Australia priced a 5 percent tax on big data centers. The public bill needs a meter

THE SIGNAL IN ONE SENTENCE

Australia now has a number for one way to make large data centers pay into the public ledger. At the request of independent ACT Senator David Pocock, the Australian Parliamentary Budget Office costed a proposal for a 5 percent tax on revenue generated by individual data centers with more than 5 megawatts of energized information-technology load. The office completed the costing on October 20 and authorized it for public release on October 21. This is not a tax law. Parliament has not enacted it. The costing is a model of a proposal, built from assumptions that the budget office repeatedly labels uncertain. That distinction is the front door to the story, not a footnote. The proposal would start on January 1, 2027. Colocation facilities, which sell customers space and computing capacity, would be taxed on their revenue. Owner-operated hyperscale facilities are trickier because a cloud company may use a data center to support revenue earned elsewhere. The proposal would therefore assign those facilities a deemed revenue rate of A$236 per kilowatt per month, indexed to inflation. The Parliamentary Budget Office estimates that the measure would improve the fiscal and underlying cash balances by about A$404 million through 2029-30, after administration costs. Its central case projects A$3.046 billion in tax revenue through 2036-37. A higher data-center growth scenario pushes that cumulative revenue estimate to A$4.125 billion. Those are tidy figures attached to a rather untidy market. The office says information about Australian data-center revenue, customer mix and future development is limited. It uses aggregate colocation data, estimated energized IT load and several simplifying assumptions. It also warns that actual results at individual facilities could differ materially. In other words, Australia has a proposed tax formula before it has a complete public meter. The plain signal is that a data-center levy should arrive with a facility-level receipt for the public costs and assumptions it is meant to address. The tax idea is not merely about grabbing a slice of fashionable technology revenue. AI data centers are physical industrial sites. They need power, cooling, transmission, land, backup systems and local services. Their owners may receive planning approvals, network connections, tax treatment and other public support. A revenue levy can be one way to return some value to the public. But a percentage by itself cannot tell voters whether the bargain is fair. Begin with the threshold. Five megawatts is large compared with an ordinary commercial building, but it is not a universal definition of a hyperscale AI facility. The proposal assumes that 85 percent of facilities are above that threshold, rising to 95 percent over the medium term as larger projects enter the pipeline. That assumption should be testable. A public register should list every in-scope facility, its owner, operator, location, energized IT load, commissioning date and whether it is a colocation or owner-occupied site. Facilities that expand across the line should enter the register. Facilities split across adjacent buildings should not become mysteriously small on paper. Then measure the tax base. For a colocation operator, audited revenue may provide a relatively direct starting point. For an owner-occupied cloud campus, deemed revenue is an administrative invention. It is necessary because the facility supports services sold through a wider business, but it is still a proxy. The proposed A$236-per-kilowatt monthly rate is based on the ratio of total revenue to billed capacity at a market-leading colocation provider. That may be a workable benchmark. It may also age badly as hardware density, utilization, pricing, energy pass-through charges and customer contracts change. Publish the annual calculation. Show the energized capacity, utilization assumption, deemed rate, inflation adjustment, exemptions, tax due and any correction made after audit. The public does not need a customer's private workload. It does need to see whether a giant owner-occupied campus is being valued by a rule that still resembles the market. Electricity belongs on the same receipt. The budget office uses a central projection in which energized IT load rises from 0.5 gigawatts in 2026-27 to 2.6 gigawatts in 2036-37. Its high case reaches 5.1 gigawatts by 2036-37. That gap is almost the size of the entire central-case industry at the end of the period. This is why the revenue range widens. More capacity creates a larger tax base, but it also changes the public infrastructure problem. A 5 percent revenue tax does not automatically cover new transmission, generation, local substations or the risk of underused assets if a project is cancelled. For every large site, publish contracted demand, actual peak demand, annual electricity use, grid upgrades attributed to the connection, financial security posted by the customer and which party pays if the load arrives late or not at all. Separate electricity bought under contracts from new generation that was actually built because of the project. The same logic applies to water. Dry cooling, recycled water and low-water designs can reduce demand, but the relevant number is facility use in the local climate. Put annual withdrawals, consumption, source, discharge and peak-day demand next to the power record. Now follow the cost downstream. The Parliamentary Budget Office says it has not quantified whether data-center operators would pass the levy to customers. Some almost certainly will try. A cloud invoice may rise. A colocation contract may add a surcharge. An Australian software company could pay part of a tax aimed at a multinational facility. That does not automatically make the levy a bad idea. Taxes often affect prices. It does mean the policy needs a distribution test. Track how operators describe the charge, how much is passed through, which customers face it and whether the tax changes investment or contracting behavior. A provider should not be able to present a broad price increase as a government-mandated amount if the actual levy is smaller or calculated differently. Revenue needs a destination too. The current costing estimates money entering the budget. It does not earmark that revenue for grid upgrades, water resilience, community benefits, worker training or efficiency programs. General revenue is a legitimate policy choice, but it should be stated plainly. Calling a tax an infrastructure response while sending every dollar into an unlabelled pool is how public arguments become allergic to arithmetic. If lawmakers choose dedicated spending, publish allocations and outcomes. If they choose general revenue, say so and judge the tax on fairness, efficiency and administration rather than implying a direct local refund. Finally, keep the uncertainty visible after the press release. The office assumes Australian colocation revenue grows 13 percent a year over the medium term. It assumes the hyperscale share rises from 42 percent of the market to 65 percent by 2035. It assumes owner-occupied facilities rise from 30 percent to 50 percent of hyperscale IT load. Each assumption can move the tax estimate. Create an annual reconciliation that compares forecast with reality. Show expected and actual energized load, taxable facilities, owner-occupied share, revenue, deemed revenue, tax collected, administration cost, pass-through evidence and appeals. Revise the forecast when the pipeline changes. That ledger would do more than improve one proposal. It would give Australia a factual base for the next fight over AI infrastructure, whether the question is tax, power, water, jobs or public support. There is something refreshing about seeing an AI policy argument reduced to thresholds and budget tables. The costing is specific enough to disagree with. Now it needs the evidence system that lets the public check whether the formula survives contact with actual buildings. A 5 percent levy is a policy choice. A public meter is basic governance. Australia should not ask communities to accept the first without building the second.