Net-Generators, Not Net-Users: Designing Data Centres for Australia's New Energy Rules

Australia has just changed the rules for data centres. Large sites now have to be net-generators, not net-users - put back at least as much energy as you take out.

On 15 July, in a speech at the University of Sydney, Prime Minister Anthony Albanese announced that Australia will legislate a single national framework for large-scale data centres. The announcement said that large sites must pay the full cost of their grid connection, and put at least as much energy into the grid as they take out: they must be "net-generators, not net-users".

Since then the policy has moved quickly. Energy Ministers agreed in late July to progress the regulatory arrangements, National Electricity Rule changes go to them in September, and the AEMC has published detailed advice setting out how the obligations would actually work.

The legislation is still to come, but the mechanics are now largely visible and nearly every one of them is a question about how the site’s energy system is optimally designed, something we’re super focused on here at Gridcog.

What the framework asks

Four energy and infrastructure obligations were flagged in the speech:

1. Pay the full share of grid connections
2. Underwrite new power supply
3. Put at least as much energy into the grid as the site takes out
4. Maximise water and energy efficiency

That ambition has since been turned into concrete regulatory design. On 28 July the Energy and Climate Change Ministerial Council welcomed the AEMC’s advice and agreed to progress arrangements mandating that data centres offset their demand through new renewable generation.

The AEMC recommends four measures working together:

1. A certificate obligation. Data centres would surrender Renewable Electricity Guarantee of Origin (REGO) certificates from new, additional generators to cover the power they use, matched to peak and off-peak periods rather than hour by hour.
2. A contract obligation. They would have to show the AER that their demand is backed by new firm capacity, with intermittent renewables attracting a low firmness factor.
3. Market registration. Large data centres would become registered market participants with telemetry obligations, though active market participation would stay voluntary.
4. Flexibility and siting. Connection agreements would encourage sites to shift demand and to co-locate with generation, with streamlined pathways for those that do.

How much of a site’s load must be offset is still open. The AEMC floats a nominal figure such as 70%, or letting each jurisdiction set its own, which would walk back some of the ambition in the Prime Minister’s July speech.

The obvious objection to these requirements is that it takes roughly three to five years to approve and build new wind and solar generation whereas a data centre can be built in two. If a project cannot bring new generation online in step with its load, the question becomes what else on site can close the gap: behind-the-meter generation, storage, flexible compute, or staged energisation.

This is where the AEMC’s design does something commercially significant: behind-the-meter generation and storage would reduce the certificate obligation, and flexibility and storage would reduce the firming requirement. 

So every megawatt of behind-the-meter generation, storage and flexibility is both a cost and a permanent reduction in the certificates and firm contracts the project has to buy for the life of the asset. This is super significant.

Even more constraints and complexity

The energy obligation is also not arriving on its own. In March the AEMC released a draft rule on technical standards for large data centre connections, sorting inverter-based loads into tiers by size and requiring them to ride through defined voltage and frequency disturbances rather than tripping offline. Above 100 MW, and for anything connecting at transmission level, the full set of access standards applies automatically. The final rule is due in late October.

The motivating case is Virginia, where 60 data centres dropped about 1,500 MW off the grid simultaneously during a single fault in July 2024. As AEMC Chair Anna Collyer put it, data centres "aren’t passive loads anymore; they’re active grid participants".

So an Australian project now has to satisfy two quite different demands at once: bring its own energy, and control its interaction with the grid at the point of connection.

The power system never sees the underlying compute load directly. It sees net load at the connection point. On-site generation and storage can make the underlying load and the net load the grid sees look very different to each other, and it is the net load that every one of these obligations is measured against.

Fluence and DNV reached the same conclusion in their joint research on the Australian market: storage that enforces ramp limits at the point of interconnection can unlock connection pathways that would otherwise be delayed or denied.

Why this is a modelling problem, not a compliance checklist

Two very different kinds of modelling are involved here, and they are easy to conflate. Before a large load connects, system and network operators require electrical studies: load-flow, dynamic and stability work, and electromagnetic transient studies down to sub-cycle timescales to capture how the site’s power electronics and protection behave. That work lives in specialist tools like PSS®E, PSCAD and PowerFactory, which produce the import limits, ramp limits and ride-through obligations written into the connection agreement.

What that grid code modelling cannot tell you is how to make the resulting project commercially optimal. What is the right asset sizing and resource mix of generation and storage? How much battery power does a ramp limit actually require? How much state of charge has to sit idle as ride-through headroom rather than earning market revenues or reducing supply costs? How much certificate and firming liability does each megawatt of behind-the-meter capacity buy back?

None of these can be answered with an annual average. A REGO liability split across peak and off-peak bands is a function of net grid import interval by interval, after on-site generation and storage have done their work. A firmness assessment turns on the shape of a contract against the shape of a load. Both are simulation and optimisation problems before they are compliance problems, and they are exactly what Gridcog is built to solve.

How Australia compares internationally

At Gridcog, we have customers modelling data centre and large load projects across many markets. Here is how some of the main international markets compare to Australia.

Ireland (CRU)

Bring your own capacity: dispatchable generation or storage matched to import capacity on a de-rated basis, plus at least 80% of annual demand met by renewables generated in Ireland, on a six-year glide path. Non-compliance can mean a cut to contracted import capacity. In force since December 2025.

Great Britain (NESO)

Flexible, non-firm demand connections where import can be reduced to a profile or dynamic setpoint, alongside "first ready, first needed" queue reform. Also considering whether a minimum level of demand flexibility should be a condition of connection.

Germany (BMWE)

Data centres above 300 kW must cover their electricity use with renewables on a balance-sheet basis: 50% since 2024, with the 100% step set to slip from 2027 to 2030 under a bill approved by cabinet in June 2026. PUE limits apply alongside it.

Netherlands (ACM / TenneT)

Flexible contracts becoming the default for large new connections — time-based and fully variable transport rights, capacity-restriction and capacity-steering contracts. Hyperscale siting confined to two designated locations, with Amsterdam closed to new data centres.

Texas (ERCOT)

"Batch Zero" maximum grid-withdrawal limits, withdrawal-limited private use networks, and a provisional controllable-load bridge to firm service.

California (CAISO)

Large-load technical requirements under consideration: ramping, power quality, ride-through and modelling.

Singapore (EMA/IMDA)

Grid capacity awarded competitively through Data Centre Calls for Application, conditional on world-leading efficiency and, under DC-CFA2 and the Green Data Centre Roadmap, at least 50% of annual demand from approved low-carbon sources.

Japan (METI)

Non-firm curtailable connections, steering new sites to regions with spare capacity, efficiency standards from FY2029.

‍

Each of these markets picks off part of the problem. Ireland asks for self-supply, Germany for renewable cover, the Netherlands for flexibility, Singapore for efficiency. Germany is the instructive contrast: it accepts renewable cover on a balance-sheet basis, pooled across the year, with no requirement that the generation be new. Australia would demand certificates from new generators only, and matched to the periods of the day when the site actually draws power. And it is proposing to do all of it, all at once.

This will be the first integrated national framework of its kind anywhere in the world. Which means the compliance case a developer builds in Australia may well be the one they are asked to reproduce in every other market they enter.

What this means for data centre developers

The framework sets the obligations but it does not tell a developer what the best project design looks like. Several decisions go into that:

• How and what new generation and firming gets contracted or built
• What storage and thermal generation sits on site, and what else it is allowed to do
• Whether and how much of the compute load can flex, and when, recognising that co-location operators rarely control their tenants’ workloads
• How large a grid connection to take, and whether part of it is flexible rather than firm
• What gets built in what order

None of these can be settled in isolation, because each one changes the others. Many designs will satisfy the obligations. They will not all cost the same over the life of the asset, and they will not all reach energisation as quickly.

This is what Gridcog is built for. We model the site interval by interval across a full year of prices, weather and workload, and use mathematical optimisation to find the least-cost combination of grid import, thermal generation, wind and solar, battery storage and flexible compute that satisfies the constraints.

Some results from our own modelling show how much the answer moves. For a hypothetical 100 MW data centre in Slough, full grid supply came to about £1.9 billion over 15 years. Swapping part of that connection for a 50 MW gas genset brought it to £1.79 billion. A microgrid-style portfolio of solar, wind and storage alongside a smaller connection cut whole-life cost to £795 million.

And academic modelling of a 250 MW AI training site on NSW market data found that a 400 MW / 400 MWh co-located battery cut days of missed committed workload from 17 in 31 to 5, and lifted day-ahead committable compute by around 38%.

That difference is worth quantifying while load requirements and connection terms are still being negotiated, rather than after. If you are developing an Australian data centre project against these obligations, book a working session with the Gridcog team to model your options.