Australian federal and state governments data centre policy responses have focused on land-first planning acceleration pathways and framing the data centre conversation as ‘sharing the benefits’. A sharp contrast to the United States which frames the data centre build out as ‘essential to national security, economic prosperity and scientific leadership.’1

Where is policy formation heading towards?

A jurisdiction offering accelerated planning is not necessarily the most helpful to getting a data centre energised. Just look at Sydney, the largest data centre hub in Australia with 90 data centres already operating, where the Investment Delivery Authority is now effectively moot followingTransgrid’s public letter to data centre proponents declaring there is limited grid capacity.

What is in motion is the Australian Government departmental executives working to enact Expectation 2 of the National AI Plan which requires data centres to:

“Secure new and additional clean energy generation and/or storage to offset demand

Cover their share of transmission and distribution infrastructure costs

Improve the overall security and stability of the energy grid, including by enhancing demand flexibility and opportunities for peak-load management”2

TNSPs are also being proactive, with Transgrid seeking a rule change that would see large load users, such as data centres, pay for their full connection capacity to the NSW grid regardless of whether they use all of it3.

‘Take or pay’ of contracted capacity regardless of use, is salve to the phantom-demand question and can help lower consumer cost, but it does not accelerate transmission build out.

The proposed rule change builds upon precedent in the United States. Drawing upon Halcyon’s analysis4 across 65 grid operator tariffs, the United States is moving towards the following five most common initiatives to reduce costs to consumers:

1. Minimum contract term lengths. For instance, Kentucky Power sets a 20-year minimum contract term for all new loads of 150MW+.

2. Minimum monthly billing demand. For instance, Indiana Michigan Power sets the minimum monthly billing demand as >80% of the capacity contracted by the customer.

3. Collateral requirements. For instance, Dominion Energy requires $1.5 million in collateral per MW of contracted capacity.

4. Exit fees. For instance, AEP Ohio allows the customer to terminate the contract for a fee equal to 36 months of minimum charges, but only after 5 years in the contract, and with 3 years of advance notice.

5. Capacity reassignment. For instance, Wheeling Power Company allows customers to reassign or reduce up to 20% of their contracted capacity without penalty under certain notice or term conditions.

The context of these tariffs is important as the United States energy market has cheap gas and clean firmed nuclear to co-locate, making it straightforward to solve for ‘bring your own generation’.

Australia’s response is instead, pay and wait for transmission.

What could data centres be advocating for instead?

Rather than land permits, it should shift towards a power-first policy response – which is the real unlock for getting a data centre built in Australia.

And, match the growing requirements for data centres to contribute with a genuine pro-investment framework.

Investors have significant influence in energy policy and they must begin to advocate for data centre and associated transmission and generation in Australia.

Let’s look at policy and funding mechanisms to accelerate transmission which draw upon United States response to the build out:

The so what for investors

Read the policy trajectory as a one-way ratchet on cost. The take-or-pay rule change, the offset mandate, and ‘cover your share of transmission’ need to be priced before a data centre draws a single megawatt. None of these policies move the date the megawatts arrive.

And for institutional investors, is the data centre offtake customer at last the right ‘shape’ for a genuine private transmission infrastructure proposition in the NEM? Could a hyperscaler vertically integrate or JV with an infrastructure fund or superannuation fund to deliver its own transmission build out?

Contestability is built into the National Electricity Market framework and the exploration of its ‘limits’ is surface level. The timing is opportune as incumbent Transmission Network Service Providers are genuinely stretched both planning for and executing major transmission projects to deliver the AEMO Integrated System Plan.

The United States is already adopting a hybrid approach with Pennsylvania Power & Light Electric Utilities and Blackstone Infrastructure creating a Joint Venture to build, own, and operate new electricity generation stations to power data centres in Pennsylvania under long-term services agreements (PPL 2025 Annual Report).

A contestable connection intersects the shared network; it does not create capacity upstream. So, the proposition works where network capacity already exists or is planned to be unlocked – such as the Hunter in NSW, the proposed VTP 2025 Truganina Terminal Station in Victoria or cutting into CopperString in Queensland.

I would privately be countering hard against the direction of data centre policy changes so government matches cost increases with ‘clearing a pathway’ and in parallel, re-running the analysis of which NEM jurisdiction offers the fastest path to megawatts.

If you would like to discuss further, I can be contacted by email: info@followthebottleneck.com.

Not investment advice. This article uses public information only and its content is general in nature.

Firmus Technologies (Firmus) is reported to IPO in September 2026 on the Australian Stock Exchange as a new, innovative offering of data centre for artificial intelligence (AI) training.

This deep dive seeks to provide insights into the assumptions underpinning its path to power to help investor capital allocation decisions.

I have purposefully kept AI revenue forecasts, immersion cooling versus other technology and further commentary about the personalities of Firmus’ leadership team out of scope in creating this report.

Personally, I wish this IPO was a routine affair in Australia, with say 4+ Firmus-shaped technology IPOs introduced each and every quarter. I hope their story inspires others to deploy first-principles thinking to complacency and push through the dark times for the chance of doing true work.

Read the deep dive and let me know what you think.

Best regards,

FTB

What is the opportunity and can Firmus capture it?

Firmus is positioning itself to be a company that can deliver AI training capacity by using Australia as the location for ‘AI school’ – bringing AI training to the location of the electrons, rather than demanding electrons be brought to training.

This AI-only breed of data centre is seeking to capture the widely reported insatiable demand for GPU compute. The thesis is: more energy enables more compute that leads to smarter, more sophisticated AI training and inference, which in turn, drives greater revenue.

The Firmus pitch hinges on solving the problem of needing to cool computation so the GPUs can deliver a higher level of performance. This problem is being exacerbated by the density of compute that AI training requires – high intensity bursts of power for the AI to learn ‘something’, such as how to drive a car.

A data centre needs to stay within a compliant temperature range to enable the GPU computation to happen. Before AI and the ChatGPT release date in 2022, this was achieved primarily through using air conditioners and fans to absorb heat from the server racks and remove it to outside the data centre to maintain safe operating temperatures. And as AMD, NVIDIA and AWS chips came into production, this brought about a vast increase in computation density and therefore more heat being created, which needed the cooling response to shift to more efficient methods, such as single-phase immersion, chiller-based direct-to-chip, 2-phase cold plate, and now with full-liquid immersion – which Firmus is targeting to deploy.

The Firmus story

Founded in 2019 in Sydney by Oliver Curtis, Tim Rosenfield and Jonathan Levee the company started out in bitcoin, where it was using immersion cooling to help improve the efficiency of running servers used for mining bitcoin tokens.

In circa 2023 it pivoted entirely by re-directing its interest in cooling server racks towards data centres housing GPUs for AI. Incidentally, this is the same origins and pivot as its Australian-American rival, IREN.

Firmus is now headquartered in Singapore, encouraged to relocate by ST Telemedia Global Data Centres, owned by the Singaporean Government investor, Temasek.

The IPO has been pushed back a few months to capture end of year accounts, with filings to the ASX to be carried out thereafter.

Firmus and the rest

For AI-only in Australia, Firmus has IREN as the key competitor who are in lock-step with NVIDIA and deploying its GPUs at sites in the United States and Canada. IREN announced in June 2026 they are to build their first data centre in Australia, a $10 billion 800 MW data centre in Bundey, 165km north-east of Adelaide, energised in 20281.

The leading data centre developers on mainland Australia such as CDC, AirTrunk, Goodman, DigiCo, NextDC, Macquarie Data Centres diversify their client base to those who require cloud and colocation solutions.

Immersion cooling technology has been deployed by IBM since the 1960s. And there are a number of companies who offer the liquid cooling data centre technology for AI compute such as Submer, Green Revolution Cooling, LiquidStack, and Vertiv.

What Firmus does, precisely

Strip away the beguiling marketing and this is what the money is going to:

• Building 3 Firmus-owned data centres in Tasmania called ‘Project Southgate.’

• Retrofitting its ‘HyperCube’ immersion cooling technology into existing data centres located across Australia.

Note this deep dive focuses on Project Southgate rather than unpacking HyperCube. My one question would be could it be used for hyperscalers making their own chips such as Amazon, Alphabet and Meta?

However, let’s unpack the plan to build 3 data centres.

Project Southgate – building the 588 MW data centre campus in Tasmania

Land has been acquired near Launceston to build the first stage of Project Southgate, a 100 MW data centre with 36,000 NVIDIA GPUs installed. The second data centre is much more ambitious, 288 MW or 500 MW (depending on the presentation) with an estimated 100,000 GPUs at George Town, and the third in Wesley Vale at 200 MW with ~75,000 GPUs. Bringing up the total of ~211,000 GPUs required from NVIDIA.

The proposed data centre campus is to be delivered under the build, operate, own, maintain model.

Path to power for Project Southgate – the risks and opportunity with the critical path

Its three locations across northern Tasmania for their first purpose-built data centres is genuinely smart. Launceston, George Town and Wesley Vale are cold ambient climate locations making it a more efficient site for cooling racks.

And as it happens, the location selected neatly sidesteps the very real and current power constraint issues playing out in metropolitan Sydney and Melbourne where it is extremely difficult to secure a 100 MW+ connection agreement, the scale of data centre required for AI training.

As recently as 17 June 2026, NSW network operator Transgrid published a letter to data centre proponents signalling metropolitan Sydney transmission is at capacity: 20260617-update-on-network-capacity.pdf

For Project Southgate, the state-owned energy retailer Aurora Energy has signed a three-year retail service agreement to supply up to 104 MW of electricity from Hydro Tasmania.

But read ‘104 MW contracted’ is capacity, not guaranteed firmed energy. The real question is whether Hydro Tasmania can serve a flat, 24/7 load – to overcome ‘dark and still’ – during 2026 - 2030.

Over the past two years, Tasmania has been in drought and there has not been enough water to pump up and dispatch down its hydro power stations, with the Tamar Valley Gas Station stepping in and generating 221 GWh in 2024 and 288 GWh in 2025, up from 71 GWh in 2023, a non-drought year, and relying heavily on Basslink by net importing 1,202 GWh in 2024 and 1,910 GWh in 2025 from mainland Australia (Source: Hydro Tasmania Annual Report 2024-25).

The lifeline for Firmus is the 750 MW Marinus Link which doubles the interconnection capacity between Tasmania and mainland Australia from 2030. However, Marinus Link is a shared, consumer-funded, two-way cable that moves energy rather than creating it. The interconnector’s business case is for the National Electricity Market, not to serve one large load customer. For instance, VicGrid write how Marinus Link unlocks Victoria’s access to Hydro Tasmania’s pumped hydro generation.

So, what does this mean?

I’m not sure the contract agreement is sufficient as currently agreed, as it is revealing an energy capacity gap when George Town is energised in 2028 – before the connection of Marinus Link.

Firmus could be relying on Tasmanian generation and storage projects via its 1 GW North West Renewable Energy Zone. But this REZ’s status is still ‘proposed’, not declared, or better yet, under construction. It requires legislation to be introduced into parliament for it to be declared. Which will be difficult to negotiate, as the Liberals are a minority government and rely on the crossbench for the passage of bills through parliament.

Firmus points to up to 5.1 GW of new renewables could be supported nationally, but it is not contracted or built and again, only 104 MW has been secured for its Tasmanian data centres.

For investors looking at Firmus, the capex bill required for additional generation for Project Southgate is the un-priced risk.

Until now.

Firmus needs to urgently review the following options.

Option one: underwrite renewable generation and storage projects either in Tasmania or on mainland Australia to fully offset its 588 MW electricity demand. This is the requirement for data centres as per Energy and Climate Change Ministerial Council meeting on 8 May 2026.

Firmus requires ~3 GW of new generation and storage costing ~$6 billion, against 104 MW actually contracted.

Or to use the other reported capacity figure for George Town – 500 MW, which requires 4 GW to offset at ~$8 billion capex cost (using same GenCost 2025-26 rates above).

But putting aside the cost, the more significant concern for investors is time to build the renewable generation and storage required for Project Southgate.

A very real stranded asset risk begins to emerge.

In the NEM, it is taking on average to execute origination, planning, connection, and construction:

• BESS: ~2-4 years

• Grid-scale solar: ~3-5 years

• Wind (onshore): ~7+ years

This could materially delay the ambitions of St Leonards, George Town and Wesley Vale data centres – as it needs more than a DA through council, it needs to be matched with 3 or 4 GW of new power.

Or Firmus could look at option two - build gas peaking power behind-the-meter, which will be quicker to deploy and deliver the energy profile that the data centres require – flat, 24/7 load. It is already building over ~300 back-up diesel generators behind the meter at St Leonards and George Town (as per DA documents), why not commit a bit further?

So far so good, $1.06 - 1.4 billion versus the ~$6 or 8 billion for renewables plus storage offset. And the time to construction is realistically 3-5 years with the binding constraint being the gas-turbine procurement supply chain due to backlog of orders at GE Vernova, Siemens and Mitsubishi.

But the very real constraint is there is only 1 pipeline supplying gas to Tasmania with a total capacity of 129 TJ/day, already serving the Tamar Valley Power Station.

New gas stations at full load would need more gas than the existing pipeline can carry – 588 MW would draw ~154 TJ/day and 800 MW ~209 TJ/day. To overcome this, Firmus may have to underwrite a new gas pipeline, alongside the fibre cable it is building to mainland Australia.2

And as proposed, if no generation and storage is secured in a timely fashion, Follow the Bottleneck analysis shows that once all three data centres are built:

• In a drought year such as 2025, Firmus could be consuming ~61% of Hydro Tasmania’s generation from its 30 hydro power stations and its sole gas-fired power station.

• In a non-drought year such as 2023, Firmus could be consuming ~49% of Hydro Tasmania’s generation.

(Source: Table 3: Generation Summary 2019-2025, page 29 Hydro Tasmania Annual Report 2024-25).

If I was an investor in Firmus, I would want a much sharper picture for the path to power for Project Southgate to understand the capex required, timelines to energisation and options to mitigate the risk of a stranded asset.

More broadly, this is an inflection point for Australia. It has already begun as hushed whispers in between sessions at energy conferences but I am calling it out, and dropping the dead possum on the table: Australia is no longer in an energy ‘transition’.

To accommodate population and industrial load growth and AI – we need it all. No matter the source, the energy reality in this country is now ‘all of the above’.

If you would like to discuss this deep dive further, I can be contacted by email: info@followthebottleneck.com.

Not investment advice. This article uses public information only and its content is general in nature.

For the open access network, the Network Service Provider – such as Endeavour, Ausgrid, Transgrid - has to offer a path for connection to a data centre. Whether there is sufficient capacity in the network is another question, but for now, this follows the connection process which adheres to the National Electricity Rules set out by the Australian Energy Market Commission which at a high level consists of 3 stages:

1. A 5.3.2 connection enquiry

2. A 5.3.4 application to connect

3. And a grid connection

This formal process also takes into account the proponents ‘readiness’ i.e. have they purchased the land for the proposed data centre, is there financing secured, construction partners in place, and maturity through the planning system. The NSP is sizing the data centre up – are they a serious player or another property developer kicking around the idea of turning an existing industrial site into the latest fashionable thing.

The data centre proponent would do well to keep in mind the commercial interests of those Network Service Providers who are looking at any opportunity to increase its regulatory asset base.

Risks involved with a connection agreement

Putting aside technology risk of the data centre facing obsolescence as power density requirements and cooling technology evolve, or concentration risk of a single hyperscale tenant moving on – important I think to spend some time looking at the connection agreement risk as it relates to the NEM.

What it gives you – an agreed maximum MW capacity in exchange for a usage charge which is determined via the Transmission Use of System (TUOS) payments.

This is the first move for data centres as it avoids the capex and complexity of building its own generation and storage required for its load profile.

Now to the risks, which need to be better understood.

With the grid connection, the data centre is now a NEM participant meaning:

The data centre can be asked by the Australian Energy Market Operator to curtail or load shed should it be required to maintain system frequency. Fine, and the response has been to use on-site diesel generators to plug any shortfall or be activated during peak demand events. Risk here is further capex spend and regulatory uncertainty as the regulator could look to challenge this default option given stacking diesel generators is incongruous to legislated emissions reduction targets.

More existential is AEMO’s due diligence could lead to energy ministers mandating data centres to ‘flex’ load which is at odds with commercial objectives which require flat, continuous, reliable data service uptime. A data centre that cannot meet 24/7 uptime for their customer base risks losing the contract, as they will shift to a provider who can.

On page 39 of AEMO’s 2025 Electricity Statement of Opportunities, the door is left open for data centres to provide ‘more demand flexibility’, to explore whether ‘potential load shifting may be plausible’.

Worth noting here AirTrunk are trying to get in-front of it and position it as a feature, not a bug, with load shifting to be adjusted so ‘it responds to the needs of the power networks, which could prevent price spikes and rolling outages’ (The Energy). AEMO agree in principle and are reviewing whether the timing of non-urgent computing tasks aligns with renewable generation peaks – i.e. the significant abundant rooftop solar that is available during the middle of the day.

Now add the layer of political risk. The energy ministers at their last Energy and Climate Change Ministerial Council meeting on 8 May 2026 agreed:

“Data centres on the NEM and WEM should invest in additional renewable generation and firming in their state of operation to fully offset their electricity demand and provide flexibility services to avoid additional costs being borne by other energy users. Further, data centres should provide transparent reporting on their energy use and emissions production.”

To what extent, and who is in scope are all live questions. But the regulatory response in Australia is becoming clearer – data centres will be forced to spend more capex on generation and storage once they are handed that golden ticket.

Interestingly this ruling does not extend to Queensland, where its Minister for Energy refused to agree to the proposal, so perhaps the state of Queensland just got more appealing for data centres to prosecute as a place to locate as connection queues remain stalled in NSW and Victoria?

And note energy ministers do not address the possibility of wholesale power prices increasing as a result of data centres soaking up available demand. If that was to happen, then the political response moves rapidly to restrictions and possible moratoriums, especially as the energy debate in Australia has shifted from reducing emissions, to reducing consumer electricity bills.

Trust is good, control is better

Now as I understand it, frontier AI data centre proponents are rather cost-insensitive, and the rule of thumb I draw upon is OpenAI’s CFO quoting the astonishing metric of 1 GW of power converts to $10 billion in revenue per year. Note however this is not a blanket rule for all data centres as colocation and cloud operators are not as cost insensitive.

So if rising costs – broadly speaking – do not phase data centres, then how to speed up the process to get those electrons flowing and revenue realised?

Can continue to sweat the existing transmission network, ramp up MW requests slowly, and accept a far greater cost and regulatory burden and political uncertainty for the next few years. And as I am laying out here, the golden ticket may not be enough to provide uninterrupted firmed power to meet the insatiable customer demand for electrons.

Or instead, do it yourself.

If your view of artificial intelligence is bullish, and in five years’ time, we are going to have more digitisation, more AI integration – not less, then think in systems.

Scrape every learning possible from the existing connection process and really commit and build your own private substation, with behind the meter storage and a gas-fired power station, while you wait for the grid to build a significant amount of new generation. The regulator is going to make you spend more for generation and storage regardless, why not do it on your own terms, rather than giving your competitors an edge?

Are you funding a ticket or a system?

For the investor, the implications are clear. The mandate to fully offset demand is a forced bid for firming and generation and co-location land.

For example, a 100 MW data centre forced to ‘fully offset’ its demand faces roughly $794 million - $1.2 billion in generation and storage capex – which runs to 50% of the facility’s own build cost – and none of it appears in the connection agreement.

The owners who can do that re-rate, the holders of a connection agreement or who are still in the queue carry an offset bill few have quantified. And while NSW and Victoria argue about it, Queensland didn’t sign.

Not investment advice. If you would like to chat further, I can be contacted on info@followthebottleneck.com

To explore this, Oxford Economics Australia’s latest research notes the Australian Energy Market Operator received 44 GW of data centre connection requests from Network Service Providers. Oxford Economics Australia posit 6 in every 7 MWs of connection requests are estimated to be ‘phantom demand’. That is, they are not expected to materialise under AEMO’s Step Change scenario. So, of the 44 GW of connection requests received, only 6 GW of prospective project capacity is expected to meet demand[1].

This churn of speculation leads to incredible confusion for all stakeholders – regulators, government, developers, utilities and investors – who are grappling with trying to understand the ‘shape’ of the data centre forecast. And to take one stakeholder perspective, it is critical to know precisely, so the transmission network planning team can accurately forecast the electricity infrastructure required to accommodate the demand.

To look at this pipeline another way, of the data centre projects that are to be energised, what type of data centre actually needs to be built in Australia?

Or to put it more bluntly:

We should start with first principles thinking – why would compute live in Australia at all?

Strip away the frothy announcements and there are exactly two structural reasons for a data centre to exist in Australia rather than somewhere with cheaper and more abundant power.

Captive data

The Australian Government’s Security of Critical Infrastructure Act imposes security obligations on critical infrastructure operators such as asset registration, mandatory cyber incident reporting and risk management programs. This is supported by the Whole-of-Government Hosting Strategy and the hosting certification framework for government data and dataset specific rules. This means government data such as from Department of Defence, or private and public sector data such as health records is captive demand that must be domiciled in Australia and not arbitraged away to a cheaper site in say the United States or India.

Latency

Enterprise applications, real-time payments and other consumer, cloud-based functions must sit adjacent to the users they serve – hence the established data centre market that is already operating in metropolitan Sydney and Melbourne – it is where the demand is.

If cloud is the base case, why then should AI training and inference be built in Australia?

JLL’s report notes that despite AI gaining daily users, it only represents about a quarter of all data centre workloads in 2025, with training driving most of that demand. JLL posit a significant shift is anticipated in 2027 when inference workloads could overtake training as the dominant AI requirement.

While an AI model represents a one-time or periodic investment, once the model is created, inference generates ongoing revenue through actual application usage. The theory being every AI model deployment creates sustained inference demand that grows with user adoption. This growth trajectory however, depends on the emergence and rapid adoption of inference applications that does not yet exist.[2]

Australia has no structural advantage for AI training

AI training is not bound by latency requirements and is the most location indifferent workload in computing. It chases the cheapest firmed power available and that contest is being won in the United States and the Middle East, not on the NEM where wholesale prices sit well above those markets and metropolitan connection timelines run to years.

To put some numbers to this claim, the average NEM wholesale spot price for 2025 was USD$58/MWh (AEMO CY25 mean converted to AUD/USD) and Texas ERCOT was USD$35/MWh. Meaning NEM energy for data centres is 66% more expensive than Texas, a key data centre growth corridor. (AEMO, US Energy Information Administration).

AI inference inherits both structural tailwinds as cloud

Enterprise and transactional inference are latency-sensitive and increasingly carry the same data-residency obligations as cloud – so it must be served from within the country. This structure favours the existing metropolitan data centres where the users and existing cloud infrastructure already sit, not a regional, remote campus in say a renewable energy zone adjacent to renewable generation.

Existing metropolitan-based cloud data centres have the right location and tenant relationships to capture inference workloads but the facilities will need significant power and cooling upgrades to host it.

Inference is the slice of the AI boom Australia could capture and it arrives through the metropolitan data centre locations, serving the same hyperscale and enterprise tenants as the cloud build out that preceded.

So sovereign cloud and co-location can win, and the investable Australian share of AI is largely inference leveraging the same infrastructure.

But because inference must sit in metropolitan areas, it stacks onto the most congested area of the high-voltage transmission network. Therefore, the unlock is not necessarily wholesale power price, it is metropolitan transmission and distribution connection capacity. And as per a previous article, metropolitan Sydney has 16 GW of data centre connection enquiries against a network forecast expectation of 800 MW of actual near-term data centre load.

And, the case for inference rests on demand that is a forecast, not locked-in contracts.

What is perhaps speculative for Australia and carries the most risk is the data centre being built to only capture training demand.

What a credit lender should take away from this article

If you lend or are looking to lend against these assets, the first-principles thinking can collapse into four underwriting rules:

1. Tenancy quality is the asset. A facility anchored by hyperscaler pre-commitments, with contracted leases and approvals in hand is fundamentally a different proposition from a tenantless build in a secondary location. The former is infrastructure. The second is a land bank with a generator or battery attached.

2. The energisation date is a claim not a fact. The information memorandum will assert an energisation date however almost none of them can be independently verified by the lender against connection queue positioning, the augmentation costs and upstream augmentation impacts. A date that slips by 18 months reshapes the entire conversation.

3. Distinguish the workload thesis. ‘AI demand’ can mean anything from contracted sovereign inference capacity for a hyperscaler or a hope that someone, eventually wants to train a model in Australia.

4. Watch the wholesale price forecasts with great interest. When the finite grid capacity is seen to be absorbed by data centres which in turn, spikes the wholesale price, the regulatory and political response will be swift. This is a live risk and the curtailment and system security impacts are not well understood.

Not investment advice, general information only. As ever, if you would like to discuss this further, or wish to stress-test a specific project, I can be contacted at: info@followthebottleneck.com

[1] 2025-11_Oxford-Phantom-Demand-Research-Briefing.pdf

[2] 2026 Market Outlook for Global Data Centers | JLL Research

Look to the Hunter region, located north of metropolitan Sydney.

Coal exits leave behind an underutilised high-voltage transmission network

What makes this region interesting from an energy system perspective is with the exception of Mount Piper, the Hunter hosts the state’s coal-fired power station generation capacity:

1. Bayswater Power Station at Muswellbrook (2,715 MW) owned by AGL. Expected closure 2033.

2. Eraring Power Station at Lake Macquarie (2,800 MW) owned by Origin Energy. Expected closure 2029.

3. Vales Point Power Station at Lake Macquarie (1,320 MW) owned by Delta Electricity. Expected closure 2033.

This generation is forecast to close thereby creating an intriguing proposition – the existing transmission infrastructure capacity could be soaked up by a large inverter-based load data centres.

Source: NEM Generation Information April 2026

The Hunter has a significant program of transmission and storage projects which are not stuck in planning, they are being delivered right now for the NSW Electricity Infrastructure Roadmap

When the coal-fired power stations retire in 2029 and 2033, energy will largely be pulled from generation outside of the Hunter region. To deliver this energy and firming capability, the key projects underway in the Hunter region include:

Waratah Super Battery

A Priority Transmission Infrastructure Project (under NSW’s Electricity Infrastructure Investment Act 2020) to alleviate constraints on the existing transmission network during periods of high demand.

Transgrid is upgrading the existing network to maximise the benefits of the Waratah Super Battery project including uprating a number of 330kV transmission lines (including the Yass to Marulan, Yass to Collector, Collector to Marulan and Bannaby to Sydney West Lines) and existing substation assets that would otherwise limit the Waratah Super Battery ability to increase power transfers into the Hunter and broader area from regional NSW.

Hunter Transmission Project Stage 1

The EnergyCo project includes a new double-circuit 500 KV line between new Bayswater South and new Olney 500 kV switching stations to help import electricity from the Central West Orana and New England REZs.

HTP is an actionable project in AEMO’s 2024 Integrated System Plan and is proceeding as a Priority Transmission Infrastructure Project. It is scheduled for completion by November 2029.

HTP scope includes:

• Establishing a new Bayswater South 500 kV switching station near the existing Bayswater substation, and cutting into the existing 500 kV network

• Establishing a new 500 kV switching station in the Olney State Forest, near Eraring, and cutting it into the existing 500 kV network

• Constructing an approx. 100km new 500 kV double-circuit line between the two new switching stations

• Installing new 500/330 kV 1,500 MVA transformers at Eraring substation

• Installing line reactors at the Bayswater South switching station on the new double-circuit transmission line

There is also a HTP Stage 2 contemplated which may be required after 2033 to unlock further generation capacity from future stages of CWO and New England REZ. Stage 2 would look to be scoped as:

• Developing a second 500 kV double-circuit between Bayswater and Eraring, using Transgrid’s existing easement by demolishing the 330 kV single-circuit line and rebuilding it as a 500 kV double-circuit line.

Hunter Central Coast REZ

Ausgrid as Network Operator is building the HCC REZ which will bring online an additional 1 GW of capacity. This is being delivered in stages, 350 MW in 2026, 280 MW in June 2028, and 370 MW in July 2028.

Source: Transmission Annual Planning Report 2025

In lieu of coal exits, there is transmission line capacity in the Hunter region

Transgrid’s TAPR 2025 notes that there is no prescribed augmentation projects planned to meet network performance requirements of NER Schedule 5.1 in the Newcastle and Central Coast region over the next 10 years.

This is significant as every other area of NSW and ACT – Greater Sydney, Northern NSW, Central NSW, Southern NSW and ACT and South Western NSW all declare a list of planned projects required to be undertaken.

This means for the Hunter, data centre developers are not relying upon Transgrid to upgrade or deliver new projects to ensure network capacity.

On a transmission line-by-line analysis, the AEMO Operations and Planning Data Management System snapshot provided in the Transgrid 2025 TAPR highlights to N-1 utilisation as a percentage of the major transmission lines across the state.

Of importance for analysing the Hunter region and its potential capacity for data centres, the below transmission lines and areas are of interest.

Newcastle to Eraring and Vales Point:

• Line 92: 49% N-1 utilisation (Eraring to Vales Point)

• Line 24: 50% N-1 utilisation (Eraring to Kemps Creek)

• Line 94: 36% N-1 utilisation (Newcastle to Waratah West)

• Line 9W: 32% N-1 utilisation (Waratah West to Tomago)

• Line 9C5: 31% N-1 utilisation (Tomago to Brandy Hill)

Liddell (former Coal Power Station site) to Muswellbrook and Bayswater:

• Line 5A4: 28% N-1 utilisation (Bayswater to Wollar)

• Line 5A3: 27% N-1 utilisation (Bayswater to Mt Piper)

Source: Transmission Annual Planning Report 2025

Data centres can raise the minimum demand profile and therefore abate the risks of system strength gaps from FY28 to FY30

In NSW, when coal generation retires, traditional network flows will change and voltage support across the network will be reduced.

During periods of low renewable energy output, NSW will increasingly rely on long-duration dispatchable capacity from Snowy Hydro units. Transgrid is seeking to build the southern backbone of HumeLink and Sydney Ring South with its ability to transfer power from Southern NSW northwards will therefore be critical to energy reliability in the state.

For the Hunter region specifically, four synchronous condensers, each providing 275 MVA fault current, or a 200 MW grid-forming BESS is required to be installed by FY28 for the Hunter Central Coast REZ.

Transgrid forecast there is a risk in the minimum level of system strength before sufficient synchronous condensers come online. The earliest possible delivery of synchronous condensers under the regulatory process is expected between March 2029 and February 2030. Under the delivery timeline, gaps to the minimum level system strength requirements are expected to occur up to 2% of the time for all of FY28 and 5% of the time for all of FY30.

When it comes to network system security and reliability, data centres are not a bug, they are a feature. Locating a data centre in the Hunter would have important, well-timed benefits of lifting the minimum load demand profile, thereby providing a critical system security benefit to the network which can act as an insurance product while Transgrid takes delivery and installs the necessary synchronous condensers required to replace coal generation.

What does this article mean for data centre developers and investors

For the next phase of data centre deployment beyond metropolitan Sydney, I would recommend beginning a three pillared search to secure land for your next data centre location. Fibre and water are a lower order issue, speed to compute, synergies with a skilled workforce and access to power remain the unlock.

I would therefore be doing the following three tasks:

1. ‘Follow the 500 kV line’ and substation build of the Hunter Transmission Project and Waratah Super Battery.

2. Review the favourable N-1 utilisation line ratings and discern if there is land available to cut in or augment an existing substation.

3. Consider co-locating at existing power station sites to leverage the substantial transmission infrastructure already in place.

Key sources used for this article: Transmission Annual Planning Report 2025, NEM Generation Information April 2026.

This is not investment advice. Happy to discuss any of this further, and can be contacted by email: info@followthebottleneck.com

Why a data centre cannot run on renewables alone

Data centres require continuous, uninterrupted power supply.

Solar and wind are intermittent by definition and neither can guarantee the flat, continuous load profile a data centre requires.

Developers can point to a solar and wind PPA and describe the facility as being powered by renewables but it is not the whole story. On a winter’s night, when solar irradiance is zero, and the wind has abated – data centres still require its large load served in full.

Without firming sitting behind that renewable generation PPA the grid is carrying that obligation – and that position is to be challenged once Eraring exits the energy system on 30 April 2029.

What constitutes firming power?

There is no single, statutory definition of ‘firming’ in the National Electricity Rules.

AEMO’s Integrated System Plan describes that in lieu of coal, firming can be derived from storage – batteries and pumped hydro - and gas. These technologies help maintain grid stability and inertia, smooth out volatile frequency and balance out fast changes in supply and demand. Gas generation also provides back-up supply during long periods of ‘dark and still’ renewable droughts and times of extreme peak demand, particularly in winter. (p.65 2024 ISP).

Storage is categorised by AEMO as:

• Shallow: grid-connected storage to dispatch electricity for less than four hours

• Medium: grid-connected storage to dispatch electricity for four to 12 hours.

• Deep: strategic reserves that can dispatch electricity for more than 12 hours, to shift energy over weeks or months or cover renewable droughts – such as pumped hydro.

The problem with batteries is they have a relatively short operational lifespan of 20 years. So the batteries installed during the 2020s will need replacing in the 2040s, ideally not like-for-like, instead with medium or deep storage.

AEMO’s need for gas

What perhaps receives little attention given it is inconvenient for the transition is AEMO backs in gas: ‘electricity from gas-powered generation is forecast to continue its important role in the NEM. After coal retires, gas will be needed to support energy supply during periods of renewable drought’.

The NEM is forecast to need 14 GW of gas-powered generation (Draft 2026 ISP) to ensure the NEM remains resilient under a range of power system and extreme weather events.

To meet the 14 GW target, 11 GW of new gas generation will need to be built to accommodate a number of plants retiring as they reach end of life.

The role of gas is to be a backup for renewable generation and storage, as the cost of gas makes it more expensive option for everyday use. Newer flexible generators can be regularly switched on to serve that back-up role, and can also operate with a clutch to provide critical system security services. (Draft 2026 ISP).

This is for a ‘strategic reserve’ for power system reliability and security, so is not forecast to run frequently. A typical gas generator may generate just 5-7% of its annual potential – an intriguing commercial proposition. A gas power station running at 5% of the time cannot recover its capex from energy sales and instead survives on capacity payments – which a data centre could potentially be funding.

AEMO is unsure how gas investment is to proceed. It models a potential projection of:

• 2 regasification terminals in south-east Australia and associated pipeline (specific locations undisclosed)

• Expansions to the existing east-coast gas pipeline network to transport gas to where it is required

• New seasonal gas storages

• New supplies of existing and additional processing plants.

If investment does not fulfill the above, existing gas power stations will need to rely upon diesel backup fuels and on-site fuel storages for continuity.

Where to get firming power as a data centre developer

1. Continue to run diesel back up gensets.

Data centres already have diesel backup generators on site as standard, sized to full load which is functionally behind-the-meter firming power. However, this will likely change as regulators push to mandate data centres to use renewables plus storage for its load. For instance, the 8 May 2026 ECMC Communique agreed (all ministers except Queensland) that data centres in the NEM and WEM should invest in additional renewable generation and firming to fully offset their electricity demand and provide demand flexibility services to avoid additional costs being borne by other energy users.

Likely pathway 2026 to 2027.

2. PPAs with existing gas and hydro assets

This is the path of least resistance – no planning approval, no capex outlay, counterparty takes the fuel risk. However the data centre developer will be competing for supply and it only offers firming on someone else’s terms.

EnergyAustralia’s Marulan Gas Power Station planned upgraded capacity to 1.43GW by 2032 is a clear signal this could be the short to medium term option for data centres following Eraring exit in 2029.

Likely pathway 2026 to 2029+.

Below are the current options for a firming PPA:

3. On-site battery storage

Offers a method to manage short-duration gaps, reduce peak demand draw and provide grid stability services that regulators are increasingly expecting from large loads.

It is likely government will mandate on-site storage to be co-located with data centre builds.

Available now onwards.

4. Co-locate with an existing coal, hydro or gas-fired power station

Option for data centres is to co-locate with an existing coal-fired power station, hydro or gas-fired power station to unlock firming capacity requirement. This has the advantage of accessing existing high-voltage switchyard and transmission infrastructure already in place.

Constraint is the Commonwealth Government now expects new data centres to fully offset their demand through investment in renewables and storage, which limits how far co-location with coal and gas can be pointed at the load.

Available now onwards.

5. Behind-the-meter gas generation

A captive gas peaker removes the dependence on gid firming availability entirely, especially with the sequencing risks that could occur following Eraring, the largest coal-fired power station in the NEM. This is the most capital intensive, dependent on fuel supply and planning approval.

Likely pathway 2029 onwards.

So is a data centre funding the capacity payments for a gas or hydro PPA? Or is it actually more in the long term interests to begin to de-risk and start the planning approvals now for a generation asset that can be fully controlled by the data centre behind-the-meter.

But where is this generation going to come from?

Let’s examine the National Electricity Market generation pipeline, focusing on NSW where the majority of data centre developments are seeking connection.

The NSW Government is endorsing 15 data centre projects valued at $51.9 billion through its Investment Delivery Authority housed in the Premier’s Department, promising data centre developers an accelerated experience through the planning system.

The Australian Energy Market Operator, has a legislated requirement to publish a quarterly report which tracks the status of generation and storage projects in the National Electricity Market. It focuses on providing a ‘commitment status’ of NEM generation, breaking down ‘commitment’ into:

• Publicly announced: projects that have been announced publicly and are not yet sufficiently progressed to be considered Anticipated or Committed

• Anticipated: generation projects that are sufficiently progressed towards meeting at least 3 of the 5 commitment criteria

• Committed: Projects with known timing and satisfy the five commitment criteria

• In commissioning: projects that have successfully completed hold point testing

• In service: production units that have fully completed commissioning as signed off by AEMO.

The commitment criteria is defined as: the project proponent has obtained land, contracts, planning, finance and a pathway for construction.

This all sounds very logical and prudent but when I audited the list, the dataset was at best a rough estimate against the proposed commitment criteria and could not be relied upon beyond a high-level status overview.

Instead, what would be a far more useful task is to follow the money, specifically, connection enquiry status.

Each project begins by lodging a connection enquiry with the TNSP and AEMO as per NER 5.3.2. Minimal money committed, really a signal of intent.

Next, should the project then move to a connection application (NER 5.3.4), this demonstrates a very strong signal with capital committed, technical modelling and feasibility studies commenced.

Then, should it proceed from connection application to connection agreement, it undergoes hold point testing and commissioning before it is energised, and electrons begin to flow.

Here is the real picture, focusing just on NSW – the most important data centre market in Australia.

There are 65 projects with a total ~16,526MW capacity at the connection application stage before the TNSP.

Of the 65 projects:

• 35 are standalone BESS projects

• 8 are hybrid solar and BESS projects

• 6 are wind farm projects

• 14 solar farm projects

• 1 gas project

• 1 advanced compressed-air project

So what does this mean?

Looking at the technology being deployed, this is not a generation pipeline for NSW. It is a storage pipeline.

35 of the 65 projects are standalone batteries and another 7 pair batteries with solar.

Storage does not create energy, it moves it. A battery charges when generation is cheap and discharges when it is not. It firms the grid, adding nothing to the annual pool of electricity the grid can actually supply.

And the storage in this queue is short duration, sitting predominately between 2 to 4 hours.

Now layer on what a data centre is: flat load that runs every hour of every day.

New load of that shape has to be met by new energy, which creates two inconvenient challenges – the queue is intermittent generation, and any new solar and wind projects that want to progress to connection application stage will take anywhere between 4-10 years to be approved by government (Herbert Smith Freehills Kramer, November 2025).

So for data centres developers and investors sweating the existing transmission network which offers a credible pipeline of BESS, solar hybrids, and standalone wind and solar projects is only going to stack up for so long.

Next instalment will explore what assets and corridors are likely to re-rate when this timeline is priced in.

Announced at the 2026 CEDA Energy and Climate Summit, Sydney has over 16 GW of data centre connection enquiries lodged with Transgrid, the high-voltage transmission network operator for NSW.

To put that in context, NSW operational demand has exceeded 13 GW only a few times in the history of the National Electricity Market. 16 GW of data centre connection enquiries lodged with Transgrid would, if realised, exceed NSW’s all-time peak grid demand. Continuously.

The demand is concentrating in Western Sydney, already one of the most congested parts of the transmission network. Transgrid is not standing still. A suite of near-term upgrades is underway — transformer upgrades at Sydney West and Holroyd, a new bulk supply point near Kemps Creek for the Western Sydney Aerotropolis, voltage stability works at Vineyard, and a new 330kV substation at Mt Druitt, spanning 2028 to 2031. According to the 2025 Transmission Annual Planning Report, these are sized to accommodate metropolitan Sydney load growth that includes 800 MW of data centre load. Not 16 GW.

The real unlock is Sydney Ring South

Sydney Ring South (SRS) is an actionable Integrated System Plan project that will deliver a new 500kV double-circuit transmission line connecting Bannaby to a proposed new South Creek substation, bringing Snowy 2.0, the South-West REZ’s 2.5GW, and interstate capacity from South Australia and Victoria via HumeLink and VNI West directly into metropolitan Sydney. When it is built, the congestion problem is largely solved.

The question is when — and the answer is not 2033.

Here is why.

SRS will have to follow the Regulatory Investment Test for Transmission (RIT-T) process.

The RIT-T is the economic assessment process TNSPs must run under the National Electricity Rules before building any major transmission project, requiring them to publish a Project Assessment Draft Report (PADR) identifying the preferred option, consult the market, then publish a Project Assessment Conclusions Report (PACR) for AER scrutiny. It exists to ensure consumers — who ultimately pay for transmission through their bills — are not carrying the cost of overbuilt or uneconomic network assets.

Stage 1 — PADR consultation: minimum 6 months, likely longer. The SRS Project Assessment Draft Report has not been published yet. In May 2025, the Australian Energy Regulator agreed to an extension by which Transgrid must publish the PADR for Sydney Ring South project to 30 April 2026, which has yet to be fulfilled. When Transgrid does publish, National Electricity Rules mandate a minimum 6-week consultation period. Given this project has to connect into metropolitan Sydney, expect a longer and more contested process.

Stage 2 — PACR published following PADR: estimate 18 months. Using VNI West as a comparable, the gap between PADR close and a final PACR has run to around 18 months.

Stage 3 — AER regulatory approval: 6–9 months minimum. Assume 9 months if no party disputes the preferred option. If challenged — and on a project of this scale, disputes are likely — add additional time to work through each submission carefully.

Stage 4 — EIS and planning approvals: 2 years at a minimum. This is where the schedule really breaks apart. Securing state and federal environmental approval for a high-voltage transmission corridor connecting into metropolitan Sydney and possible national park will not move quickly. The Snowy 2.0 Transmission Connection Project — a new 9km transmission line through Kosciuszko National Park — took 20 months from EIS exhibition to full dual-government approval. SRS is orders of magnitude more complex.

Stage 5 — Easement acquisition and land access: 3+ years. HumeLink is the benchmark. Transgrid spent over three years negotiating land access across a largely rural route. Transgrid’s Annual Planning Report notes Sydney Ring South is to connect between Bannaby – where HumeLink ends - and a new substation at South Creek, a suburb adjacent to the Western Sydney Airport. Residents at Bannaby have been engaged by Transgrid for the past 7 years on HumeLink and will likely be fatigued and resistant to another transmission project in their local community.

Stage 6 — Procurement and construction: 5+ years. Transmission equipment procurement runs 24 months for long-lead items alone. Construction of a project at this scale is a minimum of three years on top, in optimal conditions.

Add it up: if Transgrid publishes the SRS PADR in 2026, energisation before 2037 is optimistic.

There is a further complication no timeline model captures cleanly: the political cycle. Erecting 80-metre steel lattice towers required for a 500 KV transmission line into western Sydney’s South Creek is not an engineering problem — it is an electoral one. This will be a tough ‘sell’ to voters if the project RIT-T is expedited before the 2027 NSW state election.

What this means for data centre developers and investors

If your hyperscale data centre project is not already near the front of Transgrid’s Sydney West substation connection queue, the window for a commercially viable project has closed.

For those data centres not at the front of the queue, rerun your internal model against a realistic 2037 connection date. Does the equity IRR still clear your hurdle rate? Does your anchor tenant accept a four-year delivery slip without renegotiating rent? Does the renewable PPA you’ve already signed for 2033 hold its value when the load isn’t there to absorb it?

If any of those answers is no, the model in your investment committee paper is wrong.

Sydney Ring South is the most consequential transmission project for data centres in Australia. Follow the Bottleneck will continue to track its development and explore the opportunities available to data centres to actually accelerate a pathway to power.