From cybersecurity and environmental performance to the limits of large language models and the push toward edge computing, AI is becoming foundational to how critical infrastructure is secured, maintained, and evolved.
The modern nervous system of infrastructure is stretched to its limits. As the volume and velocity of data rise exponentially, so too does the risk that its stewards lose the ability to meaningfully interpret what they are responsible for. Gerard Donohue, Chief Technology Officer at Telent, does not see artificial intelligence as a silver bullet for this problem. Instead, he regards AI as a practical, imperfect ally, one that can sift through what humans can no longer see and flag what they can no longer be expected to know.
“Cybersecurity is the clearest example,” Donohue says. “People underestimate how complex it has become. AI helps identify anomalies and trends well beyond human capability. That is going to be essential because attacks are becoming more nuanced and infrastructure more vulnerable.”
Yet this is not just about defence. Donohue sees AI as necessary in addressing the dwindling talent pool in telecoms and infrastructure. With more than 60 per cent of UK telecoms professionals now over the age of 50, operational knowledge is ageing out of the system. AI automation, especially in smart cities, offers a way to replace reactive human labour with embedded intelligence.
“You might see a future city that no longer has traffic lights,” Donohue continues. “Radar and in-car systems will direct flow based on real-time behaviours. And that is not fantasy. We are already experimenting with radar for pothole detection, AI metadata tagging in drone mast inspections, and anomaly detection across 16,000 CCTV feeds for Transport for London. The possibilities are very real, but the key point is that they arise from specific use cases. You start with the problem, not the promise.”
The appeal of AI is not rooted in novelty but necessity. With fewer people entering technical roles, a complex ageing asset base, and an ever-expanding threat surface across digital and operational technology, the traditional model of managing infrastructure through spreadsheets, site visits, and human intuition simply no longer scales. What replaces it, however, must earn its place, not just through ROI or efficiency, but through trust, transparency, and relevance to the frontline decisions engineers must make every day.
A smarter machine with a lighter footprint
As much as AI offers cognitive augmentation, it also provides physical substitution, particularly in tasks where traditional operations involve safety risks or generate excessive carbon emissions. Donohue views this as a twin imperative: to reduce the need for boots on the ground while meeting sustainability obligations set by both government and customers. “Instead of climbing a mast, we use drones,” he continues. “The data they collect, with AI-enhanced tagging, is more accurate than the human eye. We can measure the size of the pole, the banding of the radio, and more automatically. That means fewer site visits and lower emissions per job, but it also means we can deploy smaller teams to do more. It changes the whole operating model.”
The shift is as much cultural as technological. For decades, maintenance in the critical infrastructure sector has been based on regular inspections, visual checks, and physical interventions. But this model is no longer viable. By applying computer vision and metadata tagging to drone footage, organisations can not only accelerate fault detection but also pre-empt future issues based on patterns that were previously invisible to the human eye.
The environmental benefits are no less significant. Telent, like many infrastructure providers, is under increasing pressure to align with national and corporate net zero goals. Yet these ambitions cannot be met by strategy documents alone. Every unnecessary vehicle trip, repeat inspection, and diesel-powered generator adds up. “We are part of a green steering group with board-level sponsors because our customers are asking us: how will you help us reach our targets?” Donohue says. “AI is part of that answer, but so are things like electric fleets and bioethanol generators. Everything is on the table.”
Donohue believes that sustainability is not a side effect of AI but a reason to accelerate its adoption. “We can do more tasks in parallel with fewer people,” he adds. “That makes the business leaner and greener at the same time. We are not just saving money; we are saving risk and emissions. That is where the real transformation begins.”
Limits of large models and the shift to hybrid
For all the excitement surrounding generative AI, Donohue remains cautious about the real-world application of large language models in the infrastructure space. “We are not developers of AI,” he explains. “We use it pragmatically. For us, tools like Microsoft Copilot help with documentation and project acceleration. But in critical systems, compliance and control matter more than novelty.”
The regulatory landscape is particularly unforgiving. Infrastructure operators are often bound by strict certification, approval, and testing processes. A new tool cannot simply be downloaded and deployed overnight. It must first be validated against a risk model, tested against legacy systems, and proven not to jeopardise public safety or service continuity.
“AI terrifies some boards, especially in heavily regulated sectors,” Donohue says. “They cannot just plug something in and hope it works. They want assurances, validations, and test data. That takes time. In contrast, marketing departments might adopt AI because it helps shape a message more quickly. So perception depends on the vertical.”
That pragmatism also extends to infrastructure strategy. Ten years ago, the mantra across the public sector and telecoms was ‘cloud-first.’ However, the realities of hyperscaler costs, data sovereignty, and performance variability have prompted a significant reassessment of these issues. Hybrid architectures are becoming the norm, with specific AI workloads retained on-premises or moved to sovereign-certified facilities. “We are re-evaluating our own workloads,” Donohue adds. “Some of what we put in the cloud could run better on a hyper-converged stack in our own data centres. But moving data out of hyperscalers is not easy. There are real repatriation costs, time delays, and integration challenges. It is not a quick fix.”
This shift is not ideological. It is rooted in the physical realities of data transfer, performance optimisation, and compliance. Donohue points out that many workloads, particularly those involving sensitive intellectual property (IPR), government contracts, or legacy systems, do not belong in a hyperscaler. “It is not about turning the clock back. It is about using the right tool for the job,” he continues.
From re-skilling to relevance
The rise of AI raises inevitable concerns about de-skilling. But Donohue rejects the idea that AI replaces skills; it changes which ones are needed. “I trained as an engineer fixing things at the component level,” he says. “That world is gone. The new world requires re-skilling to stay relevant. Yes, AI can do many things better, but it is filling gaps that humans are no longer lining up to fill.”
The underlying problem is not automation; it is attrition. The volume of new entrants into infrastructure roles is not keeping pace with the rate of retirement. The nature of the work itself once considered high-status and high-impact, is often perceived by younger generations as having low value or being invisible. “Technology is no longer aspirational,” he explains. “It is invisible. And yet, we rely on it more than ever. People complain when the internet goes down, but they rarely consider its critical importance. AI will follow the same path. It will become expected, not exceptional.”
As AI becomes increasingly embedded in the everyday operations of infrastructure, the human workforce will need to evolve in tandem with it. That does not mean eliminating human judgment; it means redesigning workflows so that human input is focused where it adds the most value: in decision-making, exception handling, and system oversight.
Governance in a future built on trust
If AI is to be accepted as part of the infrastructure stack, it must be governed like any other mission-critical system. For Donohue, this means aligning innovation with regulation without allowing either to dominate at the expense of the other. “You cannot stop innovation, but you can shape how it is used,” he says. “Governments need to be agile. Coding is agile. Regulation should be, too. But some degree of control is required. The risks of misuse, especially by state actors, are real.”
He acknowledges the tension between regulation and agility. While poorly designed rules can stifle experimentation, a complete lack of oversight invites chaos, especially when AI systems interact with operational technology assets that were never intended to be connected, let alone intelligent. “Some of our customers take two years to approve a single service,” he notes. “That is their world. And rightly so. You cannot cut corners when you are dealing with national infrastructure. Therefore, if AI is to be effective, it must integrate into those existing approval cycles. It must earn trust.”
The destination, Donohue believes, is not a future where AI is visible but one where it is intrinsic. “It will be baked into systems by default,” Donohue concludes. It will stop being discussed separately. It will just be how things work.”
But that future cannot be assumed. It must be built, use case by use case, across hybrid architectures, legacy infrastructure, and new operating models. AI must earn its place not through branding or ideology but by solving the complex, messy problems of infrastructure at scale. It is not a revolution. It is an engineering challenge. And the organisations that approach it with clarity, restraint, and realism will be the ones that make it work.
