Sustainability as Permission to Grow
Sustainability frameworks are becoming how AI infrastructure gets built. Tasmania shows paying your own way may not settle what the electricity is for.
When a community says no to a datacentre, the capital stalls — and sustainability becomes the machinery for getting it moving again. Tasmania shows what that does to the question underneath.
What happens when a community says no to a datacentre?
In the emerging economics of AI infrastructure, one answer is becoming strangely legible:
the capital stalls.
The World Economic Forum estimates that better sustainability and resilience practices could help “unlock $700 billion to $1 trillion” in planned global datacentre investment by 2030.
Unlock.
The word matters because the Forum explains what is being unlocked. Sustainability and resilience, it says, can reduce delays and cancellations by improving community relations and helping infrastructure adapt to local resource constraints, regulation and technological change.
The scale of those delays is no abstraction. Data Center Watch counted at least 75 US datacentre projects worth about US$130 billion disrupted by local opposition in the first quarter of 2026. In the second quarter, it recorded at least another 45 projects worth nearly US$68 billion.
Electricity shortages, water scarcity, infrastructure limitations, regulation and community opposition are very different things. Inside the machinery of project delivery, however, they acquire something in common:
they can stop the project.
Sustainability becomes one means of getting it moving again.
Outside the fence
To its credit, the World Economic Forum and Oliver Wyman’s Data Centre Sustainability and Resilience playbook gets something important right.
Power Usage Effectiveness isn’t enough.
PUE tells us how much electricity entering a datacentre reaches computing equipment rather than cooling, power distribution and other facility overheads. It is useful engineering information. It does not tell us whether building a particular datacentre is environmentally or socially worthwhile.
The Forum therefore expands the boundary: electricity sources, upstream energy requirements, lifecycle impacts, water, infrastructure, regulation, resilience and communities all matter.
Then comes the revealing heading:
“Constraints on AI infrastructure growth.”
The broader system boundary is real. So is the objective against which that system is being evaluated. These constraints increasingly influence, the Forum says, “where, when and how projects can proceed”.
That does not make the framework fraudulent. A datacentre that wastes less water, works constructively with its neighbours and fits its electricity system is plainly preferable to one that doesn’t.
But it exposes a tension in what sustainability is being asked to do.
A community deciding that it does not want a development and a cooling system facing inadequate water supply are fundamentally different phenomena. In an investment-risk framework, both can nevertheless appear as impediments to project delivery requiring management.
Sustainability isn’t merely functioning as a brake.
It’s becoming the lubricant.
The older idea of sustainable development contained a different emphasis. The Brundtland definition — meeting present needs without compromising future generations’ ability to meet theirs — wasn’t hostile to development or economic growth. But it put needs and limitations inside the concept itself.
The question was, in part, what could be consumed now while preserving future possibilities.
The emerging infrastructure formulation often asks something subtly different:
How can the constraints be managed so development can proceed?
Sometimes those questions produce the same answer.
They are not the same question.
Except the rules are changing
Australia complicates this story in an important way.
The Commonwealth is already moving beyond voluntary expectations towards mandatory standards for large datacentres. National Cabinet agreed in August to develop nationally consistent requirements covering energy, water and land use, with Commonwealth legislation intended for early 2027.
The proposed regime is tiered. The Commonwealth consultation paper indicates facilities below 30 MW would sit outside the national standards. Those between 30 and 100 MW would face baseline requirements. More substantial requirements could apply above 100 MW, or to a cumulative portfolio equivalent of 150 MW.
Firmus’s Tasmanian build sits squarely in that upper territory.
The Commonwealth consultation paper captures the emerging bargain with unusual clarity.
Large datacentres, it says, “cannot drive up energy bills for Australian families and businesses, threaten scarce water resources, or be built without meaningful engagement with the communities in which they are located.”
In the very next paragraph:
“At the same time, we cannot afford to forgo data centre investment.”
There it is.
The proposed standards would require large datacentres to bring forward new renewable generation sufficient to fully offset their energy demand, backed by appropriate firming; provide demand flexibility; and operate in ways intended to minimise costs imposed on businesses and consumers.
Energy regulators are moving in parallel. The Australian Energy Market Commission has recommended requirements around new renewable energy, firm capacity and demand flexibility. A separate rule-change process opened in September to determine how datacentres and other large loads should contribute to network augmentation costs they cause or accelerate.
These reforms matter.
They increasingly say:
bring the generation.
Bring the firming.
Pay the network costs.
Be flexible when the grid needs it.
Don’t quietly send the bill elsewhere.
The physical externalities are being pulled inside the project boundary.
That is substantial progress.
But there is still another question:
What should this scarce capacity be used for?
Tasmania makes the abstraction physical
Tasmania provides an unusually clean place to ask it.
Firmus Technologies is building a 104 MW AI facility at St Leonards near Launceston. George Town Council approved its 288 MW Long Reach facility at Bell Bay in August, although that approval is now under appeal. Firmus itself is challenging a condition limiting the use of its proposed 276 diesel generators largely to emergencies, testing and maintenance.
Its proposed 52 MW Wesley Vale facility remains unapproved. Latrobe Council recently raised concerns that work occurring there may have exceeded what was permitted under an existing Forest Practices Plan. Firmus said the activities were general site maintenance and did not require planning approval, but agreed to stop while council assessed the issue.
According to Firmus’s parliamentary submission, as reported by W.Media, the three facilities reach 444 MW of combined contracted capacity at full build, consuming around 3.3 TWh each year.
That is enormous in Tasmania.
Hydro Tasmania describes the state’s electricity system as currently “in balance”. It explains what that means: in general, Tasmania presently generates about as much electricity as Tasmanian customers use.
It does not mean the island has reached an immutable physical maximum. Supply changes. Demand changes. Rainfall changes. New generation can be built.
But hundreds of megawatts of new demand change the equation.
Hydro says supplying Firmus at full scale would require additional generation. It also identifies genuine potential benefits: new renewable investment, greater certainty over Hydro’s future revenue and a broader customer base over which some future system costs might be shared.
Those benefits deserve to be taken seriously.
So does Firmus’s energy policy.
The two-for-one promise
Firmus commits to supporting at least two megawatts of new renewable generation for every megawatt of contracted capacity, together with at least 2.5 MWh of new firming per megawatt of contracted load where the relevant government energy body identifies a shortfall in dispatchable renewable capacity.
Two for one sounds comfortably greater than the load.
But megawatts measure capacity.
Electricity consumption is energy.
At 444 MW of contracted capacity, Firmus’s commitment implies at least 888 MW of renewable nameplate capacity.
The company says its three facilities would consume about 3.3 TWh annually. For 888 MW of generation to produce that much electricity in a year, the portfolio would need to average a capacity factor of about 42.4 per cent, before losses or curtailment.
That is not an absurd figure for excellent Tasmanian wind. TasNetworks treats around 40 per cent as a typical annual value for network planning purposes in Tasmania, with the best sites — the Central Highlands renewable energy zone — assessed as the highest in the National Electricity Market.
But it does mean the actual generation mix matters enormously.
A portfolio containing substantial solar behaves differently from 888 MW of Tasmania’s best wind. Firming matters. Timing matters. Curtailment matters.
And so does the meaning of support.
Does Firmus sign long-term offtake agreements without which projects would not proceed?
Provide finance?
Take construction or market risk?
Contract generation that was already likely to be built?
The Commonwealth’s proposed national standard makes this more than an accounting curiosity. It proposes an obligation ultimately based on matching energy demand with new renewable electricity, and explicitly asks how “new” renewable capacity should be defined.
Firmus’s headline promise is expressed in MW.
The proposed national test is ultimately about MWh.
Whether “two for one” delivers what people intuitively hear in that phrase therefore depends on what gets built, how much electricity it actually produces, the firming behind it and whether the generation is genuinely additional.
The harder question remains:
Would this generation exist without Firmus?
Compared with what?
There is another complication, and it cuts against any simplistic argument that Tasmania simply does not have the electricity.
Existing industrial demand may shrink.
The Liberty Bell Bay manganese smelter closed in July. It had been using about 90 MW.
Bell Bay Aluminium uses roughly 355 MW, making it Tasmania’s largest electricity customer. Negotiations over its future power arrangements remain consequential for both the smelter and Hydro.
If major industrial users leave or reduce production, datacentres could absorb released electricity, underwrite grid costs and pay commercial prices.
There is a powerful financial incentive here too.
Energy analyst Marc White estimated in July that Hydro could potentially sell electricity freed up by industrial closures for two to three times what major industrial customers had been paying.
That is an estimate, not Hydro’s published price forecast. But it exposes the economic pressure in the allocation question.
Cheap electricity supplied to an industrial customer has an opportunity cost.
So does selling that electricity to a datacentre.
So does exporting it.
So does preserving it for another future industry.
If a datacentre creates genuinely new demand, the comparison might be datacentre versus industrial electrification, transport, export, hydrogen, another industry or some future use we haven’t yet imagined.
If a smelter closes and hundreds of megawatts become available, the comparison becomes:
what should replace the smelter?
There is no easy answer.
A smelter supports hundreds of direct jobs and extensive industrial supply chains. A highly automated datacentre may employ far fewer people after construction.
But a datacentre may pay more for electricity. It might offer demand flexibility. It might strengthen Hydro’s revenue base. And if coupled to research, skills, businesses and accessible compute, it could help create economic activity well beyond the facility itself.
There is another risk: replacing several industrial customers with a small number of customers from one rapidly evolving industry can create concentration risk of its own.
This is why the useful question isn’t whether AI is “good” or smelting is “bad”.
It is:
Compared with what?
Opportunity cost changes as the system changes.
That is precisely why it has to be made explicit.
Adding hundreds of megawatts changes the machine
Tasmania’s electricity constraints aren’t theoretical.
TasNetworks has been planning major augmentation around George Town after receiving large-load connection enquiries originally associated with prospective hydrogen production at Bell Bay. It says it has also received interest from other large-load proponents, including datacentres.
At its current assessment stage, TasNetworks has identified one credible network option, costed at approximately $370.6 million. It includes a new substation, reconfiguration of the existing 220 kV switchyard and 550 MVAr of reactive support.
That is not a $370 million Firmus bill.
Firmus did not create the original hydrogen-driven planning problem.
The point is more fundamental.
Adding hundreds of megawatts of demand changes the machine.
Wires, substations, voltage stability, reactive power, contingency limits and Basslink constraints are not abstractions.
Eventually the opportunity cost becomes steel, transformers and money.
Efficiency is not sufficiency
There is an even larger assumption underneath the sustainable-AI infrastructure conversation.
Demand.
AI hardware and software have become dramatically more efficient. Yet aggregate electricity consumption keeps rising.
The International Energy Agency says global datacentre electricity demand increased about 17 per cent in 2025. Consumption from AI-focused datacentres increased 50 per cent.
Its central projection has total datacentre electricity consumption rising from roughly 485 TWh in 2025 to 950 TWh in 2030, while consumption from AI-focused facilities roughly triples.
That does not prove Jevons’ paradox.
Demand can rise for many reasons while efficiency improves.
But the IEA identifies the tension directly. Simple AI queries are becoming cheaper energetically while newer uses — including reasoning, video generation and agentic workloads — can consume hundreds or thousands of times more energy per query.
A 2026 study in Communications Earth & Environment explores the rebound problem using IEA scenarios. It compares 154 TWh of efficiency-enabled savings with 318 TWh of additional demand in a rapid-deployment pathway, producing what the authors call a Jevons Ratio of 2.06 and, within that modelled scenario, a “backfire regime”.
That is not an empirical measurement proving that the actual AI economy has crossed some universal Jevons threshold.
Its narrower lesson is more useful:
efficiency cannot be assumed to reduce total electricity consumption when computational demand expands faster than energy intensity falls.
Make a token cheaper and we don’t necessarily generate the same number of tokens with less electricity.
We may generate many more tokens.
Then more reasoning.
More video.
More agents.
More workloads that become economically possible because computation became cheaper.
Efficiency matters enormously.
Efficiency is not sufficiency.
What exactly is being unlocked?
Return to the WEF’s word.
Unlock.
Community acceptance is explicitly one of the constraints in its framework.
Tasmania already has substantial public disagreement about these projects. A September EMRS poll of 517 Tasmanians found 54 per cent opposed AI datacentres in the state, while 25 per cent supported them. Seventeen per cent were neutral and five per cent unsure.
That is one poll, not a referendum.
It cannot decide what Tasmania should do.
But nor does it describe a state neatly divided down the middle.
This makes “social licence” a more demanding concept than public relations.
Consultation can improve projects. It can change designs, conditions, locations and community benefits. It can expose assumptions that would otherwise survive unnoticed.
But if the outcome that something should not proceed has disappeared before consultation begins, the process is no longer deciding whether consent exists.
It’s customer support.
So the WEF framing leaves an awkward question:
What does it mean to “unlock” investment when one of the locks is community opposition?
Where does the veto live?
Tasmania exposes a peculiar institutional problem.
When George Town Council considered Firmus’s Bell Bay project, Mayor Greg Kieser described the limit of its role bluntly:
Then:
“policy is just purely not our remit.”
Council planners similarly said broader questions involving energy policy, AI, national security and Commonwealth regulation were either addressed through other statutory systems or fell outside the planning authority’s role.
Hydro Tasmania occupies a different position.
It is not simply a private electricity retailer. Hydro is a state-owned Government Business Enterprise. Its corporate planning and governance are agreed with, and accountable to, the Tasmanian Government and its shareholder ministers, while Hydro itself operates commercially under a Ministerial Charter that sets wider public-interest and energy-security expectations.
So it would be wrong to say that nobody decides how Tasmania’s electricity is allocated.
Decisions plainly are made.
The more interesting question is how the comparison becomes visible.
A major electricity contract can allocate hundreds of megawatts through a mixture of commercial assessment, shareholder governance, energy policy and infrastructure decisions without any single step requiring a public comparison between that use and plausible alternatives.
TasNetworks asks whether the network can safely accommodate the load and what augmentation is required.
Planning authorities ask whether a development satisfies the planning scheme.
Hydro asks whether supply makes commercial sense within its obligations.
Ministers set policy and shareholder expectations.
The emerging national standards ask whether the load brings enough generation and firming, operates flexibly and pays the infrastructure costs it creates.
Every one of those questions is legitimate.
But none is quite:
Of the competing things Tasmania could do with hundreds of megawatts of low-carbon electricity and the infrastructure that carries it, why is this use preferable?
The allocation decision has a home in government.
What it does not necessarily have is an explicit comparative test.
A sequence of locally rational decisions can produce a system-level allocation without that allocation ever being publicly justified as such.
The planning application passes.
The network can be augmented.
The electricity contract makes commercial sense.
The proponent internalises its physical costs.
Every gate can work properly while the larger choice emerges from their combination.
That is the deeper question behind the veto.
Not whether an omnipotent official should be able to kill otherwise lawful projects on a whim, but whether the decision system retains an explicit capacity to conclude:
paying your costs is necessary, but it may not be sufficient.
What is the electricity for?
There is a corresponding question on the benefit side.
In its submission to Tasmania’s parliamentary inquiry into AI datacentres, the UNSW AI Institute warned that datacentres by themselves do not create a local AI industry or guarantee lasting community benefit.
Its director, Sue Keay, wrote:
“Tasmania will hold the strongest negotiating position before facilities are approved, not after.”
Her proposals give the idea of a public return some concrete form: a levy on large datacentre developments to support Tasmanian AI research, talent and capability; compute reserved for researchers, universities, startups, small and medium businesses and not-for-profits; workforce pathways through TAFE, universities, apprenticeships and PhD placements; and obligations structured so they reach the tenants actually controlling the compute.
As Keay puts it:
“The value in AI lies in compute, data, models and skills, not in the building.”
That suggests a useful metric to put beside PUE, water consumption and grid impact:
What is the social return on compute?
Not merely how much harm has been mitigated.
Not merely whether the facility pays the bill it creates.
What durable capability remains because Tasmania chose this use of its resources?
One way to describe the two unresolved questions is veto and dividend.
The veto is the capacity to conclude that a project which pays its physical costs may still not be the preferred use of scarce capacity.
The dividend is whatever enduring public value remains when the construction cranes leave and today’s GPUs are obsolete.
Neither concept predetermines the answer.
Tasmania may turn out to be an excellent place for substantial AI infrastructure. Its renewable electricity system, cool climate and hydro storage are genuine advantages. A large, flexible commercial customer could help support new generation, replace lost industrial demand and strengthen parts of the electricity system. AI itself may generate substantial economic and social value.
But those possibilities do not eliminate the comparison.
They make it more consequential.
The Tasmanian Government’s draft Expectations for Data Centres and AI Infrastructure now asks proponents to support additional renewable generation, meet their share of network and connection costs, protect energy security and grid stability, and demonstrate benefits for Tasmania.
That consultation remains open until 12 October 2026.
The Commonwealth consultation on proposed mandatory national standards remains open until 9 October.
The architecture is not finished.
It is being written now.
The policy machinery is becoming much better at answering:
How should a datacentre pay for the physical demands it places on the system?
The less settled question is:
What public return is sufficient to justify choosing this use over another?
PUE cannot answer that.
A renewable certificate cannot answer it.
A planning scheme designed around an individual development cannot answer it by itself.
Nor can an electricity supplier’s commercial assessment.
Tasmania has water, wind, wires, public electricity infrastructure and a growing queue of industries interested in using them.
A sustainability framework can tell us how efficiently those things can be converted into computation.
An investment framework can tell us whether doing so makes money.
Neither, by itself, can decide what those resources are for.
That is the decision hiding underneath all the others.