The hype for artificial intelligence continues in earnest, but the uncomfortable question, “Where’s the money?” is increasingly being asked in CFOs’ offices. Recent years in the AI industry have resembled a gold rush, where the mere fact of having a pick counted, not what you managed to dig up with it. According to predictions from experts at Colt Technology Services, 2026 will be a turning point. The time of costly experiments is coming to an end and the era of verification is beginning, where technology must defend itself in Excel tables.
Big language models and generative artificial intelligence have captured the imagination of business. However, this fascination is being followed by gigantic amounts of money that are not always returning to company coffers. Research cited by Colt Technology Services shows the brutal truth: although one in five large business groups spends an average of $750,000 a year on AI, as many as 95% of participants in the MIT study say they have not received a return on that investment.
This is a statistic that will no longer be tolerated in 2026. It is time to sober up and move from admiration of the possibilities to a hard accounting of the effects.
From ‘school’ to ‘work’, or time to apply
Until now, the industry’s attention – and most computing resources – has been focused on training models. This has been an energy-intensive, expensive and lengthy process, akin to sending an employee to a very expensive university. In 2026, that employee will finally start working.
Inference is the point at which the model stops learning and starts operating in a production environment – generating knowledge, predicting events and making decisions in real time.
This is not just a technical change, but first and foremost a business change. Shifting the centre of gravity from training to inference means moving from the investment phase (CAPEX) to the operational phase, which is expected to generate revenue or savings. McKinsey estimates that by 2030, inference will account for the majority of AI workloads. For CIOs, this means redesigning IT architecture to support fast, contextual decisions in the here and now, rather than just big data processing in the background.
Agentic AI: Automation that finally works
How to close the ROI gap? The answer may lie in the evolution towards so-called ‘Agentic AI’. Until now, we have been dealing with systems that can write lines or generate graphics. Now we are entering the era of agents that can do the job.
Instead of a passive assistant, companies are gaining a digital executor. According to the IEEE analysis cited by Colt, ‘Agentic AI’ will automate and digitise everyday tasks – from managing consumer privacy and health, to the complex organisation of processes within companies.
For business, this is a key difference. A chatbot answering questions is a convenience. An AI agent that autonomously schedules meetings, negotiates simple contracts or optimises the supply chain in real time – a real reduction in operational costs. In 2026, technology providers will need to offer tools to accurately measure the impact of these agents on a company’s bottom line. ROI models will become an integral part of the offering, not just an add-on to a sales presentation.
Infrastructure must keep up with ambitions
However, the implementation of AI into operational work raises a mundane but critical problem: how to transmit all this data? The forecasts are alarming. The volume of AI workloads moving across transatlantic cables, for example, could increase from 8% today to as much as 30% in 2035.
The traditional network is not ready for such a leap, especially if it is to be a cost-effective process. Therefore, 2026 will bring a redefinition of wide area networks towards AI WANs. We are talking about programmable networks specifically designed to manage traffic generated by artificial intelligence.
Why is this important for the budget? Because in the world of real-time inference, latency means loss. AI WAN is supposed to provide performance and security at the application level itself. Moreover, the environmental and cost aspects come into play. Increasing capacity by ‘brute force’ (adding more links) is no longer worthwhile. Innovations in sustainable networks that increase performance without a linear increase in energy consumption will become a purchasing priority.
The concept of NaaS 2.0 (Network as a Service) is also on the horizon. The traditional network-as-a-service model is evolving into an intelligent, automated platform. Colt’s research shows that almost 60% of CIOs are already increasing their use of NaaS in the face of pressure from AI. The new version of this service is expected to provide the flexibility needed to handle the unpredictable load spikes inherent in modern algorithms.
Data sovereignty as an insurance policy
The conversation about money in IT in 2026 cannot ignore risk. As technology matures, there is a growing awareness of the importance of data sovereignty (Sovereign AI). Countries and organisations increasingly want to build systems based on their own infrastructure and talent to become independent of global giants and align with local regulations.
This is a trend that is forcing changes in cloud strategy. Multicloud and hybrid models are becoming standard not only for technical reasons, but as a strategy to avoid dependence on a single provider (vendor lock-in) and to mitigate legal risks. Edge computing is gaining prominence, allowing data to be processed close to the source, which promotes both inference efficiency and compliance with data protection regulations.
Balance of the IT director
The year 2026 in the IT industry promises to be a time of great testing. IT executives will still have to balance on a fine line. On the one hand, the pressure for complex AI-driven digital transformation programmes. On the other – the absolute necessity to reduce costs and adapt to a changing regulatory environment.
The potential is huge and the infrastructure more powerful than ever. However, the winners in the coming year will not be those who spend the most on innovation. The winners will be those who move the fastest from the ‘wow’ phase to the ‘how much’ phase – effectively deploying AI where it brings measurable value, backing this up with a flexible, secure and cost-effective network.
The gap between investment and return will begin to close. For many companies, however, it will be a painful process of reviewing whether their digital strategy was visionary or just fashionable.
