Published on 13 October 2026
Over the last decade, the evolution of technology infrastructure seemed to be moving in a fairly clear direction: the traditional datacenter was gradually losing prominence while the public cloud continued to gain ground, driven by advantages that are difficult to replicate in terms of scale, elasticity, speed of innovation, and investment capacity.
The result has been a significant concentration of the market around the major hyperscalers. And rather than changing this dynamic, the arrival of Artificial Intelligence is accelerating it.
In the second quarter of 2026, AWS, Microsoft, and Google jointly accounted for 63% of the global cloud infrastructure market. The market was also growing at 43% year over year, its fastest pace in eight years, with generative AI being the primary driver of that acceleration, according to Synergy Research Group.
AI requires enormous computing, storage, connectivity, and specialized accelerator capabilities. And very few companies can invest at the pace needed to build them.
As a result, hyperscalers will continue to play a critical role.
At the same time, however, another force is emerging and reshaping the market: sovereignty.
From Choosing a Cloud Model to Choosing a Level of Sovereignty
In our conversations with clients, we are seeing how sovereignty is no longer an issue limited to public administrations, defense organizations, or certain highly regulated industries.
Questions are increasingly arising about where data resides, who has access to it, who controls the encryption keys, who operates the platform, which jurisdiction governs the service, and how easily a provider can be replaced.
This does not mean organizations want to abandon the public cloud. Quite the opposite: they want to retain its capabilities while taking a much closer look at how much control they are willing to relinquish in exchange.
Europe is accelerating this conversation.
The European Commission has created a Cloud Sovereignty Framework that evaluates sovereignty through 48 criteria grouped into eight dimensions, covering legal, technological, operational, supply chain, security, data, and AI-related aspects.
Even more significantly, the proposed European Cloud and AI Development Act introduces four different levels of sovereignty depending on requirements and risk profiles.
This concept is important because it breaks away from an overly simplistic view.
The debate is no longer about choosing between public cloud and sovereign cloud.
It is about determining the level of sovereignty required for each workload.
For some applications, a global hyperscaler will likely remain the best option. Others may require guarantees around data residency or key management. And particularly critical workloads may demand higher levels of technological, operational, or jurisdictional autonomy.
Hyperscalers themselves are adapting their offerings to respond to this new demand. Alongside them, European providers and a new generation of specialized operators and neoclouds are gaining momentum.
We are therefore unlikely to see one model replacing another.
Instead, we are moving toward a far more heterogeneous ecosystem.
AI Is Also Transforming the Datacenter
This evolution coincides with a profound transformation of infrastructure itself.
The relative weight of the traditional enterprise datacenter continues to decline. According to Synergy Research Group, on-premises facilities represented 56% of global datacenter capacity in 2018. By the end of 2025, that figure had fallen to just 32%, and it could drop to approximately 19% by 2031.
Meanwhile, hyperscalers are expected to increase their share of global capacity from 48% today to 67% over the same period.
But it would be a mistake to interpret these figures as the end of the datacenter.
In fact, the opposite is happening.
Artificial Intelligence is generating enormous new demand for infrastructure. McKinsey estimates that global datacenter capacity requirements could grow from approximately 82 GW in 2025 to around 220 GW by 2030, driven largely by new AI-related workloads.
What is gradually disappearing is not the datacenter itself, but a particular way of thinking about it.
We are moving from infrastructures designed primarily to host corporate applications toward intensive computing platforms where GPUs, energy, cooling, data, and distributed architectures play an increasingly central role.
And this new infrastructure is once again placing questions such as capacity, location, technological dependency, and control at the heart of strategic decision-making.
The More Intelligence We Add, the More Control We Need
AI introduces a second major transformation.
Until now, an architecture could largely be analyzed through its applications, servers, networks, storage systems, data assets, and cloud providers.
In modern architectures, we must also account for models, GPUs, inference platforms, vector databases, APIs, agents, and services that may be distributed across multiple providers.
The dependency chain is growing.
And with it come new questions.
Where are our models executed? Who controls the data they use? Can we change models or providers? What happens to prompts, embeddings, or logs? Which external services does an agent consume? Who can observe and govern its decisions?
The evolution toward increasingly autonomous systems makes these questions even more important.
Because the more responsibility we delegate to technology to interpret information, execute processes, and make decisions, the greater our ability must be to observe it, govern it, and maintain control over it.
This is why AI does not reduce the need for sovereignty.
It multiplies it.
And it is also changing the way we will need to manage infrastructure.
The challenge will no longer be simply operating each technology correctly in isolation. It will be about maintaining visibility and control across a distributed ecosystem composed of traditional infrastructure, multiple clouds, sovereign platforms, data, models, and agents.
This will require observability, automation, security, and AIOps capabilities to evolve toward models capable of understanding and governing far more dynamic architectures.
It also highlights another dimension that is often overlooked: operational sovereignty.
We may store data in Europe, control our encryption keys, and use a properly isolated infrastructure. But if we remain entirely dependent on a third party to operate, recover, or evolve that infrastructure, our actual decision-making power remains limited.
Sovereignty, therefore, does not mean eliminating every technological dependency. In a global ecosystem, that would probably be impossible and would not necessarily make sense.
It means understanding those dependencies, deciding which ones we are willing to accept, and maintaining control where it truly matters.
For years, we have designed architectures by balancing cost, security, availability, performance, and innovation.
Now we must add a new variable to that equation.
Sovereignty.
And the rise of Artificial Intelligence will make it even more relevant.
This does not appear to be a passing trend. All signs suggest that it is becoming a new foundational design principle for technology.
