This product has already been purchased.
- AI is not just digital, it is infrastructure. This distinction is important because of the mismatch between the development of AI technology and the infrastructure needed to support it.
- Organisations are making assumptions about where their computing will occur, grid capacity, energy costs and infrastructure resilience.
- Boards are now being asked to consider emissions reduction, electricity procurement and major capital investment as related questions.
When boards discuss AI, they usually focus on what the technology can do. Instead, they should be asking what it depends on.
An AI strategy can look compelling on paper, while resting on invisible assumptions about electricity, network capacity, computing infrastructure, water, technology supply chains and the resilience of third-party services.
That was the core theme to emerge from an AICD Climate Governance Forum panel, which delved into the growing collision between AI, data centres and Australia’s energy transition.
“AI is not just digital, it’s actually infrastructure,” said Kee Wong FAICDLife, a director of the Australian Energy Market Operator (AEMO), at the event hosted in Melbourne in late August.
The distinction matters, explained Wong, because the pace of technological change is now running ahead of the infrastructure needed to support it.
“The digital innovation advancement of AI technology is at technology speed, the infrastructure needed to host it is at construction speed,” he added.
For organisations adopting AI, Wong and his co-panelists warned this creates a potential gap between the strategy approved by the board and the physical capacity available to deliver it.
The physical footprint of cloud computing
Sabooh Whitelaw, Associate Vice President, Energy and Utilities at AirTrunk, noted that organisations are already making hidden assumptions about where their computing will occur, grid capacity, energy costs and infrastructure resilience. Too often, these assumptions never appear in the AI strategy presented to the board.
“We cannot realistically trace every AI workload to a particular data centre,” Whitelaw said. The task instead is to understand the material infrastructure, energy and supply dependencies supporting an organisation’s AI strategy, and “who carries a risk if those dependencies fail”.
Whitelaw stressed that question becomes more important as AI demand grows.
Data-centre electricity use was cited during the panel at about 3% of Australia's electricity supply today, with demand forecast to reach some 13% by 2035. At the same time, coal-fired generation is retiring and renewable generation, storage and transmission must be built to meet rising demand across the economy.
Wong described data centres as “our new smelters” – major new electricity users, but also a new engine of economic growth.
Attendees heard that for boards, the exposure will differ based on their operating model. A business using cloud or third-party AI services may lack direct visibility into the physical infrastructure supporting those tools. But an organisation committed to dedicated computing capacity will face immediate questions about electricity, water and delivery timeframes.
Ultimately, in both cases, an organisation doesn’t need to own a data centre for the dependency to disrupt its operations.
The new energy trilemma
Beyond physical capacity, AI computing creates a unique capital investment challenge.
As Wong noted, up to 80% of an AI data centre's cost goes into graphics processing units (GPUs). These chips can become obsolete within four years, yet the building and power infrastructure housing them is built for a 25- to 30-year lifespan.
Boards are caught in the middle: they have to make three-decade infrastructure commitments that are flexible enough to survive four-year technology cycles.
That financial friction bleeds directly into a company's ESG and energy targets.
Lucia Cade FAICD, chair of Infrastructure Victoria, said the rapid uptake of AI and electrification was putting new pressure on the traditional energy trilemma of security, affordability and environmental sustainability.
Previously, boards might have considered emissions reduction, electricity procurement and major capital investment as separate questions. Increasingly, she said, “they’re coming all at once”.
Cade added that the organisations able to reconcile those competing demands would be best placed to respond to the change.
“Boards that can navigate that triangle of the trilemma – being able to decarbonise while maintaining reliable supply and competitiveness – are really going to be the winners,” she said.
But those trade-offs are complex. For instance, a data centre that uses water for cooling uses less electricity, but a “dry campus” that conserves water requires significantly more energy, explained Whitelaw.
Social licence as a core capability
As AI infrastructure expands, these macro-level trade-offs are becoming local issues. Cade pointed out that community friction is often driven by an uneven distribution of benefits – while the digital service improvement is shared globally, the physical infrastructure costs and disruptions are borne locally.
“People see infrastructure in its local context as much as they do in the service context,” she said. “The service improvement is shared equally. But the local cost is born very locally.”
For Alison Dodd, a partner at Herbert Smith Freehills Kramer who works with energy project developers, that is occurring as projects themselves become larger and more complex.
“This has gone from the days when we had little solar farms,” said Dodd. “These are multibillion-dollar projects.”
Grid connection bottlenecks, workforce shortages and convoluted regulatory environments are actively slowing project delivery.
Consequently, community engagement has become part of whether projects can be delivered at all.
Dodd explained that social licence had moved from “a nice to have” to “a core critical delivery capability”.
“It’s a board level issue, not an afterthought at the end of a project,” she said. “It’s really a core part of project governance.”
Questions for the board
Ultimately, the panelists agreed the board’s task is not to predict exactly how AI technology will evolve over the next decade. Rather, it’s to stress test the physical realities underpinning the digital strategy.
“AI is physical,” said Whitelaw. “It’s growth on infrastructure that takes years to plan, approve, and build. That’s why the digital and energy transition need to be considered together.”
When reviewing an organisation’s AI strategy, the board should consider asking management:
- What physical infrastructure and energy dependencies are required to execute our AI strategy?
- Which of these dependencies do we control and which rely on third-party resilience?
- What is our contingency plan if underlying infrastructure capacity or energy projects are delayed?
- If these physical constraints alter costs or delivery timelines, where does the risk sit within our organisation?
Latest news
Already a member?
Login to view this content