offer solutions at very different levels – both technically and financially. Some platforms are developed for maximum performance in the most demanding environments. Others are designed for more flexible solutions that can be scaled gradually.

That is why it is important to understand what the technology is actually built for – and which needs it is meant to meet.

Making complex technology understandable

At nLogic, we work closely with vendors to understand how the technology works in practice – and what consequences different choices can have for architecture, operations and investments.

An important part of this work is filtering the information.

Vendors present the technology. Customers make the decisions. nLogic acts as the link between the two. By analysing needs, traffic patterns and future requirements, we can assess which technology choices actually deliver value – in both the short and the long term.

In practice, this is often about asking the right questions before investments are made – and making sure the solutions are suited to what they will actually be used for.

The goal is not to build the most advanced infrastructure possible, but to build solutions that are robust, scalable and properly suited to the organisation’s needs.

The technology will keep evolving

AI will undoubtedly continue to influence how data centres and networks develop. New workloads will place new demands on capacity and architecture.

At the same time, most organisations will meet this development gradually. The technology will mature, capacity will become more accessible and solutions more standardised.

But the need for good decisions will remain.

For many, it is not about choosing the “right AI strategy” – but about making the right priorities at the right time.

As investments grow and the consequences of choices become clearer, the need for sound assessments grows too.

Feel free to contact nLogic if you would like to discuss what you actually need – and how you can avoid investing too much, too little or in the wrong things.