Executive Summary
- Artificial intelligence depends not just on silicon, but also on power, while demand for large-scale compute is rising faster than power systems can keep up.
- The next challenge is securing power that is clean, affordable and available every hour of the year, without putting grids, or the communities that depend on them, at risk.
- Here, Alex Tancock, CEO, InterContinental Energy explains why the answer requires rethinking where, and how, AI infrastructure is built.
The International Energy Agency estimates that data centres used around 415 TWh of electricity in 2024, about 1.5% of global consumption. By 2030 it expects that to more than double to around 945 TWh, roughly the annual electricity use of Japan today, with AI the main driver.
Grids are already feeling the strain. In Ireland, data centres now account for more than a fifth of metered electricity, and new sites must bring their own generation or storage to secure a connection. Singapore regularly pauses new data centre approvals, and Johor, which absorbed much of its overflow, is now introducing its own restrictions. Australia is legislating to require large data centres to be “net-generators, not net-users”, adding at least as much new renewable energy as they consume.
For small facilities, buying renewable energy from wind and solar farms elsewhere on the network can sometimes work. At gigawatt scale it becomes much harder. The data centre still draws from the grid every hour, including after dark, and someone must pay for the backup supply and network upgrades that make that possible. Building in and around congested cities, on expensive land, with communities already wary, makes the task harder still.
Community concern is growing. Data centres’ effect on household power bills has become a public concern in many markets, and in major cities, where projects compete for land, water and grid capacity, opposition is growing. The direction of travel is clear: large data centres will increasingly be expected to bring their own power, and to do so without passing costs or risks on to others.
Moving data is easier than moving electrons
There is a different approach now emerging because not all AI computing has to happen close to its users. Services such as payment processing need instant responses and will stay near the cities they serve. But model training, fine-tuning, batch processing and scientific computing, to name just a few emergent AI use cases, depend on huge amounts of power rather than split-second response times, and they will make up a growing share of demand.
Those workloads can go where clean energy is abundant and communities are supportive. Instead of choosing a site and then working out how to power it, developers can start with the resource and design generation, long-duration backup and delivery into the campus from day one – creating a self-reliant system that strengthens, rather than strains, the wider power system.
Sharing resources across borders
Energy resources and demand for compute are not evenly distributed. Some of the world’s most important digital hubs, including Singapore, Japan, South Korea and parts of Europe, have strong demand but limited land and renewable resources. Others have land, sun and wind in abundance but smaller domestic markets.
That creates room for partnership. Singapore already plans to import low-carbon electricity from its neighbours. For compute-intensive AI workloads there is a simpler option, because it is far easier to move data than electrons. Linked by subsea cables, data centres in resource-rich regions can serve customers in resource-constrained ones, while sensitive and latency-critical services stay close to home.
Done well, these partnerships strengthen energy security on both sides. Importing countries ease the pressure on their own grids and gain access to reliable green compute, while exporting countries build new industries around their natural advantages. Long-term cooperation between governments on connectivity, data and investment already happens, so the mechanisms are in place
Putting the model into practice
In APAC, Australia has the land, renewable resources and project experience to put this approach to work at scale. The Australian Renewable Energy Hub (AREH) in the Pilbara, is a good example. Developed to supply renewable power and green hydrogen for heavy industry, the project holds state and federal environmental approval for up to 15 GW of wind and solar generation and is now being requested to host AI compute.
With daytime solar complemented by night-time wind, the combined system can deliver a 74% base capacity factor, well above what solar or wind typically achieves alone. The more consistent the generation, the less backup is needed to achieve five-nines reliability.
Green backup is possible too
Pairing AI compute with already planned hydrogen production addresses the hardest part of the problem: keeping renewable power available around the clock. Under InterContinental Energy’s P2(H2)Node design, each 1 GW node would combine up to 500 MW of data centre capacity with 500 MW of hydrogen production. The two loads behave very differently.
A data centre needs constant power, while hydrogen production can ramp up and down. When wind and solar output falls, hydrogen production is turned down first so the data centre keeps priority access to supply; during extended lulls, stored hydrogen can be used to generate power. It is a match made in heaven.
Such a system can achieve 99.995% uptime, with backup produced on site rather than drawn from a stressed grid or reliant on fuel deliveries, and benefits from shared infrastructure lowering the cost of both products. Because each node is a standard, repeatable unit (think Lego blocks), capacity can be added in phases as demand grows. The principle is transferable: a flexible industrial load can act as the buffer that makes round-the-clock renewable power practical.
A blueprint for elsewhere
Demand for AI compute will keep growing everywhere, and meeting it means looking beyond the out-dated and unsustainable model of grid-connected data centres in and around metropolitan areas. Cities will always host the services that need to sit close to their users, but they cannot supply the land, energy and resilience that gigawatt-scale compute requires.
These principles apply wherever strong renewable resources and supportive communities meet; start with the energy and land, firm it on site, and deliver reliable green power at a competitive price rather than a green premium. Australia has the opportunity to show how it can be done, meeting its own demand and, through partnerships across Asia, helping to power its neighbours’ AI requirements too.
