As the wide wave of enthusiasm that dominated the artificial intelligence sector in 2025 comes to an end, companies today find themselves facing a more difficult question: where is the real value?

After a year filled with pilot projects, scattered chatbots, and massive investments that did not always yield a clear return, 2026 looks to be a completely different phase. Companies are no longer just expected to prove that they use artificial intelligence, but to demonstrate how this usage contributes to reducing costs, improving efficiency, increasing revenues, and protecting data.

In an extensive interview, Anthony Dornan, senior managing director in the technology, media, and telecommunications sector at FTI Consulting, said that 2025 was the year of experiments, while 2026 will be the year of hard questions about value and return on investment.

Dornan said: “I think 2025 was a year for pilot projects, as every organization raced to prove its ability to execute in the field of artificial intelligence. Now, however, 2026 has become the year where questions surrounding value and return on investment take top priority for organizations.”

This new phase does not mean a decline in the importance of artificial intelligence; rather, it means companies are beginning to shift from rapid enthusiasm to strict management. The technology is no longer just a tool to appear cutting-edge, but has become a strategic decision that requires a clear data architecture, strong governance, guaranteed computing power, and measurable performance indicators.

Sovereign AI changes the rules of the game

One of the most prominent transformations facing companies in the Middle East today is the rise of the concept of sovereign AI. With governments around the world, from Europe to the Gulf states, tightening data protection and governance laws, relying entirely on centralized global cloud services as before is no longer easy.

In the Gulf specifically, investments in technical infrastructure within the UAE, Saudi Arabia, and Qatar have accelerated in an effort to ensure sensitive data remains within national borders, especially with the increasing reliance on artificial intelligence models in sensitive financial, governmental, healthcare, and operational sectors.

Dornan believes that data sovereignty has become one of the most important files for the upcoming phase. Multinational companies operating in the region are no longer able to treat local data as if it were part of a single, open global system. Today, they must build what he described as an integrated and distributed data architecture that allows models to benefit from data without breaking local rules.

To put it more clearly, the solution will not always be moving data to the model. In many cases, the model itself must move to where the data resides.

This forces companies to redesign their systems from scratch. Compliance is no longer a subsequent step, but must be part of the core design, especially when it comes to anonymization, using synthetic data, or minimizing sensitive personal information before running it inside models.

Chip and computing crises pressure companies

However, building local or sovereign artificial intelligence does not stop at laws and data alone. There is a clear physical obstacle: computing power.

With rising global demand for graphics processing units and data centers, companies are facing a real shortage of the hardware, energy, and infrastructure needed to run advanced AI models. According to Dornan, the pressures in 2026 will be greater than they were in 2025, especially with major cloud service providers monopolizing a huge share of available resources.

Therefore, Dornan urges companies to treat computing power like any other important strategic resource. This means they must secure long-term contracts, diversify suppliers, and avoid full reliance on a single entity.

Yet the solution does not lie solely in buying more computing capacity. Many companies, according to Dornan, use expensive resources for tasks that do not actually require that level of power.

He said: >>Not all work performed by companies requires high-performance computing or ultra-advanced graphics processing units<<.

This is where smart workload management comes in. Training large models requires advanced chips, but many daily operational or inference tasks do not need the same capabilities. Therefore, companies can significantly reduce costs if they distribute their technical tasks more efficiently.

The end of the era of isolated experiments

Over the past two years, many companies rushed to launch AI tools, especially chatbots, without having a clear plan for what they wanted to achieve. The result was many projects that looked modern on the outside, but changed nothing fundamentally inside the business.

Dornan believes this problem stems from the absence of the core question: what business problem do we want to solve?

If a company launches a customer service chatbot, but the internal system does not actually allow customer problems to be solved, then artificial intelligence becomes an extra layer of cost rather than a tool to improve performance.

Dornan stated that companies should not start with the technology, but with the business need. Is the goal to reduce costs? Increase sales? Accelerate operations? Improve the customer experience? Once the goal is defined, artificial intelligence can be a helpful tool, not an end in itself.

Therefore, Dornan advises companies to adopt what he calls a dual-track mindset. In the first phase, companies can experiment quickly and test ideas. But when moving to scaling, every project must be tied to clear performance indicators and a measurable financial return.

The question is no longer: do we have an artificial intelligence project?

The real question has become: where will this project create value within the company?

Agentic AI raises the risk level

As artificial intelligence transitions from content generation tools to agentic systems capable of executing tasks and making certain decisions, risks clearly escalate.

Agentic AI can send messages, analyze data, execute procedures, and perhaps make real-time decisions. This grants companies significant operational capacity, but it also opens the door to costly mistakes if clear governance is lacking.

Dornan stresses here the need to adopt the principle of graded autonomy. Not every decision can be left to artificial intelligence to the same degree.

Low-risk tasks, such as drafting internal messages or executing routine operations, may operate at a higher level of autonomy. As for sensitive decisions, such as pricing, commercial offers, financial approvals, or communicating with customers on complex matters, they must remain under direct human supervision.

He also emphasizes the need to test these systems within isolated environments before deploying them in live operations. Every action taken by the agent must be traceable and reviewable, with a clear log showing what happened, when it happened, and who approved or rejected it.

In other words, companies do not just need stronger artificial intelligence; they need a stronger control system.

Walled gardens are here to stay

In light of these risks, many companies in the region are turning toward what are known as walled gardens—more controlled digital environments that prioritize governance and data protection.

This trend does not look temporary. While open systems may become suitable for low-risk innovation, regulated sectors such as banking, energy, healthcare, and government will likely need more closed and controlled environments for a long time.

The reason is simple: the cost of error has become high. Hallucinations, data leaks, the inability to explain decisions, or relying on general models for sensitive files are all factors that make major institutions much more cautious.

Dornan believes this approach will continue in the near future, especially in the region, where national data protection intersects with considerations of competition, security, and regulatory compliance.

From enthusiasm to maturity

What is happening today does not mean artificial intelligence has lost its luster. On the contrary, this phase may be proof of the sector’s maturity.

After a period of rapid enthusiasm, companies are now entering a more realistic phase—a phase where they ask about cost, return, risk, sovereignty, governance, and the ability to scale.

Ultimately, the winning companies will not be those that launch the most pilot projects, but those that know where they use artificial intelligence, why they use it, and how they measure its impact.

Dornan concluded by saying that there is a broad realization this technology is capable of bringing about a true transformation within companies, but the biggest challenge lies in applying it in a way that ensures cost management, data protection, and improved return on investment.

In 2026, the question no longer seems to be: who uses artificial intelligence?

The more important question has become: who uses it smartly?