AI bubble: revolution, or just an expensive illusion?

News – 05.05.2026

The four largest American tech companies plan to spend approximately $650 billion on AI infrastructure in 2026. At the same time, most businesses still cannot demonstrate that AI is actually making them money. While technology companies are investing hundreds of billions of dollars in infrastructure, models and data centres, the term “AI bubble” is appearing more and more. Not because the technology doesn’t work – but because expectations are starting to collide with reality.

At first glance, everything looks like a success. The market is growing, investments are increasing and AI startups are emerging almost every week. But when we look closer, a less visible mechanism reveals itself, the so-called “circular financing”, which is driving this growth. Large tech companies invest in AI startups, which then use these funds to purchase infrastructure, often from those very same investors. The money circulates within the system, revenues grow, but the actual end market does not grow at the same pace.

An important difference compared to previous technological waves lies in the very nature of AI. It is not classic software that, once developed, generates near-zero running costs. AI is, on the contrary, extremely expensive to operate. Every query, every response means consumption of computing power, energy and infrastructure. The more companies use it, the more it costs them. This fundamentally changes the economics of the entire sector.

It is therefore no surprise that spending on data centres is growing at a record pace and tech companies are beginning to grapple not only with a shortage of chips or memory, but also with the very availability of energy. In many respects, AI is thus starting to resemble the infrastructure or energy sector more than traditional software.

When we look at real-world deployment within companies, the gap between expectations and results is even more pronounced. Most organizations are actively testing AI today, but only a small fraction has managed to integrate it into key processes in a way that has a measurable business impact. Many projects stall at the experimentation phase, because companies underestimate what matters most – working with data, changing processes and managing the entire implementation.

We see a similar picture in the Czech market. Companies have a strong interest in AI and are planning new projects, most commonly in the areas of internal process automation, marketing or document handling. At the same time, however, they run into the same limitations as elsewhere: a shortage of people, complex regulation and slow decision-making. A positive shift is at least the growing adoption of cloud, without which more advanced AI solutions are practically unworkable.

The fundamental question is therefore no longer whether AI will change the way companies operate, but how quickly the market will adjust to the reality of costs and returns. It therefore does not appear that the market is heading for a dramatic “bubble burst”, but rather a gradual sobering up. Exaggerated expectations will disappear, investments in universal solutions will decrease and more space will be given to specific, specialized applications that make economic sense.

Today it is no longer enough to simply purchase a basic license and assign one or two people on the team the task of “testing.” What is key is knowing what AI can be used for and what its real impact is. Successful organizations have chosen a specific problem, set benchmarks and are gradually scaling what works.

From our own practice we see that the question of whether AI makes sense at all is already outdated – it has now been replaced by specific questions about specific use cases.

If you are figuring out how to genuinely implement AI in your company – not just try it and forget about it but truly embed it into your processes – write to Martin Valášek. We share specific experience from projects we have gone through ourselves.

Sources: