Data centers face rising demands from AI, but innovation remains slow despite clear benefits. In this thought leadership article, Mark Venables explores how digital twins, liquid cooling, and simulation can help break the cycle of hesitation and drive sustainable transformation.
Data centres can be surprisingly ambiguous places for an industry that thrives on precision. Behind the reassuring hum of HVAC systems and rows of high-density servers, many facilities still rely on scattered spreadsheets, ageing infrastructure, and cautious habits that stifle innovation. This conservatism is inefficient and unsustainable in an era defined by exponential data growth and the rising pressure of AI workloads.
This tension is at the heart of a conversation with David King, Cadence’s Product Engineering Architect. He explained how digital twins, liquid cooling, and AI reshape data center thinking. But change, he makes clear, is anything but straightforward.
When risk aversion meets rising demands
According to Cadence’s recent report, Data Center Evolution: The Innovation Imperative, the benefits of technologies like digital twins and liquid cooling are widely recognised. Yet many operators remain reluctant to implement them. The reasons are familiar – cost, complexity, uncertainty – and yet increasingly problematic
“The data center industry, particularly at the physical infrastructure level, is quite risk-averse, and understandably so,” King explains. “Operators are responsible for keeping critical systems running, so there is a natural hesitancy around change unless there is absolute certainty about the outcome. While the benefits of innovation are increasingly clear, there is still significant uncertainty about which direction is best, which makes people cautious.”
Financial concerns magnify this caution. Even when organisations acknowledge the long-term efficiency gains of next-generation infrastructure, short-term capital investment can be challenging to justify. “Unless they have full confidence that a project will deliver measurable returns, there is little appetite to take the leap,” King adds. “That is the core of the inertia, uncertainty and cost create barriers to change, even when the potential benefits are well understood.”
Many executives find the risk-reward equation difficult to balance, particularly when legacy systems still ‘work’, even if they are operating inefficiently, consuming excessive energy, or inhibiting scalability. However, this reluctance to move forward may create deeper risks in the near future.
AI is not just another workload
With its erratic power profiles and expanding compute demands, AI poses a new threat to underprepared data centers. While 74 per cent of decision-makers believe their facilities are currently managing AI pressures effectively, King sees this confidence as potentially fragile.
“The most immediate risk is failure, outages caused by inadequate cooling or power provision as workloads increase. But even before that point, the fear of such failure can restrict growth,” King warns. “Operators install new equipment, believing they understand how it interacts with their cooling and power systems, only to see issues emerge over time. That erosion of confidence results in cautious behaviour. Enterprises start looking for new capacity or sites before fully utilising what they already have.”
Part of the solution lies in improving the fidelity of modelling and simulation before deployment. In King’s view, digital twins offer a clear path to building confidence in innovation investments, allowing teams to trial changes virtually before any hardware is touched. But here, too, misconceptions get in the way. “A common misconception is that digital twins are just visualisations, overhyped models that offer little real insight,” he continues. “That was true of early versions, which lacked intelligence. Today’s digital twins are far more sophisticated but are also perceived as complex and data-hungry, which can put people off.”
Fragmented data landscapes compound this perception. “Operators often worry that they do not have the required data in a centralised or digital form,” King says. “It is often scattered across spreadsheets, CAD drawings, or even physical documents. But those gaps in data represent real risks. If you do not know exactly how your systems are configured or behaving, you are operating blind to some extent.”
Starting small and scaling smart
The fear of complexity often results in digital twin adoption being deferred, particularly at brownfield sites. But King believes this caution is misplaced. “It is easier on greenfield sites because you are starting from a clean slate and can plan data integration from the outset,” he says. “But digital twins can be just as valuable on brownfield sites. Building a digital twin often reveals just how much critical information is held in the heads of a few experienced staff. That is a major operational risk. By capturing and centralising that knowledge, businesses make themselves far more resilient.”
King recommends starting small, maybe modelling a single room or system as a pilot, then building outwards. This approach reduces the initial investment and project risk while helping teams understand what data is required and where to find it. Once that is established, expanding the model and scaling the digital twin to cover more of the site or estate becomes easier.
Once in place, a digital twin not only supports better operational decision-making but also enables replication, reducing the time-to-value for new data centers and making modular, repeatable deployments more viable. This is especially relevant for organisations seeking to scale their AI infrastructure across multiple sites.
Cooling the complexity
Among the most talked-about innovations in the data centre world, liquid cooling holds promise but also stirs apprehension. King acknowledges the hesitation, particularly among enterprise operators. “Many enterprises are turning to colocation and data centre providers to deliver that expertise,” he explains. “But for those directly responsible, investing in people and skills is the first step. They need engineers and operators who understand how these systems work and how to manage and deploy them effectively.”
Cadence’s software is already playing a role here, helping design and simulate liquid-cooled environments. However, retrofitting existing sites remains a major undertaking. The shift from air-cooled to hybrid systems requires infrastructure changes and new ways of thinking about risk, reliability, and maintenance. “Liquid cooling can feel more daunting than innovations like AI because deploying AI on-premises often utilises a complete solution by one vendor,” King says. “In contrast, liquid cooling requires connecting multiple components across several vendors, increasing the complexity and risk to the project.”
King is also cautious about assuming early adoption equals long-term stability. “It will be interesting to see what issues hyperscalers might report after running liquid cooling at scale for 12 months,” he muses. “As well as how enterprise and other forms of data center respond to this in terms of their liquid cooling adoption.”
A living model for a moving target
Perhaps the most important message is that none of these technologies, whether digital twins, AI, or advanced cooling systems, exist in isolation. Their effectiveness is interdependent, and their success depends on continuous adaptation. This is especially true for digital twins, which must evolve over a facility’s entire lifecycle.
“The digital twin starts as a design model, then becomes an as-built model following construction, and finally transitions into the operational model,” King explains. “At each stage, updates are required to reflect reality. Once a site is operational, equipment gets replaced, layouts change and new technologies are deployed. If the digital twin does not evolve with those changes, its value quickly diminishes.”
He points to commissioning as a particularly valuable use case. “One challenge with data center commissioning is the lack of actual servers at that stage, so heaters are used to simulate load,” King adds. “These are imperfect substitutes, but digital twins can be calibrated to account for those limitations. AI workloads introduce further challenges. Their power draw can ramp up incredibly quickly, far faster than traditional systems. In many cases, it is not even possible to physically replicate these loads during commissioning. Digital twins can simulate those conditions and verify that systems behave correctly under stress, even if the physical test is not feasible.”
Imperfect data, perfect opportunity
For many organisations, the road to innovation is blocked not by ignorance but by hesitation. Yet the tools to mitigate that risk are already available. Starting small with digital twins, investing in simulation before committing to infrastructure changes, and acknowledging the real operational risks of inaction are pragmatic steps that any executive team can understand.
The future will not wait for perfect conditions. As AI demand accelerates, energy costs rise, and sustainability targets loom, the imperative to act will only grow stronger. “Even a legacy site with limited automation can gain value from a digital twin if approached correctly,” King concludes. “The key is starting with whatever data is available, then iteratively improving fidelity and granularity over time.”
It is not a question of whether innovation will come. It is whether data centers will be ready when it does.
