How to deliver data centres when infrastructure lags demand

We are in the middle of one of the largest infrastructure build-outs in modern history. Power capacity tied to data centres and AI is projected to grow from around 30 GW in 2025 to more than 90 GW by 2030. That’s more than the entire power demand of California or Great Britain. 
Yet the constraint on that growth has grown beyond demand and into the enabling infrastructure, which needs to be in place before a project can run. This extends beyond physical infrastructure to social licence, where community resistance due to concerns around energy use, water and long-term value can shape whether projects proceed at all.

In many regions, grid capacity now limits data centre growth. Projects are held back by power availability, while delays to substations and connections create multi-year bottlenecks. Wait times for an electric utility connection can run 1 to 5 years, while permitting delays can add 6 to 18 months. High demand for large equipment can take 8 to 24 months, design complexity adds another 6 to 18 months and a shortage of construction and specialised workers adds 6 to 12 months more. The result is a project paced by its slowest enabling system.

The encouraging part is that much of this time can be won back by changing how work is sequenced and which levers are pulled, including:

• Enabling parallel work by building many smaller shell data centres on one site
• Using modular design and a prefab supply chain
• Collaborating with suppliers to reduce equipment lead times
• Optimising the construction labour force and site logistics
• Partnering locally to speed up permits and power connection.
• Engaging early with local stakeholders to surface concerns around power, water and land use

Meaningful time can be recovered at every stage. For example, partnering locally to accelerate permits and power access can save 6 to 18 months against the connection wait alone. The lag is usually a function of how the work is structured rather than a fixed cost of doing business.

The second problem is more subtle though sometimes more damaging. Power, water and permitting are deeply interdependent, but too often they are still solved one at a time. A decision taken to relieve one constraint can tighten another.

Choosing a cooling technology that reduces water consumption, for instance, can increase electricity demand and vice versa. Water use is not an inherent feature of a data centre at all. It’s usually a consequence of design choices, economics and constraints, and, increasingly, how those choices are perceived by the communities hosting the infrastructure. 

A recent industry example is NVIDIA’s move to liquid cooling that runs at up to 45 degrees Celsius, warm enough to let many sites tackle heat with dry coolers and cut cooling water use close to zero in the right climates. But that’s only one aspect of the whole story, since many data centres still run on fossil fuels (which ironically use more water, particularly coal). Treating these as separate problems to optimise in isolation worsens the whole system.

Isolation also distorts the signals that utilities and communities receive. “Phantom demand” is a good example: only around one in 5 to 10 data centre applications results in a built facility. That’s a 10 to 20 percent chance of a data centre actually being built.

One Australian water utility reported nineteen live data centre applications seeking 55 megalitres a day between them — the equivalent of water for 200,000 people — yet most of that demand will never materialise. When utilities plan capacity against headline numbers like these, they risk over-investing in infrastructure that may never be used.

On the power side, a single hyperscale campus of 200 MW, 500 MW or a gigawatt can land on a grid node that was never designed for a load of that size, compressing decades of incremental growth into one connection point. That forces hard trade-offs between thermal headroom and security of supply, and raises stability questions around reactive power, inertia and grid strength that a single large load was never expected to create.

The answer is to model across the boundaries rather than within them. That means integrated, whole-of-life and whole-of-system techno-economic modelling, GIS-driven multi-criteria analysis to get siting right for water, cooling and power together, and Modelled Adaptive Pathways Planning, which combines adaptive pathways planning with probabilistic modelling to size capital under uncertainty. These tools have been used on industrial hub projects in hydrogen and mixed industries, and they exist precisely to stop teams optimising one constraint at the expense of the rest.

Data centre development needs to be treated as a fully integrated system, with constraints identified early and weighed by all parties, so people can make informed trade-offs about whether power, cooling or permitting is the limiting factor in each location. Done early, this reduces rework and, most importantly, the risk of delay, while improving confidence in delivery times.

Crucially, it cannot stop at the fence line. The parties that determine whether a project succeeds, including grid operators, regulators, local authorities and the communities directly adjacent to the infrastructure, must be part of the planning alongside the developer. This is especially important with growing community scrutiny over how data centres use shared infrastructure such as power and water, and what long-term value they return to the region. Master planning with government, as is the case on Jurong Island, builds space for data centres while maximising the shared benefit of power and cooling assets.

Developer-to-grid partnership is emerging as a potential solution too. Google, for example, has committed gigawatt-scale demand response across its US operations with utilities including TVA. In Finland, data centre waste heat is reused for district heating, turning a data centre project into a community asset. And in the UK, data centres feed free heat to swimming pools. 

Data centres don’t have to be mere consumers of energy and water. They can also be system participants and even community infrastructure assets. But it requires breaking down the silos and collaborating with developers, utilities, regulators and communities from the start.

The pattern across power, water and permitting is the same. None of these constraints are insurmountable on their own, but solved in isolation, and too late, they compound into delay. The projects that move fastest stick to these concepts:

• Start earlier and integrate land, power, water and community engagement into one coherent plan.
• Sequence for parallelism, using smaller shell builds, modular design and prefab supply chains, so no single enabling system sets the pace.
• Model the whole system under uncertainty, using integrated techno-economic modelling, GIS-driven siting and adaptive pathways planning before committing capital.
• Bring in teams that have delivered this before, and treat utilities, regulators and communities as partners.

In a build-out happening at roughly 10 times the speed and with far less room for error, alignment is the real accelerator. If you’re planning, financing or approving a data centre, the moment to align land, power, water and community is now, before the first constraint sets your schedule.