AI has moved from just a workload category for data centres to a structural force that is reshaping timelines, costs, and risk assumptions across the industry. At the same time, the pace at which AI demand is growing has outstripped the ability of energy systems, water use, and regulatory frameworks to respond.
This mismatch is now not only a technical challenge, it also creates constraints on the infrastructure and the resilience and efficiency of data centres when demand grows.
“Historically, it was more viable for data centre operators to develop or procure renewable energy sources independently, but now it’s not,” said Meiske Sompie, Asia partner for TBH, an international specialist planning and management construction consulting firm, in an interview with iTNews Asia.
“This issue requires coordinated action across the ecosystem, including governments, utilities, developers, investors, and local communities.”
Power constraints directly impact communities, data centre developers face rising costs to build and operate green facilities, and power operators need long-term planning, Sompie added.
Taken together, she said the most concerning data point is the widening gap between the pace of AI growth and the pace of energy infrastructure deployment. This imbalance is forcing a rethink of how data centres are financed, built, and operated.
From static assets to lifecycle thinking
AI is breaking long-standing assumptions about asset lifecycles and profitability.
“AI workloads require higher rack densities and significant power, which in turn accelerates the depreciation and shortens the lifecycle of key assets, particularly chips and networking equipment,” said Sompie.
One way data centre operators can maintain profitability is by moving away from a static, upfront investment approach.
Sompie said facilities should be planned and managed using a full lifecycle perspective.
This includes designing adaptable infrastructure that allows for rapid hardware refreshes and phasing capital expenditure to align with expected technology upgrade cycles, she added.
Another important factor is construction planning.
From a construction consulting perspective, a significant portion of cost risk stems from delays during the construction phase, said Sompie.
Better upfront construction planning helps ensure projects are delivered on time.
Data centre projects are highly time-sensitive, with target delivery timelines of 12 months or less.
Completing projects on schedule allows operators to enter the market earlier and capture demand sooner, which is critical for maximising profitability in a fast-growing AI-driven market.
- Meiske Sompie, Partner, Asia, TBH
Alongside power constraints, she said water scarcity has become a growing concern, particularly in Asia. Areas like Johor, parts of Peru, and India are already experiencing significant water stress.
Water planning now a standard requirement of new data centre projects
There have been recent announcements highlighting these constraints, for example, in Johor Bahru, data centre operators have reportedly been told they may need to wait until mid-2027 to gain access to a sufficient water supply.
Sompie said data centre clients are discussing renewable or recycled water solutions, including the development of on-site water recycling plants.
“This shows that data centre developers are considering the impact of growth on local communities, ensuring expansion does not negatively affect community access to water,” she said.
There are projects where data centres are developed alongside dedicated water treatment plants, ensuring a more sustainable water strategy from the outset, she added.
However, the most crucial aspect is that data centres that are not modular or AI-optimised are likely to become redundant within the next decade. This risk brings the question of stranded assets into sharper focus.
Look for signals that a facility may become a stranded asset
AI workloads, computing intensity, and modelling requirements are expected to continue rising, while AI facility technologies and design approaches evolve, increasing the risk for facilities that cannot adapt.
From a planning perspective, the role of consultants is to support operators with long-term lifecycle planning.
Sompie mentioned that operators typically assess facilities over a 20-to-30-year lifespan, with a focus on whether a facility can remain sustainable, valid, and operationally relevant over that period.
Several early warning signs can indicate that a facility is at risk of becoming a stranded asset.
One key indicator is escalating operating costs, particularly where outdated cooling systems and technologies are in use, said Sompie.
Closed-loop systems, immersion cooling, and direct-to-chip liquid cooling are the three commonly used cooling technologies. Operators may use hybrid approaches based on data centre design, AI workload requirements, local climate, and operating costs.
“Direct-to-chip liquid cooling is currently the most widely adopted in the region, reflecting its longer track record in Asia and more mature supply chain.
“Another warning sign is limited flexibility to scale, making it difficult for the facility to accommodate higher-density or more advanced workloads,” said Sompie.
Additional risks include systems or equipment that cannot be replaced or upgraded with newer technologies, as well as difficulties integrating emerging technologies into existing infrastructure, she added.
These factors point to a clearly defined and potentially shortened lifespan for the facility.
“As a result, investors are seeking projects that offer flexibility to be redesigned or reconfigured across cloud, edge, and AI computing use cases. This flexibility provides more optionality and reduces long-term risk,” said Sompie.
Who should bear the cost of future-proofing AI-ready data centres
According to Sompie, this is a complex question. In addition to operators, governments, chipmakers, and utilities, banks and investors also play an important role, particularly given their influence on how projects are financed.
From a lending and investment perspective, there is a tendency to focus on short-term profitability.
However, in this context, banks and investors may need to take a different approach by prioritising sustainability and long-term outcomes rather than maximising short-term returns, Sompie mentioned.
“Governments also need to go beyond providing regulatory frameworks and compliance requirements, and should actively support sustainable development by encouraging greener approaches to data centre projects. This may include offering incentives that encourage developers and investors to adopt more environmentally responsible solutions,” she added.
For data centre owners and operators, there is a constant tension between maintaining profitability and investing in sustainability. Taking a short-term, lower-cost approach may appear attractive initially, but failing to consider long-term sustainability and community impact can lead to challenges later.
Community opposition is a real risk, said Sompie.
For example, in South Korea, it has become increasingly difficult to develop data centres due to strong resistance from local communities who are concerned about the environmental and social impact of these facilities.
She added that there is no single clear leader who should bear all the responsibility or cost.
An example of where collaboration has worked can be seen in Singapore. The Building and Construction Authority (BCA) and IMDA, the two governing bodies overseeing data centre construction, have introduced green building standards that all data centre projects must comply with.
Compliance with these standards is effectively incentivised, as meeting the requirements is necessary to proceed with development in a constrained market like Singapore.
AI makes energy and sustainability a board-level issue
According to Sompie, AI is often framed as a burden on energy and water systems.
In practice, it is acting as an accelerator, she said.
“The scale and urgency of AI demand have pushed renewable energy, sustainability, and long-term infrastructure planning higher on executive agendas than previous technology waves ever did.”
While AI currently strains existing systems, it is also forcing faster decision-making and broader collaboration across the ecosystem.
Similarly, the rise of AI has made carbon-neutral and net-zero targets harder to achieve, but not less important, said Sompie.
These targets remain essential because they provide a clear benchmark for sustainability in an environment of rising energy and water demand.
“Carbon goals are no longer primarily about corporate responsibility. They are increasingly tied to business resilience. Climate-related risks are already affecting infrastructure reliability, insurance costs, capital availability, and global supply chains,” she said.
Organisations that embed sustainability into operational and investment decisions are better positioned to manage these pressures, she added.
