NEM data centre demand explained: how to analyse it using Ko
Data centre demand is becoming one of the largest sources of new load in the NEM (1.6 GW by 2030). As projects connect, they add new load to the grid and increase the need for generation, storage and network capacity.
The current pipeline totals more than 26 GW of nameplate IT load across 74 projects. The opportunity sits in understanding which projects are likely to connect, where they are located, who is developing them, and how quickly they ramp once online.
You can now query Ko for our full NEM data centre pipeline. Ko can break the pipeline down by region, development status and operator type, compare nameplate capacity with probability-weighted expectations, and calculate year-by-year grid demand under different scenarios.
Ko can query the full NEM data centre pipeline
The pipeline register covers 123 project phases across all five NEM regions, including nameplate IT capacity, commissioning year, development status, operator and region.
Project status is an essential measure of projects moving ahead. Operational and under-construction projects are closest to real grid demand. Planning and enquiry-stage projects carry more uncertainty around timing and delivery.
Ko combines the project register with scenario assumptions and operator ramp-up curves to convert announced IT capacity into expected grid demand. That makes the pipeline comparable across regions, status, operator types, and scenarios.
Analyse NEM data centre demand with Ko
The examples below were produced by asking Ko questions against the NEM data centre pipeline. Ko identified the right tables, wrote the SQL and produced the written interpretation. Each chart was built from the same underlying data, and you can ask Ko the same questions directly.
Output: The timing of data centre demand depends on project status. Operational, committed and under-construction projects are closest to real grid demand, whilst planning and enquiry-stage projects carry more delivery risk.
2026 and 2027 contain the most credible near-term increase. Under-construction capacity totals 1.7 GW across the two years, with committed projects adding further confidence through 2028.
From 2028, the pipeline shifts toward planning-stage projects. Annual additions rise to 1.5-2.4 GW, but those projects are less certain.
Output: A small group of operators accounts for most planned data centre capacity. NextDC is the largest single operator in the pipeline, with 2.2 GW of nameplate IT capacity across New South Wales, Victoria and Queensland.
Microsoft and AirTrunk follow with 1.4 GW and 1.2 GW respectively, concentrated in NSW and Victoria — and AirTrunk is set to add a further 1 GW through its pending Mamre Road acquisition.
South Australia and Tasmania show a different type of demand growth. IREN’s 606 MW Bundey campus in South Australia and Firmus’s 296 MW facility in Tasmania are the only AI Cloud operators in the NEM pipeline.
Output: Operational data centre demand is much lower than connection capacity. The operating fleet holds 2,581 MW of grid connection capacity but draws 809 MW on average.
Two adjustments explain the gap. Connection capacity sits 974 MW above the 1,607 MW IT rating, reflecting headroom for cooling and redundancy. Ramp-up and under-utilisation then remove a further 798 MW.
Average grid demand is lower than IT capacity because facilities are not fully utilised on average.
Output: The 26 GW nameplate pipeline translates to 1.6 GW of expected grid demand by 2030 under the Central scenario, equivalent to 14.2 TWh of annual energy consumption.
New South Wales and Victoria account for 93% of the total across all scenarios, with New South Wales at 764 MW and Victoria at 739 MW in the Central case.
The scenario fan widens from 2028. The Low-to-Central spread is 29%, whilst the Central-to-High gap is 16%, reflecting larger downside risk than upside.
Output: NEM data centres follow an S-curve ramp from commissioning to mature load — compute is installed and activated only as chip supply lands and customer demand is locked in. The pace varies enormously by archetype, from ~5 years for AI Cloud to 14+ for legacy colocation.
Three forces set the pace:
- Demand lock-in: AI cloud operators fill capacity with pre-contracted workloads, while colocation operators wait for tenants to arrive organically.
- Chip supply: AI workloads are chip-constrained, so once GPU allocations are secured the ramp is fast.
- Phasing: large projects are built hall by hall, each with its own S-curve, so the biggest projects grow continuously over 10–15 years rather than in step changes.
Scenario sensitivity is significant — the High case compresses timelines (AI Cloud full by year 4), the Low case stretches them. Modo sees 2030–2035 as the steepest acceleration, when existing projects reach maturity and new projects begin ramping at once.
Output: Amazon is the only data centre operator with recorded battery offtake contracts in the NEM. It has nine PPAs totalling 337 MW, announced in 2026.
The contracts are split across Victoria, with six assets and 191 MW, and New South Wales, with three assets and 146 MW. All are off-site PPAs pairing remote battery sites with Amazon’s data centre consumption.
No other hyperscaler or colocation operator has recorded BESS offtakes in the NEM to date.
Try these questions with Ko
Ko can query the data centre pipeline directly. Ask a question in plain English, and Ko will select the right data, write the SQL and return results without setup.
The examples above cover project timing, operator concentration, operational demand, 2030 scenarios, ramp-up curves and battery offtakes. Ko can run the same analysis across any region, development status, operator type, scenario or time horizon in the dataset.
Starting questions:
