Urbanity-26: From AI to Modular Delivery
At Urbanity-26, we explored AI in construction, modular construction and digital engineering, focusing on practical adoption and better project outcomes.
In Episode 5 of the Kapitol Conversations podcast, our Directors Andrew Deveson and David Caputo sit down with Simon Cooper, Group Chief Development Officer at NEXTDC, to discuss the future of data centres and what it will take to build for the AI era.
The conversation covers NEXTDC’s growth, the move from air cooling to liquid cooling, modular delivery, planning, workforce productivity and the need for better information across the full project team.
More broadly, it asks a practical question: if data centre demand keeps growing, can the construction industry keep delivering these projects in the same way?
Simon Cooper joined NEXTDC in 2011, shortly after the business opened its first data centre, a two-megawatt facility in Brisbane. At the time, the organisation had around 20 people and was still building the operating structure needed to support its first facilities.
The position looks very different now. In the podcast, Simon discusses current data centre developments valued in the billions of dollars and an annual capital program that continues to grow. He describes the change in scale as significant, but the more difficult issue isn’t the size of the numbers. It’s the pressure to deliver capacity quickly enough.
The computing equipment supporting AI requires major investment. Once that equipment has been purchased, customers need it operating and generating value. An unused rack isn’t simply empty space. It’s expensive equipment waiting for power, cooling and a completed facility.
That pressure flows directly into data centre construction. Programmes become tighter. Decisions need to be made earlier. Supply chains must be considered during design. The project team also needs to distinguish between genuine speed and activity that looks fast but causes rework later.
For builders, this is where scale becomes a responsibility. Larger projects need stronger systems, clearer roles and more detailed planning. Increasing the workforce without improving the structure around it can create more congestion, weaker supervision and lower productivity.
Our experience in data centre construction has shown us that technical complexity can’t be managed through effort alone. It needs buildable design, coordinated information, disciplined programme management and teams that understand the detail before it reaches site.
One of the clearest changes discussed in the episode is the move from air-cooled to liquid-cooled data centres.
Traditional data halls move cold air towards servers. Fans inside the equipment pull that air across components, and the resulting hot air is removed and cooled before being circulated again. This approach has supported data centres for years, but the density of new computing equipment is changing the equation.
Simon Cooper explains that earlier racks commonly operated at relatively low power densities. Current commercially available equipment can require far more power per rack, and equipment under development may push that requirement higher again. Air can only carry away so much heat within a fixed space.
Liquid cooling moves water or another cooling liquid closer to the equipment. The liquid passes through or near the hottest components, takes on the heat and carries it away. Water transfers heat more efficiently than air but bringing liquid into that environment creates a different set of design and construction requirements.
The pipework needs a high level of cleanliness. Connections, testing and commissioning need close control. Mechanical systems become part of the direct protection of expensive computing equipment. The availability of specialist materials and fittings also becomes a supply chain issue as multiple projects seek the same products.
This affects far more than the mechanical design. It changes spatial planning, installation sequences, trade interfaces, quality controls and commissioning. A team that has delivered air-cooled facilities can’t assume that the same methods will work unchanged for a liquid-cooled build.
That’s why detailed coordination matters. The benefits of digital engineering are clearest when the model helps the team resolve physical constraints, service interfaces, access and temporary conditions before installation starts. Our in-house Digital Engineering team uses coordinated models and Autodesk Construction Cloud to review design information, identify clashes and provide controlled information to project teams in the field.
Technology helps the team see the problem earlier. The value comes when experienced people use that information to make a better decision.
As the amount of infrastructure grows, the industry needs to reduce the volume of complex work completed in live construction environments.
Simon Cooper describes off-site manufacturing as one of the most important ways to improve safety and speed. Instead of delivering individual pieces of equipment to site and building rooms around them, complete assemblies can be manufactured, tested and then installed in position.
NEXTDC and Kapitol have applied this thinking to data centre delivery before. In the episode, Simon recalls a project where 24 switch rooms were installed over several long weekends. This avoided a long sequence of separate transformers, switchboards and other equipment arriving and being assembled progressively on site.
But the group is also honest about the limits of modular delivery.
A module isn’t automatically cheaper or faster. If a conventional design is divided into boxes late in the process, the project may duplicate structure, create difficult interfaces or shift problems rather than remove them. The design needs to start with manufacturing, transport, cranage, installation and connection in mind.
This is where early contractor involvement can make a practical difference. Bringing builders, designers, planners, manufacturers and key trade partners together early creates the opportunity to test the methodology before the design is locked in. It allows the team to review logistics, temporary works, procurement lead times, buildability and programme impacts while those decisions can still be changed.
For data centre projects, the value isn’t in being able to say that something is modular. The value sits in the work removed from site, the risks controlled in a factory and the certainty created during installation.
NEXTDC D1 Darwin, constructed using modular construction methodologies.
A large part of the podcast focuses on an uncomfortable construction problem: as projects get bigger, productivity doesn’t always improve.
In theory, scale should create efficiency. In practice, a crew that works well with 20 people may struggle when it grows to 100 or 200. The supervisor can’t hold the same level of detail in their head. More reporting lines appear. Work areas become congested. Responsibility becomes less clear.
David Caputo explains that a good foreman who can personally coordinate a smaller job may not be able to manage a workforce several times larger without better systems around them. The answer isn’t to expect one person to work harder. It’s to break the work into clear zones, define responsibilities and give supervisors accurate information at the right level.
One example shared in the episode involved a formwork crew of around 40 people struggling to complete an eight-day cycle. After reviewing what each person was doing, the workforce on the deck was reduced to around 20 and the team began achieving four and five-day cycles. The workers removed from the deck weren’t dismissed. They were allocated to other work.
The improvement came from understanding the task, not pushing more labour into the area.
Another example involved a structure that required a five-day cycle but had been taking seven days. The team mapped the five days hour by hour, agreed what needed to happen and then achieved the target on the next cycle. This wasn’t a new piece of software or a larger crew. It was better planning at the level where the work was happening.
These examples get to the heart of the productivity issue. People can be working hard while the system around them works poorly.
When construction waste is discussed, the focus often goes to surplus materials. That matters, but waste also sits in the time spent correcting incomplete designs, responding to avoidable RFIs, moving labour between poorly prepared work fronts and rebuilding work that wasn’t right the first time.
During the podcast, Andrew Deveson describes an RFI as a type of defect in the information. A drawing or specification has reached the project team, but something within it can’t be built or doesn’t coordinate. Resolving that issue then creates work for the builder, consultant, client and affected trade partners.
One RFI may appear manageable. Hundreds of them create a separate workflow running beside the build.
The same is true of physical rework. Completing a task twice is already expensive. Completing it again after access has tightened, surrounding work has progressed and other trades have moved into the area is worse.
This is an important area for the wider innovation in construction industry conversation. Better outcomes don’t always come from adding another platform. They often start with clearer design information, earlier coordination and a shared understanding of what the project team is trying to achieve.
The technology matters when it helps the team catch an issue before materials are ordered or people arrive on site.
Simon Cooper argues that major projects would benefit from spending more time planning and less time repeatedly changing the plan during delivery. The challenge is that project teams often feel pressure to start before the design, procurement strategy and work sequences have been examined at the required level.
Planning is also a specialist skill. A strong project programmer needs to understand the software, but they also need practical construction knowledge. They need to know what a change means for access, labour, procurement, safety, testing and every activity sitting downstream.
The potential role for AI is to help those experienced planners test more options. Simon discusses AI-supported constraint modelling that could allow a planner to compare different programmes, adjust work areas, test resource assumptions and understand the effects of each decision. The human planner still provides the construction logic and judgement. The tool helps them iterate faster.
The next step is connecting that programme to work on site.
If a project identifies nine active work fronts, each one should have a clear task, planned crew and expected result. At the end of the shift, the team should be able to compare the plan with what happened. Were the right trades present? Was the task completed? If not, was the assumption wrong, did a constraint appear or did the work front lack the required information?
That feedback needs trust. Clients, builders and trade partners won’t share meaningful productivity data if they believe it will only be used to shift blame. The commercial model, programme and site reporting need to point towards the same outcome.
Data centres were once largely invisible outside the industries that built and operated them. AI has changed that.
The public conversation now includes where data centres are built, how much power they require, how they are cooled and what they mean for communities. Simon Cooper discusses the work undertaken by NEXTDC and other operators to provide clearer, more consistent information to governments and communities through Data Centres Australia.
That visibility brings scrutiny, and it should.
The sector needs to explain the infrastructure clearly, listen to concerns and show how standards are being met. Builders need to manage the effects of construction on surrounding communities while maintaining safety, quality and programme. Project teams also need to keep improving how materials, energy and labour are used.
For us, building for the AI era comes back to the basics done at a higher level: plan earlier, coordinate tighter, test the methodology and give people information they can use.
Demand may be moving quickly. The work still needs to be built properly.
