How we model data centre contracting
This page sets out the assumptions behind our analysis of what the proposed national data centre energy rules would make NEM data centres contract. We apply the rules as proposed in late 2026 and build the least-cost book of new supply each state's covered data centre load would need in 2035.
Which data centres the rules cover
We assume the national obligations start in 2028. Legislation is due in early 2027, and the Clean Energy Regulator needs about a year to set up the renewable obligation.
- Covered: every data centre phase at a site of 30 MW or more of ultimate connection capacity that switches on in 2028 or later. The AEMC places the obligation on data centres
that connect to the grid after the legislation commences
, and the government aims to cover data centresthat are yet to connect to the electricity system
. - Exempt: phases already running before 2028, including their later ramp-up. Both the AEMC and the government rule out retrospective obligations.
- Firming timing: the firming obligation needs rule changes and may not apply until about 2029. We keep the 2028 cut-off for it too, because the firming request aims for a rule that
covers all data centre load over the threshold size
, whichmay include both existing and future data centres
. Sites switching on in 2028 would need firm cover once the rule starts. If the rule reaches older sites as well, our covered load is conservative. - Phase by phase: a new hall at an operating campus is covered when it switches on, because it adds new load to the grid, which is the load the rules target.
- Tiers: every covered project connects at over 100 MW, so all covered load faces the federal top tier.
Obligations follow consumption
A covered site's obligations track its actual consumption as it ramps up, not its connection capacity. The renewable obligation applies to annual electricity consumption from the time they begin operating
, the government says obligations could be linked to this growth
, and the firming request lets the AER use the demand ramp profile in each site's connection agreement.
Example: a hyperscale site with a 100 MW connection that switches on in 2029. Its obligation (bars) follows average consumption on Modo's hyperscale ramp, reaching full load of about 43% of connection capacity after eight years.
Where covered load comes from
Covered load comes from Modo's Central data centre demand forecast, built from a register of project phases. Each phase carries a build probability by status and planning stage, a commissioning delay, a mature draw per MW of IT capacity and a ramp-up curve by operator type. Demand the forecast adds beyond the named pipeline comes from projects not yet identified, so it is covered in full.
We apply the national rules in every state. Queensland data centres buying from state-owned generators could apply to the AER to modify their obligation; we model the national standard as written. Tasmania's proposed campuses exceed 30 MW and carry the same obligations. Our dispatch runs cover the four mainland regions, so Tasmania's covered load takes Victoria's mix per GW, and is left out of cost figures.
What covered data centres must contract
| Requirement | What we model | Why |
|---|---|---|
| Energy | Long-term contracts with new wind and solar in the data centre's state, producing at least 100% of its annual consumption. | Energy ministers agreed to mandate that data centres offset their electricity demand by investing in additional renewable generation located in the jurisdiction. The proposed rule only requires certificates, but once all must come from new plant, their price has to cover the gap between a new project's cost and its market earnings. Under the Renewable Energy Target, a similar certificate scheme, new build was mostly financed through PPAs bundling energy and certificates. Contracting the energy also lets that output count towards firm cover in every hour. Annual matching gives data centres flexibility on timing at the lowest compliance cost, and the firm cover test already secures supply in every hour. |
| Firm cover | The book's own supply covers at least a set share of load in every hour: 70%, 90% or 95%. The article leads with 90%. We stop at 95%, because the last few hours depend almost entirely on how well an operator foresees wind lulls. | The firming request asks for cover appropriate to their demand shape and profileand lets the AER assess the full 24-hour profile. AEMO will advise the level. 90% is a strict every-hour test that leaves the market to fill the last tenth in the hardest hours. The request does not set a time step. It would assess contracts, not delivery: the AER would check a portfolio's firmness against the site's demand profile at connection, at reviews every 5–10 years and when load or contracts change. We test hourly because the request asks for supply that meets the timing, duration and magnitudeof demand across the full 24-hour load profile, which a daily or monthly energy balance cannot show. Our test checks that the book can physically deliver cover in every hour, a stricter version of the AER's contract test. |
| Location | All contracted supply sits in the data centre's state. | Ministers made in-state supply the default, unless a state opts out. |
| Storage minimum | Battery power of at least 25% of contracted wind and solar capacity. | New South Wales' data centre guidelines set this ratio, the most specific storage rule in the NEM. Batteries sit above this minimum in every state at 90% cover. |
| Battery ceiling | With gas, contracted batteries are capped at 1.5 GW per GW of data centre load. Without gas there is no cap: batteries are sized for firming only, then scaled up until cover holds. | Beyond about 1.5 GW, most of each battery's energy is sold to the market rather than serving the data centre: about 60% at 1.3 GW and up to 73% at 2 GW. More batteries selling surplus would be cheaper than the gas they replace, but the ceiling limits the capital and market exposure a data centre takes on. |
| Generation cap | Contracted wind and solar output of at most 150% of consumption. | Keeps the book to volumes a data centre would sign for, rather than a large overbuild to cover a handful of hours. It does not bind in any state. |
| Firming technology | Four- and eight-hour batteries in every case; gas, modelled as reciprocating engines, in the with-gas case. | The standards allow firming through gas, batteries or hydro. Reciprocating engines run more efficiently than open-cycle turbines at the thousands of hours a year the cover test asks of gas, and come in modules sized for data centre loads. The without-gas case shows the cost if gas is excluded. |
| Market trades | Every contracted asset trades in the spot market when the data centre does not need it. Wind and solar sell surplus output, gas engines sell when the price beats their running cost, and batteries charge from the grid and sell back. Market purchases for the load are limited to the share above the cover level. | A data centre holding a portfolio of contracts trades its residual position. The firming request does not restrict how storage is charged, and battery contracts such as tolls leave the operator free to trade capacity the data centre is not using. |
| Battery earnings | 88% round-trip efficiency and at most one full cycle a day. Each discharge is valued 9–16% below the hourly price, so a standalone battery in our model earns the 2035 energy revenue in Modo's Central forecast for four- and eight-hour batteries in each state. FCAS revenue from the same forecast is added per MW. | Our model sees every hourly price in advance. Modo's forecast trades batteries at five-minute resolution with realistic price foresight and is calibrated to the Modo NEM BESS Index, so matching its revenue keeps battery earnings realistic. |
| Turn-down (tested) | The data centre can cut up to 25% of its load for up to 2 hours a day, which counts as cover. Tested, not shown in the charts. | The firming request counts dependable demand flexibility, and New South Wales proposes this condition in connection agreements. Two hours a day cannot bridge a multi-day wind lull. Under realistic forecasting it changed gas capacity by between −16% and +1% by state and cost by $0.6–2.6/MWh; without gas it trimmed batteries by 1–4%. |
How the firm cover test works
In every hour, wind and solar output, battery discharge and gas must supply at least the cover level. Batteries charge from contracted wind and solar or from the grid, lose 12% of the energy that passes through them, and must be full enough to carry the load through wind lulls. Output the data centre does not need, including from gas and batteries, is sold to the market.
Illustration: three typical April days for a New South Wales data centre at 90% cover with gas engines, share of hourly load (AEST). Supply is shown renewables first, so wind and solar serve the load before batteries and gas. Operated with realistic forecasts, as described below. Across the year batteries serve about 16% of this site's load and gas about 6%, with gas rising to over half during the longest wind lull.
Checking the book with realistic forecasts
The least-cost model plans the whole year at once, so it knows exactly when every wind lull arrives. A real operator does not. We keep the model's wind and solar and re-run 2035 the way Modo's NEM forecast trades batteries. The book re-plans every hour and acts on the current hour only. It sees that hour exactly. Beyond it, prices are a smoothed six-hour shape, and wind and solar are forecasts that blur towards today's conditions over the next day or two.
- With gas: run this way, the least-cost book misses cover in 27–192 hours a year, because trading batteries sell their charge before lulls the forecast misses. Where cover fails, we test adding batteries (up to the battery ceiling) against adding gas engines, and keep the cheapest book that holds cover. Across the NEM at 90% cover this raises gas from 0.5 to 1.3 GW and adds about $4/MWh.
- Without gas: batteries are sized for firming only. While choosing the book they earn no trading income, so none are bought as market plays. Where cover fails, we scale them up with no cap until it holds, which in most cases means doubling them. At 90% cover this takes NEM storage from 27 to 54 GWh, and cover then holds in every hour in all but one case.
The charts show these realistic books. Our wind forecast is off by about 10–11% of capacity a day ahead, similar to a single wind farm's forecast; a diversified portfolio would do better. The check also uses one weather year and sizes top-ups with hindsight, so treat firm capacity as a floor.
New and existing certificates
The renewable obligation lets data centres use some certificates from existing plant at first, rising to all new. We define the new share as the share of a site's annual certificate liability that must come from generation or storage reaching financial close after the obligation starts.
Six years follows the Irish glide path the government cites. All certificates are new from 2034, so the wind and solar contracted for 2035 is all new build. Firming comes from new plant in the data centre's region, the firming request's default.
How contracts are priced
Each contract is priced to cover the asset's build and running costs plus investors' required return, set by the cost of capital, over the asset's life. Lenders size debt to the cash flows a long-term offtake provides, so the cost of capital covers both debt and equity.
- Cost of capital: 7.5% for wind and solar and 8.5% for batteries, CEPA's rates for a typical PPA. CEPA gives no rate for gas, so gas takes the battery rate.
- Build costs: CSIRO GenCost 2025-26, including interest during construction. GenCost excludes grid connection, and we leave it out for every technology so all are on the same basis, in line with Modo's wind research. Batteries use GenCost's total battery and balance-of-plant cost for each duration. Wind and gas capital is spent 50/50 over the two years before operation; solar and battery capital is spent in the year before.
- Gas: $13.25/GJ (Modo Central, 2035). Engines at GenCost's $2,022/kW, 41% efficiency, $29.4/kW a year fixed and $8.5/MWh variable costs.
- Contract vintage: covered load arrives steadily from 2028 to 2035, and each site signs its contracts when it switches on, at that year's build cost. Each state's contract prices are weighted by when its 2035 load arrived, using GenCost's cost path. Earlier, dearer builds lift wind, solar and battery prices by 4–11% against pricing everything at 2035 costs. GenCost holds gas engine costs flat, so gas is unchanged.
- Indexation and dollars: contracts are indexed to inflation, so prices are flat in real terms. All figures are real June 2024 dollars; GenCost costs are deflated by 2.1% to match.
| Technology | Contract price, covered-load average ($/kW a year) | Running cost ($/MWh) |
|---|---|---|
| Wind | 291 | – |
| Solar | 133 | – |
| Four-hour battery | 162 | – |
| Eight-hour battery | 259 | – |
| Gas engines | 246 | 125 |
What the data centre pays
This is the all-in cost of compliance. We show it two ways for the same book. With sell-back, surplus wind and solar, spare battery discharge and spare gas output earn the hourly price, and batteries earn FCAS. Without sell-back, the data centre earns nothing from the market and pays only for its purchases. Both are compared with the average spot price.
Weather, prices and load
- One consistent year: wind and solar output and hourly prices come from the same Modo dispatch run for 2035, two sites per state, adjusted for marginal loss factors. A windless evening and the price it causes arrive together.
- Hourly: the model works hour by hour across the full year, which captures every wind lull, including multi-day lulls in winter.
- Flat load: data centres run as near-constant 24/7 load, as AEMO and Transgrid model them.
- Price-taker: the contracted assets do not move market prices. At the scale of storage contracted, real trading would compress price spreads, so battery earnings and the without-gas cost are best read as a lower bound on cost.
Sources
- PM&C, Getting it right: Building AI infrastructure that works for Australia, consultation paper, September 2026
- AEMC, Advice on data centre regulatory pathways, July 2026
- Rule change request: data centre contracting obligation for firmed energy, October 2026
- Energy and Climate Change Ministerial Council communiqués, 2026
- Clean Energy Regulator, Renewable Electricity Guarantee of Origin
- NSW Government, Data Centre Guidelines
- NSW Government, data centre network consultation
- Victorian Government, Sustainable Data Centre Action Plan, August 2026
- CSIRO, GenCost 2025-26
- CEPA, WACC assumptions for AEMO Services, 2023
- Modo Energy, Data centres: how much will they increase demand in Australia?