A new report commissioned by Amazon Web Services and prepared by Frontier Economics concludes that Australia’s National Electricity Market (NEM) charging framework is well-designed to ensure data centres bear the network and wholesale costs they create, shielding residential consumers from cross-subsidies.
Available through Data Centres Australia, the report is a credible piece of regulatory economics and a useful reference for anyone navigating the NEM’s charging architecture. But a closer reading suggests it leaves several questions unanswered – questions that are directly relevant to where the industry and its regulators go next.
Where the report earns its credibility
On the mechanics of cost recovery, the report is solid. It accurately describes how distribution network service providers apply capital contribution frameworks and cost-revenue tests to large negotiated connections, requiring data centres to cover incremental connection and upstream augmentation costs that their ongoing network charges won’t recover.
It correctly explains that transmission charging – through exit service charges and maximum-demand-based locational tariffs – means large sustained loads pay a materially higher share of shared network costs than households, whose contribution is diluted through aggregation at distribution connection points.
The report also points out that major hyperscalers including Amazon, Microsoft, Google and Equinix have become significant counterparties for renewable Power Purchase Agreements in Australia, providing revenue certainty for new generation projects and genuinely supporting the investment pipeline.
Its overall conclusion, that the regulatory framework is intended to prevent data-centre-related network costs from being shifted onto residential bills, is broadly accurate as a description of how the rules are designed.
What the framework doesn’t resolve
However, the burning question – and the one the report does not fully engage with – is whether cost recovery adequacy and system readiness are the same thing. They are not, and it is a distinction that runs through much of AEMO’s recent thinking on large load growth.
The report’s analysis is largely static and average in its framing as it addresses whether data centres pay their share. AEMO’s planning concerns are increasingly dynamic and locational: for example, whether 300 MW appears in Western Sydney, whether 500 MW clusters around the same zone substation in Melbourne’s west, and whether transmission augmentation is ready when the load arrives. Paying the right tariff does not resolve a congestion event that occurs before the augmentation is complete.
On wholesale prices, the report applies standard economic reasoning – new supply will follow new demand, higher prices will incentivise entry, and grid-scale batteries can respond relatively quickly. That is theoretically sound in long-run equilibrium. But AEMO’s practical experience across the transition has been that transmission arrives late, renewable project timelines slip, and supply and demand growth rarely synchronise as cleanly as the model implies. The transition risk is real precisely during the adjustment period, not after equilibrium is restored.
There is also an important distinction between the report’s treatment of network costs and wholesale costs. On network charging, Frontier walks through the relevant regulatory mechanisms in considerable detail, explaining how capital contributions, connection charges and transmission tariffs allocate costs to large users. The wholesale argument is necessarily less definitive. Rather than demonstrating that data centres fully internalise any wholesale market impacts they create, the report argues that wholesale price pressures need not emerge if new generation, storage and firming capacity enter alongside demand growth.
That may well prove true, but it depends on future investment occurring at the right scale and pace. In effect, the report establishes cost causation and recovery for networks far more directly than it establishes the same proposition for wholesale electricity markets.
The report places emphasis on installed capacity growth without fully exploring differences in capacity value and firming requirements. Counting 130 GW of committed and anticipated generation entry to 2035 as evidence of system headroom is meaningful only if that capacity is firm when it is needed. A 500 MW data centre requires energy in every dispatch interval. A 500 MW solar farm does not deliver that on a winter evening.
The renewable spill argument – that data centres can absorb curtailed energy and strengthen the investment case for new renewables – has merit in principle, but the load profile question is left unexamined. Most hyperscale facilities operate around the clock. Peak system stress typically occurs at precisely the moments when renewable output is lowest. The report presents only one side of that dynamic.
Finally, and perhaps most notably, the report is almost entirely silent on AI training loads, gigawatt-scale cluster demand, and the load forecasting challenges those present. These are the scenarios AEMO has been grappling with openly. Treating data centres as just another large industrial load may have been reasonable five years ago. Many planners no longer see them that way.
The debate that still needs to happen
None of this is to say the report’s core regulatory findings haven’t added to the debate. The NEM’s charging frameworks are broadly well-structured for cost recovery, and data centres are, by design, not free riders on the network. But rapid and geographically concentrated demand growth still creates planning, reliability and timing challenges that cost recovery rules alone cannot resolve.
While the framework tells us who pays, it doesn’t tell us whether the infrastructure will be ready, in the right place, at the right time. That is where the harder work lies, and it is a conversation the industry, AEMO and regulators will need to keep having more and more regularly.