We’ve been sitting on our latest episode of State of Systems one for a while, we think it’s a good one and particularly resonant for right now. Over the past few months, we’ve had some fascinating conversations about where AI is inside organisations, not just where the headlines say it is. One of those was with Ian Heath at Dell, who’s right in the thick of things when it comes to infrastructure and what it takes to make AI work at scale – he knows becuase they did it themselves at Dell (he talks about it in the episode).
There’s a lot of chat about strategies, use cases, competitive advantage, productivity gains and all of it matters. But what stood out in this conversation was something much more grounded and, frankly, unavoidable: as organisations move deeper into AI, infrastructure and particularly the availability of hardware is starting to shape what’s possible, what’s delayed, and in some cases, what’s being rethought entirely.
AI adoption isn’t theoretical anymore
The debate about whether AI is relevant is long gone. Every major organisation is already doing something with it, even if it’s still early. The real question has shifted from “should we?” to “how do we actually make this work with what we’ve got?”
And that’s where the story gets interesting, because once you go beyond the proof-of-concept stage, it stops being a conversation about clever models or shiny tools and starts becoming one about pure capability and whether your infrastructure can handle the scale, performance, and responsiveness that AI demands.
Can your environment really deliver the compute you need? Is your storage fast enough? Is your data architecture ready for workloads that don’t wait around for latency? The answer, for a lot of organisations, is: not yet. And that’s before you even get to the biggest hurdle of all…the state of your data. We know it’s boring but as Ian talks about it’s probablythemost important step of all.
The data reality check
It’s easy to talk about “AI-ready data,” but most organisations are still wrangling the basics. Datasets are fragmented, inconsistent, and sitting across systems that were never designed to work together. And because AI models are only as good as what you feed them, poor data doesn’t just create bad insights it accelerates them.
That’s where a lot of the unseen effort is happening right now. Teams are reworking pipelines, cleaning, consolidating and revalidating datasets so that when AI truly scales across the organisation, it’s standing on solid foundations.
The unexpected slowdown: physical reality
For all the talk about cloud elasticity and virtual scale, the physical side of AI… the hardware itself is becoming a real constraint. As more businesses move toward production-level AI, demand for GPUs, high-performance storage, and networking gear is skyrocketing, and supply chains are feeling the pressure. We can’t say this more seriously, our partners are battling this right now, today… there are big back-logs in hardware.
What we’re seeing across the board is that lead times are stretching, certain system configurations are harder to get hold of, and some projects are quietly being redesigned around what’s available rather than what was originally planned. It’s not that the ambition has gone away, it’s that the logistics have caught up with it.
Hardware isn’t an afterthought anymore
Traditionally, infrastructure came after the project, you’d define the need, then build the environment to match. That doesn’t really work in this moment in time. Hardware timelines, availability, and component strategy are now influencing project scope, architecture decisions, and delivery dates before development even begins.
It’s forcing a new kind of planning discussion. The smart teams are thinking 12–24 months ahead, reserving infrastructure early, modelling around hardware lead times, even designing modular architectures to swap components in as availability changes. It’s not about slowing down innovation; it’s about keeping momentum when reality hits.
What this looks like on the ground
Much of the work we’re doing right now mirrors this change. It’s less about hardware quotes and more about scenario planning, helping organisations understand what’s actually on the table, how long it’ll take to get it, and what their options look like if things move faster (or slower) than expected.
Sometimes that means locking in Dell infrastructure early to secure supply. Sometimes it means adapting the rollout so projects don’t stall mid-way. Either way, it’s about recognising that infrastructure, down to the physical components has become a living part of the AI strategy.
Where this leaves us
AI will keep accelerating. Infrastructure will keep evolving to meet it. But right now, there’s a clear tension between the pace of innovation and the pace of supply. The organisations that acknowledge that early are shaping their strategies around it; those that don’t often hit the gap later when it’s far harder to fix.
Our chat with Ian is fun, hopeful and insightful;. His energy and enthusiam was palpabale we’d love to chat to him anytime.
Watch the full discussion on our you tube channel: https://www.youtube.com/watch?v=XvNbh5osf8k