Tasmania's AI bargain · a reading of the public record November 2026 · working slice

The power · the promises · the missing price

Tasmania's AI bargain.

A new electricity customer of extraordinary scale has asked the state to make room for it.

The investment is real. So is the opportunity. What is not in the public record is the bargain itself: what Tasmania has agreed to, what it will receive, and who checked whether the deal was worth doing.

444MW Proposed Three sites, one proponent, no binding commitment to build.
104MW Under an electricity contract Three years. Price undisclosed. A cap, not a reading.
? Publicly priced net return Not a claim that the return is zero. A statement that no assessment is published.
One

The size of the ask, and what has actually been agreed

Four hundred and forty-four megawatts is announced. One hundred and four is contracted. Nothing is metered.

Firmus Technologies has proposed up to 444 MW of electricity demand across three Tasmanian sites: St Leonards in Launceston, Long Reach at George Town, and Wesley Vale. Firmus has proposed approximately 444 MW of future demand, Hydro Tasmania told a parliamentary inquiry, naming the three sites.

Scale is easy to lose in a number that large. A submission to the parliamentary inquiry put the request at as much as 30% of Tasmania's current electricity use — and attached that comparison to 450 MW, one of three totals the same portfolio carries in the record. The state's supplier puts the number more carefully: an additional load of up to 444 MW would represent a material increase in demand.

This is not a story in which nothing is happening. A council approved a 288 MW campus six votes to one. A 26 MW building was approved by staff delegation, unadvertised, without a vote. About ninety contractors a day were working on the Launceston site in May. Land is being prepared; rates are being paid; substations are being upgraded.

But a proposal is not a commitment, and an approval is not a contract. The distinction is the whole subject. In the documents, three different things are routinely described with one number, and the difference between them is the difference between a press release and a legal obligation.

What has actually been agreed, site by site
Site Proposed Planning approved Electricity contracted Metered
St LeonardsLaunceston 104 MW Existing site
Connection upgrade in progress
104 MW Not published
Long ReachGeorge Town 288 MW 288 MW None Nothing built
Wesley ValeTwo 26 MW facilities 52 MW 26 MW by delegation
26 MW referred upward
None Nothing built

In force Approved by staff delegation Referred upward, not decided No instrument

Read the columns, not across them. “Contracted” means an executed instrument fixes the number — here, a three-year retail agreement with Aurora Energy, capped at 52 MW from 1 August 2026 and 104 MW from 1 December 2026. Across the three sites: 444 MW proposed; 314 MW planning-approved, 26 MW of it by staff delegation; 26 MW referred upward; 104 MW contracted. Nothing has been metered anywhere: at Long Reach and Wesley Vale no facility exists to meter, and at St Leonards load has been permitted since 1 August 2026 with no reading published. The last column records absence of evidence, not evidence of absence. Sources: Hydro Tasmania; George Town Council; the Tasmanian planning record; Hydro's released board papers.

One hundred and four megawatts is 23 per cent of the proposal. It is also the only part of the plan where money changes hands on terms anyone has agreed to — and even those terms are largely withheld. The price is redacted. The value of the hedge behind it is redacted. The premium the customer pays for refusing a demand-reduction notice is redacted.

Which leaves the question the rest of this piece keeps returning to. Four hundred and forty-four megawatts of demand is being planned around. What has the state been promised in return, and where is the document that says so?

Two · Chapter one

The business case nobody can see

A $10 million threshold, a rule that requires a business case, and a released page that shows the rule being set aside without showing the decision.

On 6 March 2026 the chair of TasNetworks wrote to the state's shareholding ministers about three connection enquiries from Firmus. The first was modest in the telling and large in the building: an upgrade of the existing St Leonards connection from 20 MVA to 104 MVA — from the capacity of a small industrial site to the capacity of a city block.

By the time Treasury wrote to the Treasurer, the works had been classified. They were a “major capital investment” under the guidelines that govern state-owned businesses: above the $10 million threshold, and involving the private sector. That classification matters, because for investments of that size the guidelines require supporting documentation — including a business case, the document that asks whether the thing should be built at all.

The minute records what happened next in one sentence. It is the sentence this chapter exists for.

Transcription of the released page · typeset here, not a facsimile the document's own withholding the editor's note
Department of Treasury and Finance Minute to the Treasurer
Critical Date: No Critical Date

Subject: Firmus Technologies Pty Ltd connection enquiry – St Leonards expansion

Approved / Not approved / Noted  ·  Treasurer comments
Approved
Not approved
Noted

Editor's note — No tick, no comment and no date appear in the released copy. That is a fact about this page, not evidence that no decision was made.

Recommendation

It is recommended that you:

s35

Background

Firmus Technologies Pty Ltd is progressing a data centre expansion project in partnership with Nvidia and CDC Data Centres. On 6 March 2026, the Chair of TasNetworks, Mr Roger Gill, wrote to Shareholding Ministers regarding three transmission connection enquiries submitted by Firmus.

The first project proposed by Firmus is an upgrade of its existing St Leonards connection from 20 MVA to 104 MVA.

Treasury comments

  1. The St Leonards connection upgrade project is a “major capital investment” under the Guidelines for Tasmanian Government Businesses: Capital Investments as it has a total value in excess of TasNetworks' major capital investment threshold ($10 million) and it involves the private sector.
  2. For major capital investments, businesses must provide Shareholding Ministers with supporting documentation to justify the investment, including the business case. It is noted that TasNetworks has not provided a business case.
  3. Under the Guidelines, a Government business may request approval for an exception to the requirement to submit a business case. Exceptions must be approved by the Treasurer. To seek an exception, a request must be submitted to Treasury. The burden for demonstrating circumstances that justify an exception to the Guidelines sits with the business.
  4. The National Electricity Rules set out TasNetworks' roles and responsibilities for investment in transmission connection assets. Under the NER, connection assets are not regulated services and the load proponent is required to pay the full cost of the investment.

List of attachments

1. out of scope
2. Letter from Shareholding Ministers to Chair of TasNetworks – Firmus Technologies Pty Ltd connection enquiry - St Leonards expansion

Document reference: 26/61125 Contact officer: Director, Shareholder Policy and Markets Branch
An excerpt, not the whole document. Reproduced from the Treasury minute released under right-to-information request 2526-023, six pages, obtained by this investigation in full. The black bars are the document's own withholding, not an editor's. The rust mark under one sentence is the editor's — the only mark on this page that is. The question it raises is not about any individual. It is about a decision path that left no public trace.

Three separate things are in play in this minute, and it rewards keeping them apart. Collapse them and the document says something it does not say.

The requirement
A major capital investment above $10 million must be supported by documentation including a business case, provided to shareholding ministers. Treasury records that none was provided. That is a statement about paperwork — not a finding that the project was unwise, and not an approval given in the absence of analysis.
The exception
The guidelines allow a business to request an exception, and an exception must be approved by the Treasurer. The burden of justification sits with the business, and an exception may be considered appropriate where the investment is already subject to another robust planning and reporting regime. Nothing in the released material shows whether an exception was requested.
The decision
Not in the record. The recommendation is withheld under s35. The joint letter from the Treasurer and the Acting Energy Minister to TasNetworks' chair is listed as attachment two and released as out of scope. The decision row carries no mark. Whether the Treasurer approved an exception, refused one, or never considered the request is not established by anything this investigation has seen.

What remains is a narrow sentence that survives all three readings. A public authority asked its shareholder minister a question about whether to build something, and the answer is not in the file.

“Exceptions must be approved by the Treasurer.”

Treasury minute to the Treasurer, c. March 2026

What the documents establish is narrow, and it needs to be stated narrowly. A major investment crossed a governance threshold without its normal supporting analysis among the papers before the decision-maker. Whether an exception was sought, granted, refused, or never decided is not established by any document this investigation has seen.

What the investigation will not say — and once did, and withdrew — is that the Treasurer approved the investment without a business case. That is the claim the missing record would settle, which is precisely why it cannot be asserted in its absence.

The sequence, as far as the record shows
Knew Said publicly Committed Scrutinised
6 Mar 2026 TasNetworks chair writes to ministers
c. Mar 2026 Treasury minute: threshold crossed, no business case
c. Mar 2026 ?The Treasurer's decision
29 Oct 2026 Committee: modelling must be updated

Knew is what a department, minister or state business demonstrably knew. Said publicly records public assurance statements — not diligence performed in private, which is why this lane is empty: nothing was said publicly about this decision either way. Committed is a decision, approval or contract that bound someone. Scrutinised is independent scrutiny identifying an unresolved question. The dashed marks a decision the public record does not contain. It sits in the Committed lane to show where such a decision would appear, not to imply that one was made. Scrutiny is not always afterwards: the committee examining these matters was established in July 2024, before the commitments it later tested.

There is a twist here that matters, because it cuts against the suspicion the document invites. The same Treasury comment that records the missing business case also records a protection: connection assets are not regulated services, and the proponent pays the full cost of the investment. Households do not pay for the wire to the data centre. The rules were designed for exactly this, and on this point they work.

So the governance question is not who pays for the wire. It is whether anyone asked whether the wire was worth building — and whether the answer, if it was given, can be read by the people whose electricity system is being rearranged around it.

What would settle it

Three records, none of them sealed by law, all held by the Executive:

The Treasurer's signed response to this minute — the s35 redaction on page one covers the recommendation, not the decision, and a returned copy would show which box was marked. The joint letter from the Treasurer and the Acting Energy Minister to TasNetworks' chair, listed as attachment two and released as “out of scope”. And any exception instrument, if one exists, which would name the circumstances that justified it.

Until one of those surfaces, the honest description of this decision is a blank: a public authority asked a question of its shareholder minister, and the answer is not in the file.

Three · Chapter two

The factory arrives before the wind

In June 2026 the state's own supplier put in writing what the years 2028 to 2030 would look like. The load is expected to be there. The generation is not.

On 10 June 2026 the chief executive of Hydro Tasmania wrote to Firmus. The subject was the price of a long-term electricity supply for the company's Tasmanian facilities. It was not an offer. It was, in the letter's own framing, a progress update on the commercial, technical and risk work underway — sent before any numbers, because the numbers depended on risks that were still being worked out.

That order of operations is reasonable, and worth stating plainly before anything else: Hydro declined to price a decade-long obligation to a company whose demand, financing and construction schedule were all still moving. What it did instead was describe, in specific terms, the window in which the system would be under strain — and the arrangements it was considering to get through it.

From the letter · 10 June 2026

“The period of greatest focus sits between 2028 and 2030, when the full Firmus load is expected to be online but before new wind generation is commissioned. During this window, the Tasmanian system's energy balance is tightest.” Page 2

“Under normal - high inflow conditions, Hydro Tasmania's existing portfolio — supplemented by gas generation and Basslink imports — can accommodate the Firmus load. The risk in this scenario is manageable.” Page 2

“Under extended low inflow conditions or limitations on Basslink, energy availability may be constrained.” Page 2

Those three sentences are the chapter. They say that the state's electricity system expects, for a period of up to two years, to serve a very large new customer with generation that does not yet exist — and that whether this works depends on rain and on a cable under Bass Strait.

Hydro's proposed way through was not to delay the whole commitment. It described a structure in which the second tranche of load is released on a conditional basis - triggered by availability of Basslink and critical plant, and the commissioning status of new generation. The first tranche would be supplied on a firmer basis; the second would wait on the things that make it safe.

The bridging fuel is named without embarrassment: the Tamar Valley combined cycle gas turbine. Hydro set out what it still had to work out — the volume and duration of gas running, whether gas would be available and at what price through 2030, and the emissions consequences — and said where those costs land: These costs will be reflected in the pricing.

Editor's note — A widely repeated version of this letter says Hydro could supply “half” the load unconditionally. That formulation is the ABC's, reporting the letter in October; the letter as released speaks of a “second tranche” released conditionally and does not divide the load in half. Both claims point at the same arrangement. They are not the same sentence, and the weaker one is the one the document supports.

Load, as described 1 2 3
2028–2030 · full load expected, wind not yet built
New wind 4
  1. 1August 2026 — 52 MW permitted under the retail contract. S83, term sheet
  2. 2December 2026 — the cap rises to 104 MW. S83, term sheet
  3. 3Mid-2028 — Firmus's stated target of ~400 MW load. S84, p. 1 — a target, not a contract
  4. 42030 — new wind not available before this date. S57, the ABC's report of the Chief Executive
  • Aug 2026 · 152 MW permitted under the Aurora contract
  • Dec 2026 · 2104 MW cap under the same contract
  • Mid-2028 · 3~400 MW load, Firmus's stated target, not a contract
  • 2028–2030Full load expected, new wind not yet commissioned
  • 2030 · 4New wind not before this date, per the ABC's report
A schematic of what the documents say, not a forecast. Positions are the dates the sources state; nothing is interpolated between them. The hatched band is the window Hydro named in June — the load expected online, the generation not yet built — with dashed guides dropping to the wind track. Filled pins mark dates the documents state; the dashed pin marks a target rather than a contract. Every pin carries its own source.

The delivery problem is not only weather. Hydro identified three prospective wind farms that could supply the renewable energy behind the arrangement, and said of each that it was well advanced in its development approvals and connection agreements. Then the qualification: all three want the same heavy-lift cranes, the same civil contractors and the same connection works, and Simultaneous construction is not achievable.

That is an unusually honest paragraph for a document of this kind. It says the portfolio is not three independent bets that diversify risk; it is one construction queue. The timing of the whole arrangement therefore depends on a sequencing decision that had not been made.

Hydro expected

  • A staged ramp: about 220 MW average by late 2027, ~400 MW by mid-2028.
  • The full load online between 2028 and 2030.
  • The system's energy balance to be tightest in exactly that window.

Conditionally manageable

  • Normal-to-high inflows: existing portfolio, gas and Basslink imports can carry the load.
  • A second tranche released only as Basslink, plant and new generation allow.
  • Firming from the Gordon and Poatina storages once renewables are running.
  • Load relief from the customer in defined risk scenarios.

Commercially agreed

  • 104 MW, capped, staged, through the Aurora retail agreement — three years.
  • Nothing at 444 MW: no contract, no price, no terms.
  • The hedge behind the retail deal, on terms that remain redacted.

Still unknown

  • The price, and the cost of the bridging period inside it.
  • Whether the formal supply proposition promised for July was made, and on what terms.
  • Whether any load has been consumed at St Leonards since August 2026.
  • What Hydro's hydrological modelling concluded about acceptable thresholds.

The case for Hydro, at its strongest

The documents here do not show a supplier gambling with the state's system. They show one declining to promise what it could not deliver, and saying so in writing.

  1. It answered the question it was asked. A customer requested 444 MW; Hydro assessed the system against that request, named the binding constraint, and set out its reasoning to the customer before pricing anything.
  2. It held the condition that matters. Two megawatts of new renewable capacity for every megawatt of load is the basis on which Firmus asked it to contract — a real requirement, not a slogan, and the mechanism by which new generation gets built at all.
  3. It sequenced rather than refused. Releasing the second tranche only when Basslink, plant and new generation allow is the prudent version of the same commitment — and it preserves the option to supply more if the system can take it.
  4. Bridging with gas and imports is ordinary practice, not a scandal. The Tasmanian system already leans on Tamar Valley gas and Basslink imports at the margins; the question is the price and the emissions accounting, both of which Hydro says it is quantifying.
  5. It took the decision to its board. A formal supply proposition, with indicative pricing and conditions, was to follow Board consideration — the opposite of a commitment made by press release.

None of that is contradicted by the record. What the record does not contain is the answer to the question the counter-case makes unavoidable: if the bridge is real, what does crossing it cost, and who pays if the wind arrives late?

Hydro said it would take a formal supply proposition to its Board in June, and be in a position to present one — with indicative pricing, contract structure and conditions — in July. Whether that happened is not in the public record. The 444 MW decision is now expected by the end of the year.

So the position is this. The state's supplier has set out, in writing, a two-year window in which a very large load is expected and the generation behind it is not yet built; a bridging plan whose cost it says will be in the price; and a set of conditions it considers sufficient. It has not said what the bridge costs. The customer has not said what it will consume. And the state has not published an assessment of whether the arrangement, taken as a whole, is worth having.

Four · Chapter three

What the machines were for

Three decisions were made about the same 276 diesel generators. Each rested on a different idea of what they would do.

On 25 August 2026 George Town Council approved a 288 MW data centre at Long Reach, six votes to one, after residents raised concerns about noise and emissions. The approval carried fifteen conditions. One of them was about the backup power: 276 diesel generators, standing on the site behind the halls.

Condition 13 is the only public document that states what those machines may do. It permits them during an unscheduled failure or interruption of the mains supply, during routine inspection, testing and maintenance, or in response to a genuine emergency. Then it says something plainer than any of that:

“The backup diesel generators must not be operated to provide electricity during normal operations, or to augment the mains electricity supply.”

George Town Council, condition 13, 25 August 2026

That is the planning permission. Two other decisions were made about the same fleet, by different bodies, on different assumptions — one before the approval, one after it, and one that is still being decided. The question this chapter turns on is narrow: were they about the same machines?

276diesel generators, as approved

One square, one machine. The public record does not state their individual or combined capacity, so this shows a count and nothing else — not megawatts, not fuel, not emissions, not run hours.

Installed is not the same as operated. The council approved the presence of 276 machines and a narrow envelope for running them; testing is confined to weekday business hours and to no more than 24 hours a year of simultaneous running. Anything said about emissions from this site is a statement about a future operating pattern, not about steel on the ground.

Three decisions, one fleet

Environmental

The EPA's gateway

Before approval · 2026

The Environment Protection Authority decided the proposal did not require environmental assessment. It understood the generators were for just for emergency use in the event of a power failure, and not for normal operation beyond scheduled weekly runs of under two hours, annual load tests and utility outage events.

Implies: a fleet that is almost never run.

Planning

Condition 13

25 August 2026

The council codified roughly that understanding, and went further: a list of occasions, a clock on testing, a reporting duty, and a prohibition on augment the mains electricity supply.

Implies: a fleet bounded by what the grid cannot supply.

Commercial

The hedge's assumption

March 2026

Hydro's board paper describes the contract it was approving: It also includes a clause that incentivises Firmus to switch to onsite co-gen options in response to Lack of Reserve (LoR) notification from AEMO or trip of Basslink.

Implies: a fleet that runs on instruction.

Three different kinds of instrument, deliberately set side by side: a regulator's gateway, a planning condition and a board's commercial assumption. They are compared because they describe the same machines, not because they carry the same legal weight.

Set the dates beside each other and the first thing to notice is order of operations. The commercial assumption is five months older than the planning condition it would have to live under. The board paper was written in March, when there was no Long Reach approval and no condition 13; the council decided in August, on an environmental framing that the earlier document does not share.

The states a generator can be in

“Emergency only” and “runs on instruction” sound like the same thing if you do not separate the states. They are not the same thing. Six are in play here, and only some of them are permitted.

Emergency operation permitted
The mains supply fails, or there is a threat to safety. The condition names it expressly. It is also what the EPA was told to expect.
Scheduled testing permitted, bounded
Weekly exercise and annual load tests. Weekday business hours only; no more than 24 hours a year of simultaneous running; usage reported. This is the largest permitted use — and it is for maintenance, not supply.
Loss of reserve not permitted
AEMO forecasts that Tasmania's operating reserve will fall below its threshold. The grid is intact; the margin is not. Condition 13 does not name it, and it is not an interruption of supply. Contract trigger: the hedge does name it.
Demand response not permitted
Hydro issues a notice to Aurora to communicate a desired demand reduction; Firmus can comply by supplying itself. The condition is silent on the mechanism. Contract trigger: a premium applies if Firmus does not comply.
Market stress under appeal
High prices or network constraints while supply is available. Condition 13 does not permit it. This is what Firmus's appeal seeks: augmentation to offset curtailment, manage energy security and maintain operational reliability.
Routine generation prohibited
Running the machines to displace grid electricity in normal operation. The condition forbids it in terms.

Badges record the planning condition only. Where a row reads contract trigger, the commercial instrument reaches an event the condition does not — that is the mismatch, not a judgement about whether the event is real.

Two of the six states are contractual, not planning permissions. That is the whole difficulty in one line: the contract can be triggered by an event the planning condition does not recognise.

The sequence, and what each party knew

Each entry states what the document shows the decision-maker had in front of them at the time — not what was in anyone's mind. Where that is an inference, the word “knew” is doing less work than it appears to.

  1. March 2026

    Hydro Tasmania's board approves the hedge

    The paper states the demand-reduction mechanism and the diesel substitution it relies on: Firmus can switch from taking energy from the grid to using onsite co-generation in the form of diesel gensets when instructed by Hydro Tasmania. At this point, no Long Reach approval exists.

  2. Before approval · 2026

    The EPA decides no assessment is required

    On the basis that the machines are for emergency use, weekly exercise and annual tests.

  3. 25 August 2026

    George Town Council imposes condition 13

    Permitted occasions, testing windows, a reporting duty and the augmentation prohibition — the narrowest of the three descriptions.

  4. 10 September 2026

    Firmus appeals, seeking a wider envelope

    It asks the Tribunal to vary the condition to allow limited augmentation of the mains electricity supply to offset curtailment, manage energy security and maintain operational reliability during periods of energy market stress.

  5. 7 October 2026

    The EPA board applies to join the appeal

    It says the variation, had it been in the original proposal, would likely have resulted in the Board making a different decision in relation to whether to assess the proposal.

  6. 10 November 2026

    Firmus gives evidence to the parliamentary inquiry

    Its energy manager says the generators are backup and are not intended — as others have claimed — for us to run these when commercially convenient, and that the company is seeking “clarification” of a broad definition of emergency.

What the record answers

No — the three decisions were not made on compatible assumptions. Not because anyone behaved badly, but because three bodies described the same machines three ways and nothing reconciled them until the EPA read an appeal it had not expected.

The incompatibility is specific. The trigger: the condition knows outages and emergencies; the contract knows reserve notices. The decider: the condition leaves running to the operator's judgement about an emergency; the contract puts the decision in Hydro's hands and prices a penalty for refusing. The purpose: the condition permits power for maintenance and failure; the contract wants power for system relief.

What the documents do not show is a misdescription. The EPA's statement is about the adequacy of its own gateway — it says it would likely have decided differently, not that it was deceived. Firmus's appeal is a lawful request to widen a condition, made in a tribunal that exists for that purpose. And Hydro's hedge is a commercial assumption, not a permission: a contract cannot authorise what a planning condition forbids.

The case for the machines, at its strongest

Standby generation is ordinary. Almost every large data centre has it, and network operators often expect it. The record here is not a story about diesel being sneaked onto a site.

  1. The condition already permits what Firmus says it wants. Genuine emergencies, outages and maintenance are allowed. An operator asking for the definition of “emergency” to be clarified is not the same as an operator planning to run.
  2. Self-supply during scarcity helps the system. If the grid is short, a large customer that runs its own machines is removing load, not adding it. Hydro's clause is a system protection dressed as a commercial term.
  3. The appeal is bounded by its own words. “Limited augmentation … to offset curtailment” is not “run whenever convenient”. A tribunal can grant a narrow variation.
  4. Nothing is running yet. Condition 13 binds until it is varied. The environmental risk is prospective, and the EPA retains a seat at the table.
  5. The company is funding research on reducing diesel reliance at its Wesley Vale site. That is not the behaviour of an operator planning to treat the fleet as baseload.

The counter-case is strong, and it still does not answer the question. Grant every point above and the compatibility problem remains: the assessment was made on one description, the contract on another, and the permission on a third. Which of them governs the machines is exactly what the Tribunal has been asked to decide.

Four things remain unknown, and each is a record rather than an argument. The capacity of the fleet is not published, so no emissions estimate can be anchored. Whether the EPA was ever shown the March board paper is not established either way. Whether the Tribunal has heard the appeal is not public at the time of writing. And whether a reserve notice and a “genuine emergency” would ever coincide is a question of electrical fact that no document in this file settles.

The physical object is not complicated: 276 machines behind a fence, permitted to run on maintenance and failure, contracted to run on instruction, assessed as though they would rarely run at all. The state made three decisions about them. The question is whether it made them about the same thing.

Five · Chapter four

The promised return, and the machinery it needs

Six benefits are claimed. Each needs something built to produce it. Across six rows, one thing has been written down, and it is a payment.

The case for the development is not mainly about the buildings. It is about what the buildings are said to cause: careers that keep young Tasmanians, local firms that grow, research that becomes possible, manufacturing, creative industries, a more resilient economy. None of those is a property of a data centre. Each is a property of an economy with particular institutions attached to it — and the attachment is a contract, a course or an allocation, not a building.

The state's own submission concedes the point more plainly than any critic: benefits should not be assumed from gross investment alone. If they are not to be assumed, something must produce them. So the question is not whether data centres create ecosystems elsewhere. It is what Tasmania required.

What each promised benefit needs, and what the record contains
The promise What would produce it Secured Observable
A career worth staying forthe claim that carries the others Employers deep enough to offer a ladder, and pay bid up by scarcity No mechanismproponents are expected to engage with TAFE and UTAS — a hope addressed to someone who has not signed it Promiseda data portal on local employment and skills — not yet operating
Local firms that growprocurement, not proximity Procurement obligations, supplier development, payment terms a small firm can carry No enforceable termlocal procurement is listed only among the commitments proponents report against Promisedthe same portal, per stage, with ABNs and values
Compute Tasmanians can usethe mechanism the adjacency argument assumes Reserved capacity at a price a researcher or a start-up can pay No mechanismrecommended twice; not in the policy Noneno allocation, no utilisation figure
Manufacturing that staysthe one channel already working Fabrication and assembly inside the state None soughtno local-content term in any instrument Documented elsewherethe modular build is prefabricated in regional New South Wales
Creative industriesrender, inference, latency Wholesale or subsidised compute, and a customer pathway No mechanisma games fund exists; no access terms to the facility Noneno take-up, no usage
An economy that diversifiesthe claim underneath all of them Retained capability that outlives a single customer Payment, not capabilitythe community fund binds money, not capability Noneno published net assessment
Secured is not the same as observable. A supplier register would show whether local firms won work; it would not create a supplier. An employment return would show whether people stayed; it would not give them a career. Two rows have a measurement promised and no row has a mechanism in force. The emptiness of the third column is a statement about the record, not about the future: it lists what no instrument requires, not what cannot happen.

The distinction between the last two columns is the useful one. Of the instruments that would change an outcome rather than reveal it, three are named in the record and none is written: reserved compute, so that the capacity is available on the island; a procurement or local-content term, so that the build buys something here; and a training commitment tied to real posts, rather than an expectation that proponents will talk to TAFE.

Everything else is reporting, and reporting is worth having — it is how anyone would find out. But measurement is not a substitute for mechanism. If the policy depends on an outcome, it needs either an instrument that helps produce it, or a clear theory of why it will happen anyway. The record names the first in three places, and argues the second nowhere. The state's own submission rebuts the assumption that benefits arrive with the investment — which is not the same as rebutting every argument for how they might, but it does move the burden: the benefits should not be assumed from gross investment alone.

The expectations are the test case. The policy says proponents must enter agreements covering renewable generation, transparency and local benefits, and the Government describes them as binding. That is more than an aspiration and less than a mechanism: no agreement has been published, so what it requires, how it is enforced, and who carries the downside cannot be read. On that last question the record is unusually specific. Hydro has described firming — filling the gaps when there is little wind or sun — as its own role under one of the two structures it set out, and says it is still deciding whether an agreement would be profitable, sound for its water storages and good for Tasmania. Until an instrument exists, a mechanism has been announced. It has not been shown.

This is not an argument that the benefits will not arrive. It is a list of what would have to be written down for them to be produced rather than hoped for, and an observation about how much of it has been. The investment is real. The machinery that would turn it into the thing being promised is not yet written — and it is ordinary machinery, of the kind a state writes when it intends to be paid in something other than activity.

Six

The chapters ahead, and one thing that has already happened

Four chapters are built, two more and a checklist are mapped against the same documents. One development has arrived that changes what the map is for.

  • Live · 18 November

    The moratorium that is a motion, and the bill that is not yet law

    The House of Assembly voted 17 to 13 to call on the Government to immediately stop approving new AI and data centres above 5 MW until Parliament agrees fit-for-purpose guardrails. A resolution binds nobody: no permit was stayed, no contract touched. The day before, an independent member introduced a bill that would bar new approvals — while exempting applications already lodged and Panels already established, and blocking extensions to approvals already granted. It has been read once. The Government refused the motion outright the next day — it won't be doing it. Between a motion and an Act there is an interval, and in that interval a proponent can lodge, contract and begin.

  • In preparation

    Forty-six billion dollars, and who gets it

    A headline economic contribution whose own author describes it as indicative, as a contribution rather than a net benefit, and as value that substantially accrues to the facility's owner. No net public benefit figure has been published.

  • In preparation

    The price of the transition, and who carries it

    A contingent $370.6 million network project triggered by committed load. A community fund set at $3,000 per megawatt-year on a base that is not defined. A bridge period with no published cost.

  • Checklist

    What would settle the questions

    Seventeen specific records, who holds each one, and what each would establish — assembled as a checklist a committee, a journalist or a constituent could use.

Seven

What this piece got wrong, and how it knows what it knows

An investigation that cannot show its corrections is asking to be believed on its tone.

The most useful thing this project has produced is not a finding. It is a list of the things it had to withdraw. The hypothesis that gave the project its working name — that Tasmania is being run as a digital resource colony — is, on the documents reviewed, not substantiated. No subsidy has been identified. An objecting regulator, hard-negotiating state businesses and five appeals are all in the record.

  • 17 Nov 2026 A site name. The 104 MW site was published as Westbury; it is St Leonards, a different town about 40 km away. Corrected in the registry and the published page, and recorded here.
  • 16 Nov 2026 444 MW was described as committed. It is a request. One hundred and four megawatts is contracted; the rest is proposed.
  • 16 Nov 2026 The jobs comparison was withdrawn. An earlier edition compared a forecast 8,900 jobs with a 10,800-person tech workforce. The arithmetic was right and the inference was meaningless: different populations, different years, different definitions.
  • 15–16 Nov 2026 A $370 million augmentation was withdrawn, then re-documented. No network document in the corpus supported the figure as originally described. It is now cited with its correct meaning: a contingent project for voltage stability at George Town, not a general northern augmentation, and not committed spend.
How the evidence works — open

Every factual sentence here traces to a registered source, and every citation opens a panel showing the document, its date, its file, a cryptographic hash of the bytes that were read, and whether the quoted words were found in the document's text by the build that produced this page.

A verified quotation means the words match the document. It does not mean the document's own assertions are true. Where a source could not be retrieved — one government release is served behind a bot challenge and exists here only as a search-index entry — the citation says so rather than implying a reading that did not happen.

Figures are not typed into this page by hand. The totals, the site counts and the contracted cap are injected from the project's canonical claim registry at build time, which fails closed if a citation does not resolve or a total does not add up.

Appendix

Documents cited in this extract