The infrastructure behind the UK’s AI ambitions

Could you introduce yourself and Park Place Technologies, and your role in supporting enterprise infrastructure globally?

I am Chris Carreiro, the Chief Technology Officer at Park Place Technologies. I've been with the company for over two decades, starting out as a technical support engineer, so I've watched both the business and the wider infrastructure landscape transform in that time.

Today, we’re a global IT infrastructure services specialist, dedicated to helping organisations monitor, maintain and optimise their technology environments - from legacy systems to next-generation AI platforms. As the principal technical leader, I oversee global Corporate Innovation, Research and Development (R&D) and the design of new portfolio offerings. During my tenure here, I am proud to have helped create our flagship predictive monitoring service, called ParkView, which was built around an AI-driven monitoring tool I designed to cut through the analysis paralysis of managing OEM estates at scale. Innovations like this have been a crucial catalyst in transitioning Park Place from a hardware maintenance provider into a comprehensive, global Data Centre Services company.

How do you see the UK’s sovereign compute ambitions translating into real-world enterprise infrastructure requirements over the next five years?

The UK government’s push to expand sovereign compute capacity by 2030 and designate data centres as Critical National Infrastructure is a major strategic milestone. However, translating this national vision into enterprise reality requires a fundamental architectural shift. For the past two decades, organisations have relied heavily on a cloud-first model dominated by non-European hyperscalers. Today, regulatory scrutiny, tightening data localisation laws and high-profile service outages mean that concentrated public cloud dependency is now seen as an operational vulnerability.

Over the next five years, sovereign compute requirements will drive UK enterprises towards deliberate, hybrid IT architectures, with a decisive shift to intelligent workload segmentation. Highly sensitive, regulated data and core AI workloads will increasingly migrate to private clouds or secure on-premises environments where data residency and access are absolute. Ultimately, genuine sovereignty rests on the resilience of the infrastructure beneath it. Organisations must map their physical dependencies and diversify their infrastructure providers to ensure they can weather geopolitical and resource volatility.

Why are so many organisations still struggling to move AI initiatives beyond the planning stage, despite strong strategic intent?

There is currently a stark disconnect between board-level AI ambition and practical infrastructure readiness. Many organisations are finding that while building a proof-of-concept is relatively straightforward, scaling those workloads to production is far more difficult. One significant compounding factor is the global memory crisis. To power AI infrastructure, manufacturers have redirected production away from standard enterprise DRAM to high-bandwidth memory. This has caused standard DRAM prices to surge by 171% year-on-year, turning hardware procurement into a moving target with lead times stretching from weeks to months.

Consequently, organisations locked into multi-year digital transformation plans built on legacy pricing and procurement assumptions are finding the hardware is either unaffordable or simply unavailable altogether. In addition, since the infrastructure requirements for production-scale AI are so complex and costly, we are seeing AI readiness slip down the priority list. Many IT leaders are opting to divert budgets to simpler technology initiatives that offer faster, more predictable commercial returns, leaving ambitious AI strategies stranded in proof-of-concept limbo.

Where is the accountability gap emerging in AI infrastructure ownership, and how should enterprises be structuring governance to resolve it?

Historically, major digital transformations like cloud migrations had clear owners. Everyone knew who signed the cheques, who managed the risk and who ran the systems. AI is fundamentally different. Because AI is beginning to reach every part of modern business, ownership has become deeply fragmented. Responsibilities are split between traditional IT, dedicated innovation teams, R&D, service operations and individual business units. The lack of clear demarcation has created a significant accountability gap that can result in critical sign-off bottlenecks for infrastructure investments. 

As a solution, enterprises must move away from ad-hoc governance. An effective model requires establishing a cross-functional steering committee that bridges the gap between commercial strategy and operational delivery. This governance body must clearly define who owns the underlying infrastructure decisions, who is responsible for data compliance and who manages the physical risks. Without this, AI initiatives will continue to stall because there is no single entity to align capital expenditure on infrastructure with the business’s long-term strategic goals.

In your view, what does “good” cross-functional ownership of AI infrastructure look like between IT, innovation and operational teams?

Successful cross-functional ownership is built on a shared understanding that AI is not purely a software or data science challenge but a physical infrastructure challenge. In a successful model, the innovation and business teams define the requirements - the business cases, performance needs and desired outcomes of AI models. As the architectural gatekeeper, IT examines whether the existing network and compute environments can support these demands without compromising day-to-day operations.

Meanwhile, operations and facilities teams must be integrated into the conversation from day one. Because AI workloads demand unprecedented power and cooling, facilities teams must assess constraints like power availability, rack space and thermal management before any software is deployed. Genuine cross-functional ownership means that no AI project moves past the ideation phase until IT, innovation and operations have co-signed a comprehensive readiness plan.

How prepared are UK enterprises for the shift towards high-density computing environments, particularly in relation to AI workloads?

Most UK enterprises are unprepared for the practical realities of high-density computing. For years, standard enterprise data centre strategies were built around predictable, low-density rack configurations cooled by traditional raised-floor air systems. AI workloads have completely upended those assumptions. Modern GPUs and high-performance AI chips operate at power densities that legacy facilities were simply never designed to accommodate.

Despite record-breaking expansions in the UK's data centre pipeline, many enterprise-grade facilities are carrying hidden risks. They are attempting to run next-generation workloads on legacy infrastructure foundations. Too often, IT leaders treat power and cooling as facilities issues rather than critical board-level technology risks. Preparedness requires an honest assessment of current limitations. Many organisations will soon realise that they cannot scale their AI ambitions within their current physical footprint without a significant, near-term overhaul of their power and cooling architectures.

Why is liquid cooling becoming a critical infrastructure consideration now, and what are the risks of delaying adoption?

Liquid cooling - whether direct-to-chip or immersion - is rapidly shifting from a niche, high-performance computing solution to a mainstream operational necessity. This matters because AI and high-density chips generate heat at a scale that traditional air-cooling systems literally cannot dissipate. Trying to air-cool these dense environments is inefficient and rapidly accelerates hardware degradation, driving energy costs to unsustainable levels.

Delaying adoption carries multiple risks. First, organisations will face operational bottlenecks. They will be forced to artificially throttle chip performance to prevent thermal overload, killing the ROI on expensive AI hardware. Second, they will hit a sustainability wall. In an era where carbon reporting, energy efficiency and environmental impact are under intense regulatory scrutiny, relying on inefficient air cooling for high-density workloads is a major liability. Delaying this transition will leave organisations holding legacy assets that are incompatible with the future of compute.

Looking ahead, what practical steps should enterprises be taking today to align infrastructure strategy with both AI ambitions and sustainability requirements?

Organisations must design for persistent uncertainty and volatility rather than assuming supply chains and energy markets will normalise. The first step is to sweat existing assets harder. By using third-party maintenance and predictive monitoring, enterprises can safely extend hardware lifecycles. This reduces exposure to increasingly volatile procurement windows for new components, cuts down on e-waste and lowers the capital expenditure required to keep standard operations running.

Additionally, IT leaders must broaden their sourcing. Rather than relying solely on OEM-direct channels, forward-thinking teams are looking at alternative sourcing, including pre-owned hardware, to keep projects moving. Finally, sustainability and infrastructure planning must be unified. When designing future data centre footprints, enterprises must plan for high-density liquid cooling from the outset and select regional sites that mitigate risks like energy constraints. Resilience is an overused word, but here it comes down to one thing: building flexible infrastructure that scales to meet computational demands while maintaining a responsible, transparent energy footprint.