AI Data Centres: Black & Veatch on the Infrastructure Crisis

What happens when an entire industry’s power consumption quadruples in six years? The data centre industry is about to find out. After maintaining a steady 2% share of US electrical power for the past decade, these facilities are set to consume 8% by 2030: a transformation driven by an explosion in AI workloads that now form as much as 50-70% of computing demands.

This dramatic shift has exposed infrastructure vulnerabilities not seen since the 1980s. Operators can now be forced to wait years for grid connections, forcing them toward self-generation strategies that were unthinkable just five years ago. Water scarcity has moved from an operational footnote to a primary element in site selection as AI chips generate unprecedented heat loads. Regulatory frameworks designed for modest facilities now confront multi-gigawatt developments that more closely resemble power plants than traditional data centres.

The challenge isn’t just scale – it’s interconnection. Each infrastructure problem amplifies the others, creating cascading effects that traditional development approaches can’t address. The industry’s response requires fundamental changes in expertise, partnerships and technology to maintain the reliability that underpins modern digital infrastructure.

Introduction: The New Data Centre Reality

Picture a large data centre from 2014. At 10MW, it was considered substantial. Fast-forward to today, and that same facility likely wouldn’t register as a rounding error on hyperscale campus plans. 

AI workloads demonstrate fundamentally different behaviour patterns compared to traditional computing. Where standard servers typically cruise at 30-40% capacity utilisation, machine learning algorithms drive processors to sustained peak performance levels around the clock. GPUs designed for AI training can consume 300-700 watts per chip compared to 150-200 watts for traditional server processors. All of this resulting heat generation requires more cooling capacity and demands entirely different approaches to thermal management, power delivery and facility design.

Phil Fischer, Client Executive in Black & Veatch’s data centre and mission critical team, has watched this transformation accelerate beyond industry predictions. “If we go back 10 years, something that was in the double digits of megawatts was considered a large data centre. So 10-50MW was an enormous campus,” he explains. Today, these facilities are dwarfed by hyperscale data centre campuses reaching “500MW to multiple gigawatts in size.”

The pace of change has caught infrastructure systems unprepared. Electrical grids that expanded steadily through the 20th century have remained largely static whilst other sectors improved efficiency, freeing capacity that AI workloads now consume. “Data centres will add about 65GW of capacity, and they’ll represent 8% of total consumption by 2030, if not more,” Phil describes. “The change that is here has become very swift, and the responses in terms of building more capacity – whether that’s distribution, transmission or even generation – is not something that can be done overnight.”

Water systems face comparable pressures as higher heat densities challenge cooling methods that functioned adequately when servers operated at lower temperatures and data centres maintained smaller footprints. The predictable development patterns that defined the industry for decades have been replaced by complex infrastructure constraints that demand new solutions.

The power crisis: Energy access and generation

The numbers tell a stark story. A 500MW campus requires roughly the same power as a mid-sized city, delivered with the reliability that utilities typically reserve for hospitals and emergency services. Now imagine dozens of such facilities planned across regions that haven’t seen significant load growth in decades.

Traditional utility planning assumes gradual, predictable growth patterns. A new manufacturing plant might add 20-30MW over several years, giving utilities time to plan upgrades and secure funding. Data centres shatter this model by demanding hundreds of megawatts with deployment timelines measured in months rather than years. The mismatch between business requirements and infrastructure reality has created a bottleneck. 

“Now the wait for many data centres is potentially up to five to 10 years to deliver the energy that they need directly from the grid,” Phil says.

Five to 10 years might as well be forever in the AI race. Companies investing billions in machine learning capabilities today can’t wait for utilities to build transmission lines that may face permitting challenges, community opposition and financing constraints. 

These narrow timelines have forced a fundamental reconsideration of how data centres source power. Data centre operators, who once focused exclusively on computing and networking, now need expertise in generation technology selection, fuel procurement, environmental compliance and utility coordination. 

Supply chain constraints add another layer of complexity. Power generation equipment now carries lead times of five years for some technologies, forcing procurement decisions before final site design completion, essentially requiring operators to bet on capacity requirements years in advance.

But what if these constraints could become opportunities? Behind-the-meter generation enables arrangements that benefit both data centre operators and utilities, while virtual power plant concepts allow utilities to access data centre backup generation during peak demand periods, providing grid support without affecting normal operations while creating revenue streams for operators. 

Could nuclear power hold the answer? Small modular reactor (SMR) technology promises clean, reliable baseload power without the massive infrastructure requirements of traditional nuclear plants. 

“Nuclear will be done dramatically differently than in the past, with small modular reactors,” Phil acknowledges. “That does take time, though we see significant support for this becoming an emerging energy market centred around nuclear technology.”

Case Study: Client-Owned Substations Accelerate Campus Development

When a global co-location provider faced utility delays stretching five years or more, they made a decision that would have seemed radical just a decade ago: take ownership of electrical infrastructure traditionally owned and operated by utilities. Black & Veatch was selected to build substations at six expanding data centre locations, allowing the client to control construction schedules and costs while enabling utilities to connect directly with transmission lines. The approach cut campus construction timelines by half or more, demonstrating how infrastructure ownership can become a competitive advantage when traditional approaches fail to meet business requirements.

Water scarcity and sustainability

Water usage in data centres once followed a simple logic: use what you need for cooling, discharge the rest and pay the bill. AI workloads have made that approach untenable: generating heat loads that would overwhelm traditional cooling systems. But the challenge goes deeper than technical requirements: touching on community resources, environmental sustainability and public perception in ways that can catch operators unprepared.

A modern hyperscale facility might consume millions of gallons daily for cooling while the surrounding community faces drought restrictions. The optics alone create problems, but the underlying resource competition poses bigger questions about sustainable development and community impact. How can data centre campuses justify massive water consumption for computing when residents can’t water their gardens?

Heather Cheslek, Global Industrial Water Solutions Leader at Black & Veatch, sees this tension regularly. 

“Water consumption is a huge concern,” she says. “We have several clients that are looking at various ways that they could improve efficiencies within the cooling system itself.” 

Site selection has evolved accordingly. Where operators once prioritised connectivity and power availability, water assessments now carry equal weight. Mikeal Vincent, Client Account Manager in Black & Veatch’s Technology and Data Centre Group, explains the shift: “Water stress and water scarcity. It has to be thought about because obviously you don't want to build in a place where you won’t have long-term reliable water supply.”

The solution often lies in redefining what constitutes acceptable water sources. Recycled water from wastewater treatment plants addresses both availability and perception issues by using water that would otherwise be discharged. “We have worked with a hyperscaler in a community where we performed a study recently to look at whether or not they could utilise a recycled stream from the wastewater treatment plant – treated effluent – to basically replace their use of potable water, drinking water, at that site,” Heather describes.

But water and energy consumption create complex trade-offs that vary by location and season. Air-cooled systems reduce on-site water usage but increase electrical demand, potentially shifting consumption to power generation facilities that may use more water per megawatt hour than direct cooling. 

“If you’re increasing your electrical load or your power generation or your power that you’re utilising, based on an air-cooled or mechanical system, your off-site water demand is increasing,” Mikeal explains. “Power generation off-site – I think a lot of people do not think about that, but there’s a significant amount of water use in the power generation process at the utility.”

Regional variations must be considered, too. A facility in Arizona faces different water constraints than one in Oregon, and cooling strategies must adapt accordingly. Free cooling using ambient air temperatures works well in northern climates but offers limited benefits for AI workloads that generate heat year-round.

When it comes to thermal management, direct-to-chip liquid cooling represents perhaps the most significant technological shift. Rather than cooling entire rooms, these systems capture heat at the processor level with dramatically higher efficiency. 

“It’s the liquid cooling direct-to-chip that’s the future here, as we’re moving forward,” Mikeal notes. This approach enables waste heat recovery for space heating or other applications, turning a disposal problem into a resource opportunity.

Design optimisation becomes crucial when every gallon matters. Computational Fluid Dynamics (CFD) analysis reveals inefficiencies in airflow patterns that waste cooling capacity. Mikeal describes: “We do a lot of CFD analysis and look at where the hot zones and the cold zones are. If you’re trying to ensure that the hot air isn’t mixing with the cold air and making those zones more efficient, you're moving the warm air more efficiently.” These improvements can reduce cooling loads by 10-20% through better airflow management without additional equipment investment.

The discharge side of the equation has become equally important. Utilities need capacity to treat blowdown water, or facilities require independent discharge permits that can affect project timelines and costs. 

“We also have to be concerned with discharge as well,” Heather explains, “and whether or not the utility that they may be partnering with is going to be able to treat that discharge long-term."

Case Study: Comprehensive Water Infrastructure Solutions

A global technology leader's multi-site expansion revealed the full complexity of modern water management. One project required a 2.5-mile wastewater pipeline connecting cooling tower blowdown to the local treatment plant while coordinating NPDES permit modifications across multiple agencies. Another involved alternatives analysis for discharge options, evaluating whether blowdown could discharge directly to receiving bodies or required on-site treatment systems. A third project developed conceptual design for a wastewater treatment plant to supply recycled water, examining existing utility infrastructure and identifying new requirements to meet both current demands and sustainability goals.

Regulatory and permitting complexities

Imagine trying to get approval for a nuclear power plant, but the regulatory framework assumes you’re building a warehouse. That’s essentially what data centre developers face when facilities reach the scale and complexity of modern hyperscale campuses. The regulatory systems simply weren’t designed for facilities that consume more power than small cities while incorporating generation, water treatment and telecommunications infrastructure that would normally be spread across multiple projects.

As a result, air quality permits that once required anything from three to six months now stretch to up to 18 months for full approval. Mike Rinkol, who specialises in air emissions at Black & Veatch, explains the systematic problem: “We’re also seeing some instances based on jurisdictions, staffing issues and so getting permits into those and getting them reviewed has taken up a lot of time.”

The Authority Having Jurisdiction (AHJ) variations create particular challenges for operators developing facilities across multiple regions. What passes regulatory muster in one county might violate standards in the next, even within the same state. Kathleen Margolis, Client Account Manager within Black & Veatch’s Data Centre Mission Critical Group, describes this fragmentation: “AHJs by nature are regional... There are communities that are ready to accept data centre developments, and there are those that are not.”

This variation forces location-specific expertise and community engagement strategies that can't be standardised across developments. A facility design that works perfectly in Northern Virginia might face months of regulatory challenges in a market where data centres represent unfamiliar technology.

The move toward on-site generation fundamentally changes the regulatory landscape by introducing power plant regulations to data centre projects. “A lot of these data centres are now looking at producing the power by themselves,” Mike says. “So there’s a whole different realm of regulations that may be applicable for the actual power plant side of things.” Air quality, water discharge, noise control and safety standards that may be entirely unfamiliar to data centre developers suddenly become project-critical requirements.

Acoustic regulations also present particularly thorny challenges due to lack of standardisation and limited regulatory expertise. Andrew Truitt, Subject Matter Leader for acoustics and noise control at Black & Veatch, highlights the complexity: “The problem, especially in the US, with acoustical noise regulations, is that there’s almost no reciprocity across any states with noise regulations. It varies state by state and, even within a state, it’s going to vary county by county and city by city.”

Even established data centre markets struggle with this issue – Andrew notes that data centre capitals like Prince William County and Loudoun County “have changed their noise regulations almost yearly for the past five to six years.”

Community engagement has become essential because data centres create public concern due to unfamiliarity with the technology and infrastructure requirements. “People don't understand data centre development,” Kathleen explains. “So there’s big reactions to a data centre coming in because it looks like something new and ominous.”

But what if this community concern could be turned into support? Educational outreach that explains data centre operations and economic benefits often builds the community backing that facilitates regulatory approval. The numbers are compelling: construction generates US$200-300m in local revenue while operational facilities contribute over US$1m annually in local taxes per building.

Advanced engineering and design solutions

The stakes for getting data centre design wrong have never been higher. A single design error in a multi-billion-dollar facility can cost tens of millions to fix while causing operational disruptions that affect global services. This reality has driven adoption of engineering tools that enable virtual validation of every system before the first shovel hits the ground.

Kyle Kropf, Associate Vice President and Data Center Portfolio Solutions Lead, describes the transformation: “The most significant advancement is how we use data to make good engineering decisions, versus in the past where we relied more on 'industry best practices.’”

CFD modelling has become essential because traditional rules of thumb simply don’t work at current scales and heat densities. The integration of air cooling, liquid cooling and hybrid systems creates thermal interactions that can’t be predicted without detailed analysis. 

Kyle explains: “We build a digital twin model to analyse both the air and liquid cooling systems within and external of the data hall to identify hotspots and cold spots. We then mitigate these inefficiencies through further refinement of the mechanical cooling design.”

The complexity multiplies with direct-to-chip liquid cooling for AI workloads. These systems remove approximately 80% of heat through liquid while the remaining 20% requires air cooling, creating thermal management challenges that require both cooling methods to work together seamlessly. Changes in workload or distribution in power (aka transients) affect thermal patterns in ways that can't be intuitively predicted, making detailed modelling essential for reliable operation.

Digital twin technology extends this analysis beyond initial design into lifecycle management. Building Information Modelling data created during design provides the foundation for construction tracking, commissioning validation and ongoing operations optimisation.

But perhaps the most critical analysis focuses on reliability. Five nines availability – 99.999% uptime – means less than five minutes of downtime per year. At current facility scales, even brief outages can affect millions of users and cost millions in revenue. Reliability, Availability and Maintainability (RAM) analysis provides data driven quantitative assessment of whether designs can actually achieve these targets.

“We’re modelling the systems all the way down to the server racks,” Kyle says. “So everything from where the electrons are generated all the way to where it’s being consumed at the chip level.”

Campus infrastructure integration complicates reliability modelling because behind-the-meter generation, substations, water treatment systems and telecommunications must all coordinate to achieve overall availability targets. Equipment failures that might have minimal impact in traditional facilities can cascade through integrated systems to affect overall performance.

The scale of modern facilities amplifies acoustic concerns even when individual equipment units meet acceptable standards. Andrew describes the cumulative effect: “It’s not just the size of the individual units, but it’s the quantity of units that really drives a lot of the noise concerns to these sites.” Hundreds of cooling units can combine to create noise levels that exceed regulatory limits despite individual compliance, requiring system-level analysis rather than component-level evaluation.

The consequences of skipping acoustic analysis can be sobering, Andrew warns of an operational data centre which was acoustically modelled post-construction due to noise complaints and ultimately “had to replace hundreds of roof-mounted HVAC units with new units.” 

“Apart from the cost to buy those new units as well as install, I don’t even want to think about the operational downtime impacts of incorporating that on a live data centre project,” he says.

The integration of these analysis tools enables optimisation across competing objectives – energy efficiency, reliability, environmental compliance and capital cost. 

“Making good engineering decisions early in the concept design and the detailed design,” Kyle emphasises. “So that way, we know when we go build that data centre, it’s going to be the most efficient, most sustainable, best design we can do with today’s tools and technology available.”

Case Study: World's First Zero-Water Consumption Data Centre

Nautilus Data Technologies took an entirely different approach to the water challenge by building their data centre on a barge. The 7-megawatt facility at the Port of Stockton uses patented TRUE™ technology that recirculates water from the San Joaquin River as a heat sink, eliminating water consumption entirely. Black & Veatch's commissioning oversight verified that during testing at 1.45MW load, room temperature remained consistent at 73 degrees Fahrenheit even during generator backup operation. Rob Pfleging, President of Nautilus Data Technologies, explains: “Commissioning was the final step in bringing our vision of delivering an energy and cost-efficient data centre to the market. It was important for us to know our equipment functions as intended.”

Testing and validation provide final verification that design intent translates into operational performance, but they also reveal the interconnected nature of modern data centre systems. Jenn Cahill describes: “With these two programmes, we feel like we can design a really solid data centre that is going to ensure that whatever they're promising there – if it’s a colocation – whatever they’re promising the end user or a hyperscaler, that they're going to get what they ask for.”

As Jenn concludes: “These innovations are really fun because we're looking at virtual power plants and the data centres are working with utilities to allow the utility to actually pull from their backup generation when they're not using it during peak hours. And so they’re supplementing the grid, they’re making our grid more reliable and more resilient without impacting the ratepayers.”