Data Centre World 2026: five takeaways from the fastest-growing demand source in the NEM
Data Centre World 2026: five takeaways from the fastest-growing demand source in the NEM
Data centres are the fastest-growing new source of demand in Australia, and will fundamentally reshape the NEM, providing opportunities for new generation and storage. Modo Energy's Central case forecasts NEM data centre demand growing by 400% to 28 TWh in FY2036.
On 8 and 9 September, Data Centre World Australia in Melbourne brought industry participants together to discuss data centre construction and operations, including discussions on energy sourcing, alongside an Australian Energy Council (AEC) lunch on data centres and the grid.
This article shares five key takeaways from the event: trends in data centre forecasts, the importance of fast energisation and its main obstacles, new AI-load build moving to regional areas, and the rapidly evolving policy landscape.
1. Data centre demand forecasts are being revised up, reflecting growing developer ambition
Power is the limiting factor of data centre growth. Data centre proponents expressed a variety of opinions on data centre growth. The consensus, though, was that access to reliable, compliant energy is the bottleneck to buildout.
Energy industry bodies are increasing data centre demand forecasts to reflect the growing ambition. AEMO's 2026 ESOO Step Change forecast is up to 50% higher than its 2025 ESOO (and 2026 ISP) projections. This shifts from below Modo Energy's central case to sitting in the middle of Modo Energy's central and high cases. The forecast increase is driven by a larger announced project pipeline, 67 GW against 38 GW a year ago. Faster ramp assumptions, a higher FY26 baseline and a new allowance for Asia Pacific demand served from Australia also contribute.
2. Speed-to-power is more important than energy cost
Modo Energy forecasts lower electricity prices in the next few years as renewables come online before coal retirements. However, any energy cost savings from building later are dwarfed by expected additional early-year data centre revenue.
NEXTDC, an Australian co-location landlord, bills its tenants approximately three times the all-in electricity cost per MWh of load. An AI tenant it passes the power bill onto is expected to spend only around 5% of revenue on energy, as estimated from disclosed US GPU-cloud contracts.
In contrast, a contracted campus forgoes a year of revenue for every year without power: up to 35 times a year's grid power bill at A$100/MWh.
Therefore, a data centre is likely willing to pay for more expensive power that allows for faster energisation. Conference speakers and attendees repeatedly stressed that energy price is secondary to speed to power.
3. The speed of grid connection and energy buildout limit data centre deployment
Builders want power far faster than the system delivers it. AirTrunk indicated its customers want capacity in 12 to 18 months. Grid connection alone can take 2 to 3 years.
Another issue is the significant gap between data centre load energisation and the commissioning of the energy resources it incentivises. This leads to increased grid stress in the interim, especially for data centres that require major transmission augmentations.
Data centre developers expressed their eagerness to pay for their share of network augmentation costs based on a transparent methodology. A VicGrid representative clarified at the AEC event that due process, rather than funding, is the reason for the long augmentation schedule. They encouraged creative solutions from data centres and energy developers.
In the US, the shortcut is on-site gas, built quickly enough to energise a site before its grid connection. In the NEM, many data centre proponents think this is difficult. They cite long lead times, high gas prices, reliability concerns, and emerging regulation that favours renewables. However, one company at the conference also claimed that they can deliver on-site gas generation and supply gas with a lead time of one year.
4. LLM-based data centres are going regional as metro capacities fill up
Data centre developers expressed a preference to increasingly site large language model (LLM)-serving data centres in regional areas. They favour proximity to large energy and transmission infrastructure over proximity to users.
Fibre connection is unlikely to be an obstacle for regional data centres, several stakeholders confirmed. Recent builds put long-haul fibre at about 75,000 AUD per kilometre, approximately 100 times cheaper than HumeLink.
LLM training and inference are both insensitive to latency. Anthropic shared that since agentic tasks run for “many minutes, potentially hours”, network round-trip time is immaterial for inference loads. Siting the load close to reliable grid supply matters more.
The named pipeline is beginning to show this trend. In 2026, 5.4 GW of new plans were announced outside metro sub-regions, against 0.9 GW of new metro plans. Before 2026, regional plans totalled only 0.13 GW. Most of these new sites self-describe as AI factories or AI cloud data centres.
5. Energy policy is rapidly evolving, and many customers want renewables
Rules and policies on data centre development are under consideration at every level of government.
Energy is the bottleneck, so both the data centre and energy industries are watching the type and strength of regulations favouring renewables. Representatives from data centre developers expressed uncertainty at the current state of policy flux.
At the same time, several data centre developers and compute buyers said they prefer renewable power whenever available. In their vision, gas would only play a firming role only when renewables are insufficient.
What is coming next
Modo Energy is developing a global data centre model that forecasts the all-in power cost of a data centre campus. It considers factors such as wholesale price forecasts, offtake structures, on-site resources, and demand response. Deployment of the model is starting in the US, followed by Australia.
Please reach out for a demo, or to evaluate the energy cost of your data centre. We compare on-grid and off-grid power mixes. That includes speed-to-power arrangements where a campus runs on behind-the-meter resources while it waits for a grid connection.
This work builds on our US research on how data centres navigate grid connections and the case for behind-the-meter gas generation. Our previous NEM data centre research is below.
