Mark Lewis, CMO at Pulsant, explains why rising costs, AI and data sovereignty are forcing organisations to rethink cloud-first strategies and take a workload-first approach to digital infrastructure.

For more than a decade, the prevailing wisdom in enterprise IT was simple: move everything to the public cloud. Hyperscale platforms promised unlimited scalability, lower costs, agility and freedom from the burdens of managing infrastructure. Cloud-first rapidly gained momentum as the de facto path to a modern digital footprint. Until now.

Data sovereignty, AI acceleration and a more complex and uncertain world mean that something which was billed as a simpler model for workloads has become inherently complex. The net result is that we’ve effectively reached ‘peak cloud’ and, in light of this, organisations are having to review, redesign and evolve their infrastructures.

Today, rather than asking how quickly workloads can be migrated into hyperscale environments, CIOs are asking: which workloads actually belong there in the first place?

Of course, hyperscale isn’t going anywhere any time soon. Public cloud remains an essential component of modern digital infrastructure. However, businesses are now recognising that a one-size-fits-all approach no longer fits with the realities of cost, compliance, performance and AI-driven risk.

Recent research commissioned by Pulsant and conducted by Vanson Bourne reflects this shift, with 87% of organisations planning to move partially or fully away from public cloud over the next two years. Among those reconsidering their approach, 54% are evaluating hybrid models led by private cloud, 38% are increasing reliance on their own data centres, and 36% are assessing colocation as part of the mix. These findings point towards a significant recalibration of infrastructure strategies. And with good reason.

The end of cloud cost certainty

One of the original promises of hyperscale cloud was predictable, consumption-based economics. Yet many organisations have discovered that, at scale, ‘predictability’ can be anything but.

Unexpected egress charges, escalating storage costs, inter-region data transfer fees and the expense of maintaining always-on workloads have resulted in unwelcome bill shocks for enterprises across every sector. What initially appeared to be an operational cost advantage has evolved, in many cases, into a substantial and difficult-to-forecast cost burden.

This is largely, but not exclusively, owing to the training, inference and data-intensive profile of AI workloads, which demand vast quantities of compute and memory. The ongoing shortage of specialised hardware and high-performance RAM has increased infrastructure costs across the market, while competition for GPU capacity continues.

For organisations operating entirely within hyperscale environments, this can create an uncomfortable dependency on pricing structures and resource availability that they cannot directly control.

As digital demand continues to evolve, many businesses are seeking to spread risk across multiple environments, combining public cloud with colocation, private cloud, regional infrastructure and Infrastructure-as-a-Service (IaaS) platforms to achieve greater financial predictability and reduce exposure to any single infrastructure model.

An evolving risk equation

Data is among the most valuable assets a business possesses. Yet, as AI systems become embedded into workflows, organisations are rightly scrutinising where their data resides, who can access it and how it may be used.

The question facing many executives is increasingly important: could others potentially benefit from or gain insight into proprietary workloads? For organisations handling proprietary algorithms, product designs, customer interactions or sensitive operational data, greater control over how that information is accessed and used is becoming a priority.

Whether perceived or real, concerns surrounding AI training, model exposure and data leakage have heightened anxieties around hyperscale dependency. Enterprises are demanding greater transparency, control and sovereignty over their information assets.

In parallel, the regulatory environment continues to bring complexity. In fact, research from Vanson Bourne found that 79% of organisations identify sovereignty and residency considerations as a major factor influencing infrastructure investment decisions, placing data governance firmly at the centre of infrastructure strategy.

Beware data sovereignty-washing

As sovereignty concerns rise, complexity is being compounded further by so-called ‘data sovereignty-washing’.

Claims that data is sovereign simply because it is stored in-country can be misleading. Sovereignty extends beyond geography. It encompasses jurisdiction, legal access rights, operational control, governance frameworks and an understanding of how data moves across interconnected systems.

Firms must, therefore, work with partners who understand the jurisdictions, regulations and compliance mandates relevant to their sector and geography. More importantly, they need partners capable of designing digital ecosystems which support those requirements, balancing sovereignty, performance, resilience and innovation.

Hybrid: the goal, not the compromise

Against this backdrop, colocation, connectivity and collaboration are becoming increasingly important in 2026. Hybrid is, therefore, emerging as the starting point, the goal, the journey and the operating model; not simply a compromise.

The challenge for organisations is determining which workloads should run where. Infrastructure decisions must be driven by use case and desired business outcome, not by forcing business outcomes to fit a predetermined technology framework.

Certain applications will, of course, continue to benefit from hyperscale elasticity. Stable, predictable workloads may prove more economical on private cloud or colocated infrastructure. Latency-sensitive applications increasingly belong at the edge, closer to users and data sources. AI inference workloads often require regional proximity, while large-scale AI training typically needs hyperscale capacity.

But hybrid architectures only succeed when organisations can move workloads, applications and data seamlessly between environments. In this context, regional data centres are playing an increasingly strategic role.

Far from operating in isolation, modern regional facilities can provide high-performance connectivity into hyperscale platforms, SaaS ecosystems, enterprise networks and critical data sources. They can act as digital exchange points, enabling organisations to retain local control and sovereignty while preserving access to global cloud services.

This interconnected model can allow businesses to scale sustainably, reduce latency and mitigate concentration risk without sacrificing flexibility. And, crucially, it can reduce dependency on any single provider or infrastructure model, so organisations are better equipped to respond to changing commercial, regulatory or technological conditions.

Embracing a workload-first world

In summary, what we’re seeing now is a shift from peak cloud to smarter cloud. That means a cloud-centric ecosystem supported by hybrid architectures, regional infrastructure, colocation, edge computing and IaaS services, which collectively deliver greater control and resilience. For today’s enterprises, the future may no longer be cloud-first; it is increasingly workload-first.