AI and the grid: when the solution is also the problem

There is an unusual tension building in Australia's energy sector. The same technology that is helping manage grid complexity is also creating new demand that the grid is not yet built to handle. AI is both the tool and the load, and that duality is starting to shape real policy decisions.

In February, we wrote about where AI stands in the Australian energy transition broadly. Since then, several developments have sharpened the picture. AEMO has updated its forecasting methodology. The AEMC has proposed new grid connection rules specifically for data centres. And a UNSW research partnership is testing AI-driven energy management under real operating conditions. Each of these tells part of a story that matters for anyone working in energy in regional Australia.

The demand side: data centres as a new category of load

AEMO's Draft 2026 Integrated System Plan, released in late 2025, includes a significant change in how the market operator thinks about electricity demand. For the first time, data centres are being forecast and reported as a standalone demand category, separate from industrial or commercial load. That is not a technical footnote. It is an acknowledgement that AI-driven computing infrastructure is becoming a structural feature of the National Electricity Market.

Under AEMO's Step Change scenario, data centre electricity demand is projected to grow from around 2.2 per cent of grid-supplied NEM consumption to 6 per cent by 2029-30. The major projects are concentrated in Sydney and Melbourne for now, but investment is spreading. Regional centres with available land, cooler climates, and proximity to renewable generation zones are increasingly attractive for data centre developers.

For regional energy businesses, this creates a complicated picture. More demand should mean more opportunity, particularly for renewable generators. But it also means more pressure on transmission infrastructure that is already strained in many regional corridors. And data centres are not like traditional industrial loads. They run continuously, they are sensitive to power quality, and when they disconnect during a grid disturbance, the sudden loss of load can itself cause instability.

New rules for a new kind of consumer

The Australian Energy Market Commission recognised this problem in March 2026 with a draft rule proposing new technical standards for large data centres connecting to the grid. The core requirement is straightforward: facilities drawing 30 megawatts or more must be able to "ride through" grid disturbances rather than disconnecting. If a voltage dip hits the network, a large data centre that suddenly drops off can make the problem worse, potentially triggering cascading outages.

The AEMC's proposal also includes provisions for data centres to provide grid support services. A large facility with sophisticated power management systems could, in theory, become a valuable grid asset rather than just a consumer. That possibility is still theoretical for most facilities, but the regulatory framework is being built to accommodate it.

Stakeholder feedback on the draft rule closes on 7 May 2026, with final rules expected by mid-year. For energy businesses in regional areas, particularly those in renewable energy zones where data centres might co-locate, these rules will shape the terms of engagement.

The supply side: AI helping manage what it consumes

While AI drives new demand, it is also becoming more capable on the management side. The partnership announced in March between UNSW researchers and Aussie Solar Batteries is a useful example of what this looks like in practice. The collaboration, funded through the Australian Government's Trailblazer for Recycling and Clean Energy (TRaCE) initiative, is focused on building AI-driven energy management platforms for solar and battery systems.

The project is specifically about virtual power plants (VPPs), the coordination of distributed energy resources like rooftop solar, home batteries, and eventually electric vehicles into a system that behaves like a single, dispatchable power source. VPPs have been discussed in Australia for years. What has been missing is the intelligence layer that makes coordination work reliably at scale.

The UNSW project is testing forecasting algorithms, demand-side management, optimisation, and digital-twin modelling under real operating conditions. The aim is to move beyond the pilot stage and toward commercially deployable systems. The project runs until the end of 2026, with commercialisation pathways designed to keep the technology and its benefits within Australia.

This matters for regional Australia because distributed energy is disproportionately a regional story. Rooftop solar penetration is high in many regional areas. Battery uptake is growing. But the value of these assets depends on coordination. A household battery that charges and discharges on a fixed timer captures some value. The same battery managed by an AI system that reads wholesale price signals, weather forecasts, and local grid conditions in real time captures significantly more. Multiply that across thousands of installations and the economics shift meaningfully.

AEMO's evolving toolkit

At the grid level, AEMO continues to develop its own AI capabilities. The market operator is building energy simulation tools for the entire east coast network, incorporating detailed models of wind farms, solar farms, and battery storage to simulate their behaviour when connected to the grid. In Victoria, Jemena is using machine learning to monitor grid health across its distribution network, detecting stress points before they become failures.

These are not speculative capabilities. They are operational tools being deployed and refined. But they are also concentrated in the hands of large network operators and the market operator itself. The challenge for smaller players in the energy sector, whether community energy groups, smaller retailers, or regional renewable developers, is accessing the benefits of these tools without the data infrastructure and technical teams that the larger organisations possess.

The honest assessment

The AI-energy relationship in Australia in early 2026 is genuinely complex. AI is not simply "helping" or "hurting" the energy transition. It is doing both simultaneously, and the net effect depends on where you sit in the system.

If you are a large renewable generator or network operator, AI is becoming essential infrastructure. Better forecasting, predictive maintenance, and optimised dispatch are real and delivering value. If you are a data centre developer, the regulatory environment is tightening, but the fundamentals of demand growth remain strong. If you are a regional energy business or community energy group, the tools are improving, but the gap between what is available to large operators and what is practically accessible to you remains wide.

The UNSW VPP project and similar initiatives offer a path toward closing that gap, by building AI management tools designed for distributed energy assets rather than centralised generation. But commercialisation timelines are measured in years, not months.

For regional energy operators right now, the practical advice has not changed much since our February article. Start with generative AI tools for documentation, compliance, and grant applications, where the value is immediate. If you have operational data (generation records, consumption patterns, maintenance logs), explore what AI tools can tell you about that data. And pay attention to the regulatory changes. The AEMC data centre rules and AEMO forecasting updates are reshaping the market context that every energy business operates within.

The tension between AI as consumer and AI as management tool is not going to resolve. It is going to intensify. Understanding both sides of that equation is increasingly part of operating in the Australian energy sector.