A 2025 report on the ‘State of AI’ by McKinsey looked at how organizations are “rewiring” themselves for AI and, among various useful insights, it highlights the pace at which organizations are changing.

As recently as 2023, the emphasis was on experimenting with generative AI through small pilots and employee productivity tools. Fast forward to today, and attention has turned to embedding AI in production environments, with one of the major talking points being the associated infrastructure constraints.

As far as Managed Service Providers (MSPs), Value-added Resellers (VARs), and other channel providers are concerned, customers are no longer simply asking for infrastructure capacity; they also want guidance on how best to support increasingly demanding, real-world AI workloads. Compared to conventional enterprise systems, these applications have very different infrastructure requirements, with GPU resources and cloud costs quickly becoming much more significant considerations.

What has changed?

There can be no doubt that many organizations and indeed their channel partners initially assumed that AI initiatives could be accommodated within existing cloud strategies. Looking back, that’s completely understandable, but while it may have been sufficient during the experimentation phase, production AI is a different beast.

AI projects also move at an unprecedented pace. Development teams want to experiment quickly, train models, test new capabilities, and deploy successful applications without waiting weeks or months for additional infrastructure to be provisioned. Add to that growing concerns over AI token spend as organizations scale production deployments, and infrastructure decisions must urgently balance harsh economic realities with technical capability.

Embedding AI into business-critical processes also brings new questions about data sovereignty, particularly around where information is processed and who controls the supporting infrastructure. The stakes are also higher. AI applications are increasingly handling business-critical data, meaning organizations need confidence that they retain appropriate control over their data and the infrastructure that supports it.

Collectively, these factors are encouraging organizations to reassess long-held assumptions about infrastructure. For instance, is it still the case that public cloud is the default answer for every workload or that compute resources are effectively unlimited and readily available? Is infrastructure cost relatively predictable as applications scale? The list goes on.

Staying focused

If the answer to any of these questions is no, some strategic adjustments may be required to ensure AI projects remain properly specified. Let’s be clear, this isn’t necessarily about wholesale change or yet more transformation. Take hyperscale cloud providers, for example, which remain central to enterprise IT because they enable organizations to deploy applications on demand and at scale without significant upfront infrastructure investment.

But it’s also clear that AI workloads can become expensive to run in public cloud environments, particularly where resources are required over sustained periods rather than for short-term or intermittent demand. This situation is unlikely to change soon and may become even more pronounced as agentic AI is integrated.

The answer is to look more closely at where AI workloads are running and what is driving infrastructure costs. In this context, visibility is increasingly important, particularly as environments inevitably become more complex. Organizations want guidance around whether their applications are best suited to hyperscale cloud platforms or would benefit from specialist infrastructure that is better aligned with their specific requirements. The mindset is somewhat different in that it is workload-led rather than platform-led.

Dealing with uncertainty

This is precisely the kind of scenario where channel partners have always added the most value. Customers increasingly want to understand the commercial implications of different infrastructure choices, not just their technical characteristics. They also expect clear, objective recommendations that reflect their business priorities rather than a preference for a particular platform or vendor.

So where does that leave us as channel stakeholders? One of the biggest challenges facing customers is that AI infrastructure decisions are being made against a backdrop of exceptional technological change. Assumptions that appear reasonable today may need to be revisited within months.

Let’s be honest, few organizations can be completely certain how AI adoption will evolve over the next year, let alone the next two or three. Infrastructure strategies therefore need to accommodate uncertainty rather than optimize for a single expected outcome. This explains why the industry is seeing much more made of the advantages that infrastructure flexibility offers. Customers increasingly make decisions based on how well they preserve future options rather than locking them into a particular technology or commercial model.

This changes the nature of conversations between customers and channel partners. Rather than recommending a destination architecture, partners increasingly help customers make decisions that remain resilient as AI takes its next exciting turn. Getting this right will increasingly influence positioning and competitive advantage.