This is investment research, not personal financial advice.
Centuria Capital Group rose 6.0% on 13 August, from A$1.49 to A$1.58, after announcing that its half-owned ResetData venture had signed for 7MW of data-centre capacity and documented A$165 million of GPU bridge finance. Turnover rose to about 6.5 million securities, more than four times the prior session's volume. The market added roughly A$90 million to Centuria's equity value in one day (ASX 2026; Centuria 2026b).
The reaction recognised genuine progress. ResetData now has a named data-centre provider, a named lender and ordered hardware. Three practical bottlenecks have moved closer to resolution.
But the announcement did not disclose one new binding end-customer contract, a minimum spend, a project margin or Centuria's full guarantee exposure. The 7MW master services agreement gives ResetData a place to deploy GPUs. It does not prove that customers will use them profitably. The 3MW addition in the headline was a six-week reservation, not a completed long-term order. A prior 13MW heads of agreement had already been abandoned.
That makes the 6% rise broadly proportionate as an execution relief rally, but too early to treat as an earnings reset. At A$1.58, Centuria's established funds-management and co-investment platform can explain most of the valuation. ResetData only earns a large part of the remainder when paying utilisation, refinancing and project returns appear together.
Seven signed megawatts, three reserved megawatts
The legal quality of each number matters more than the total. ResetData's agreement with CDC Data Centres covers an initial 7MW under an executed master services agreement. The extra 3MW came through a letter of intent that reserved capacity for six weeks. If the parties did not execute a binding order form by the start of September, that reservation would expire (Centuria 2026b).
The same distinction applies on the customer side. Centuria referred to an unnamed investment-grade corporation that might procure about 2MW at Centuria-owned facilities, but only under a non-binding memorandum. Other enquiries were still in commercial and technical discussions. The announcement's own wording said revenue could expand as demand converts into contracted capacity.
So the stack at 13 August looked like this:
| Disclosure | Capacity | Contract quality | What remains missing |
|---|---|---|---|
| CDC master services agreement | 7MW | Executed capacity arrangement | Customer offtake, price, term, utilisation commitment |
| CDC letter of intent | 3MW | Six-week reservation | Binding order form before September |
| Potential corporate customer | about 2MW | Non-binding memorandum | Full commercial agreement and credit terms |
| Existing AI-F1 customer | first 0.7MW phase disclosed as leased | Revenue-generating, terms undisclosed | Identity, tenure, price and concentration |
| Wider infrastructure pipeline | more than 250MW | Development pipeline | Sites, approvals, power, customers and project finance |
This is better evidence than the prior 13MW heads of agreement, which Centuria chose not to pursue after identifying larger alternatives. It is also smaller in firm near-term megawatts. The episode is a useful warning: pipeline can improve in quality while shrinking in quantity.
The deployment table showed roughly 0.7MW operating at AI-F1, about 0.15MW at another Centuria facility and about 2MW at CDC. Targets totalled about 10.6MW before the conditional 3MW. Hardware comprised NVIDIA H200 and B300 systems. Customer revenue from the larger sites was not expected until the second half of FY27, leaving finance costs and equipment ageing ahead of much of the disclosed revenue ramp (Centuria 2026b).
A$165 million funds equipment, not demand
Macquarie Bank's Specialised and Asset Finance division signed facility documents for A$165 million of senior bridge GPU finance. W.Media quoted a Macquarie executive confirming the lender's involvement, which gives independent support to the identity of the financing party (W.Media 2026).
The facility is a limit, not an amount already drawn, and certainly not revenue. Centuria did not disclose pricing, tenor, amortisation, security, advance rates, covenants or the refinancing deadline. Parent-company support may become available subject to customer contracting, senior-debt refinancing and other conditions. Dell Financial Services also funds existing procurement with Centuria support.
That structure creates an awkward timing question. GPUs lose economic value quickly as newer hardware arrives. A bridge loan assumes that contracted cash generation or longer-dated refinancing appears before the assets become less competitive. If customer utilisation lags, debt remains while the earning power of an H200 or B300 cluster falls.
The power announcement reaches further into the future. Centuria said 72MW of dedicated generation units had been secured for 2028 delivery, within a pipeline exceeding 250MW. No supplier, purchase price, fuel source, site, ownership vehicle, planning status or connection arrangement was disclosed. Running 72MW continuously would consume 631GWh a year before cooling and facility overhead, although actual use will depend on load, efficiency and whether the units provide primary or backup power. That scale is almost one-sixth of the estimated 3.9TWh used by all Australian data centres in FY25 (Oxford Economics 2025).
Australia does need more data-centre power. AEMO estimated that NEM data centres used about 4TWh in 2024-25 and projected 21.4TWh by 2034-35 in its central Step Change case. Yet AEMO also warned that connection applications can double-count projects or include developments that never proceed (AEMO 2025). ResetData's 250MW pipeline belongs in that uncertain category until sites, permits, customers and finance converge.
The property platform still carries the valuation
Centuria is not a pure AI infrastructure company. It manages property funds, makes co-investments, provides property and development finance through Centuria Bass Credit, and runs an investment-bonds business. ResetData is a 50%-owned venture added in 2024 for consideration of up to A$21 million, including working capital and earnings-linked payments (Centuria 2026d).
The established platform had A$21.8 billion of AUM at December 2025: A$18.3 billion in property funds, A$2.5 billion in real-estate finance and A$1.0 billion in investment bonds. It also had more than 15,500 active private investors and over 1,200 advisers. Repeat participation matters. Some 1,600 investors held at least three Centuria funds, which lowers the distribution burden when the group launches another vehicle (Centuria 2026e).
That network is a defensible advantage in property. It is not automatically transferable to GPU cloud services. The overlap lies in sites, power procurement, development and financing, not in proof of customer demand for accelerated computing.
Centuria's 50% stake also changes how AI disclosures should be read. ResetData's revenue, capacity and project values are generally stated at 100%. CNI ordinarily receives half the equity economics. Parent guarantees can run the other way and expose the listed group to more than half of a project shortfall. Meanwhile, a data-centre building may sit inside a Centuria-managed fund, producing management fees and co-investment income rather than the full property return for CNI.
Four years show a dependable franchise without per-security compounding
The financial record separates the stable platform from the new story. CNI's management-defined operating revenue rose from A$292.6 million in FY22 to A$355.4 million in FY25. Operating profit, however, fell from A$114.5 million to A$100.8 million. OEPS declined from 14.5 cents to 12.2 cents while AUM ended FY25 exactly where it began FY22, at A$20.6 billion (Centuria 2022; Centuria 2025).
| Reporting period | Operating revenue | Operating NPAT | Statutory NPAT | AUM | OEPS | Distribution | Operating gearing | Interest cover |
|---|---|---|---|---|---|---|---|---|
| FY22 | A$292.6m | A$114.5m | (A$37.9m) | A$20.6bn | 14.5c | 11.0c | 13.2% | 6.8x |
| FY23 | A$311.6m | A$115.6m | A$105.9m | A$21.0bn | 14.5c | 11.6c | 10.6% | 5.0x |
| FY24 | A$305.0m | A$94.7m | A$102.1m | A$21.1bn | 11.7c | 10.0c | 12.1% | 4.0x |
| FY25 | A$355.4m | A$100.8m | A$82.7m | A$20.6bn | 12.2c | 10.4c | 12.3% | 3.8x |
| 1H FY26 | A$195.5m | A$54.6m | A$49.8m | A$21.8bn | 6.6c | 5.2c | 12.4% | 4.1x |
All figures in the table are company-reported except the roic_pct entries in frontmatter. Those are author-computed return-on-book-equity proxies, calculated as operating NPAT divided by average book net assets, or closing book net assets for FY22. They fell from 8.35% in FY22 to 6.81% in FY25. The label is deliberately conservative: CNI consolidates controlled vehicles and carries large intangible and co-investment balances, so a conventional industrial ROIC denominator would imply precision that the structure does not support.
The FY25 statutory revenue presentation also changed because interest revenue appeared separately. The operating-revenue series is the cleaner comparison across years. Statutory NPAT is volatile because property revaluations and other non-operating items move through the accounts. That is why operating NPAT matters here, but it should not be mistaken for cash.
HY26 improved. Operating NPAT reached A$54.6 million, AUM hit A$21.8 billion and operating interest cover recovered to 4.1 times. Base management fees barely changed, at A$70.6 million versus A$69.8 million. Transaction fees rose from A$1.5 million to A$9.2 million and performance fees from A$3.9 million to A$11.1 million. The stronger half therefore included a large cyclical contribution. Sovereign AI lost A$5.9 million at EBITDA level while it scaled (Centuria 2026e).
Owner cash narrowed before the equity raise
A cash bridge is more revealing than the operating-profit label. Using CNI's operating-segment cash flow and subtracting purchases of property, plant and equipment gives an author-computed owner-cash proxy. Co-investment purchases and disposals sit outside this bridge because they are capital allocation rather than maintenance spending.
| A$m | FY22 | FY23 | FY24 | FY25 |
|---|---|---|---|---|
| Operating cash flow | 182.1 | 83.4 | 122.0 | 128.4 |
| Less PP&E purchases | (2.7) | (2.3) | (0.2) | (33.9) |
| Owner cash before co-investment recycling | 179.4 | 81.1 | 121.8 | 94.5 |
| Cash distributions | (90.5) | (93.5) | (86.2) | (85.8) |
| Residual after distributions | 88.9 | (12.4) | 35.6 | 8.7 |
FY25 owner cash was close to operating NPAT, but only A$8.7 million remained after distributions. The A$33.9 million of PP&E purchases captured the start of AI infrastructure spending. This calculation does not charge the further cash tied up in fund underwriting or co-investments, so it is not a generous definition of surplus capital.
Centuria then raised A$300 million in June 2026 through a placement and entitlement offer at A$2.00. About 150 million new securities increased the pre-raising base by 17.6%. Net proceeds of A$292 million were assumed to repay debt initially, taking company-pro-forma operating gearing from 12.4% to 3.4%, before capital was redeployed into property strategies, credit and ResetData (Centuria 2026c).
That reset improves near-term liquidity but raises the per-security hurdle. A A$100 million earnings base spread across 1,001 million securities is about 10 cents each. Absolute earnings can rise while per-security earnings remain below FY22 if the new capital does not earn enough. The A$1.58 closing price also sits below the A$2.00 issue price only weeks after the raising.
Headline corporate gearing understates economic exposure. At HY26 CNI had A$288 million of cash and undrawn debt, operating gearing of 12.4% and look-through gearing of 37.9%. Property vehicles carried non-recourse debt, and the listed parent may support GPU finance. Legal non-recourse limits matter, but lower property values can still reduce distributions, fee income and refinancing flexibility at the same time.
A sum of the parts leaves ResetData on probation
A conventional earnings multiple alone double-counts co-investment income if the underlying assets are also added. The valuation therefore has three parts: normalised EBITDA for funds management, private credit and investment bonds; a discounted value for property co-investments; and a separate probability-weighted ResetData value. Corporate debt, post-raising cash and commitments are adjusted around those pieces.
FY25 segment EBITDA before property investments and sovereign AI was about A$89 million: A$59.6 million from property funds management, A$27.0 million from property and development finance and A$2.6 million from investment bonds. HY26 annualises higher, but transaction and performance fees explain much of the improvement. A base range of A$85 million to A$95 million is more appropriate for normalised core EBITDA.
Applying 9 to 10.5 times gives A$800 million to A$1.0 billion for those operating businesses. The multiple recognises recurring fees and distribution, while discounting flat FY22-FY25 AUM, co-investment intensity and cyclical fees. Charter Hall's FY25 A$84.3 billion of FUM, A$271.5 million of funds-management EBITDA and 6.0% balance-sheet gearing show why a larger property manager can sustain a scale premium over CNI (Charter Hall 2025).
Centuria's proportionately consolidated property investments were about A$1.41 billion at HY26. After A$573 million of non-recourse loans and an A$81 million valuation adjustment, equity anchored near A$753 million. Applying discounts for illiquidity, office exposure and look-through leverage supports A$575 million to A$675 million in the base case.
ResetData cannot yet be valued with a project DCF because GPU cost, debt terms, customer price, contract tenure, utilisation, power cost, maintenance capex and margins are not disclosed. A probability-weighted proxy is the honest alternative. CNI previously indicated that a fully used 1.1MW AI-F1 could generate A$15 million of annual revenue on a 100% venture basis, or about A$13.6 million per MW (Centuria 2026d).
Assume, for analysis rather than as company forecasts, a 20% to 40% mature EBITDA margin, a 10-times project multiple and 50% CNI ownership. Before debt and probability adjustments, the value attributable to CNI is:
| Live paying capacity | 20% margin | 30% margin | 40% margin |
|---|---|---|---|
| 3MW | A$0.04 per security | A$0.06 | A$0.08 |
| 7MW | A$0.10 | A$0.14 | A$0.19 |
| 10MW | A$0.14 | A$0.20 | A$0.27 |
| 20MW | A$0.27 | A$0.41 | A$0.54 |
These figures use live, paying megawatts, not infrastructure reservations. They also ignore project debt, guarantees and future equity. The table explains why 250MW cannot be inserted into valuation: even 10MW needs customers, utilisation and positive margins before it creates meaningful per-security value.
What A$1.58 already assumes
At A$1.58 and 1,001.394 million securities, CNI's market value is A$1.582 billion. A reverse sum of the parts helps identify what is embedded:
- A$90 million of normalised core EBITDA at 8 to 10 times implies A$720 million to A$900 million.
- Discounted co-investments contribute roughly A$550 million to A$650 million.
- Post-raising corporate cash, debt and commitments are treated around nil pending deployment.
- The established-business subtotal is therefore A$1.27 billion to A$1.55 billion.
The residual for CNI's half of ResetData is A$30 million to A$310 million, depending on where the core sits in those ranges. The midpoint is about A$145 million, equivalent to A$290 million for all of ResetData before project-level debt.
Four scenarios show how wide the uncertainty remains:
| Scenario | Main assumptions | Value per security |
|---|---|---|
| Severe downside | Core EBITDA falls to A$75-85m, property marks weaken and guarantee exposure creates negative ResetData value | A$0.70-A$1.05 |
| Bear | Flat AUM, 7.5-9x core EBITDA, deep co-investment discounts, little ResetData value | A$1.00-A$1.42 |
| Base | A$85-95m core EBITDA, conservative asset marks, A$100-250m for CNI's ResetData stake | A$1.47-A$1.97 |
| Bull | Better capital flows and several customer-backed AI deployments, with A$300-600m for ResetData | A$2.05-A$2.75 |
The ranges are outputs of driver assumptions, not destination prices. They were built independently and then compared with A$1.58. The current price sits inside the base range because a credible core can carry most of the valuation while leaving a moderate allowance for AI optionality.
The anti-thesis is harsher than simple project failure. ResetData could become a working, growing compute business and still earn a poor return after GPU depreciation, bridge interest, customer ramp costs, parent guarantees and dilution. Centuria's property distribution network is valuable, but it has not yet demonstrated a moat in selling compute. Nor did FY22-FY25 establish per-security compounding in the core.
Three clocks now decide whether capacity becomes value
The first clock expires in early September 2026. The extra 3MW CDC reservation either becomes a binding order or disappears. More important than that option is evidence of at least 3MW of paying customer capacity by December, backed by contract duration and minimum-spend protection.
The second clock runs through the second half of FY27. ResetData needs utilisation above 60% of installed capacity and EBITDA break-even by June 2027. Disclosure should include the Macquarie draw, refinancing terms, Dell exposure and the maximum CNI guarantee. Revenue without those details cannot establish an equity return.
The third clock is slower. The 72MW generation units are due in 2028. Before delivery, investors need a site, approvals, customer contracts, fuel and emissions details, and financing without open-ended parent recourse. Until those pieces exist, 72MW is a procurement milestone rather than an earning asset.
The core has nearer tests. FY26 results are scheduled for 27 August. They should reveal actual post-raising liquidity, the split between base and cyclical fees, ResetData's loss and whether operating cash covered distributions. Over FY27, base management fees need to grow faster than 3%, operating interest cover should remain above 3 times, and gearing should stay below 15% unless new deployments carry binding earnings.
Source notes: confidence stops at the private contracts
Confidence is high for the filed financial history and the terms Centuria chose to disclose. Confidence is lower for ResetData economics because the decisive contracts are private. No public CDC contract, Macquarie facility agreement, GPU purchase schedule or power-equipment contract was available. The independent event coverage confirms named-party involvement, not the missing commercial terms. ABR records confirm active ResetData entities, including two entities with MSA in their names, but not ownership or contract economics (ABR 2026).
The May ResetData briefing was retrieved and read through Centuria's investor-feed endpoint, but that reader URL returns HTTP 403 to the gate's liveness probe. The article therefore carries verification: partial despite the document having been opened. The valuation uses filed figures for the core, author calculations for owner cash and return on book equity, and explicit assumptions for ResetData margins, utilisation and multiples. It assigns no valuation-grade status to the 250MW pipeline. Those limits are the missing-information boundary, not a claim that the contracts are unfavourable.
Centuria's 6% rise was a reasonable response to fewer execution obstacles. It was not proof that 7MW has become a profitable customer book. The market now prices the property platform plus a measured ResetData option. September's order form, FY27 utilisation and disclosed guarantee terms will decide whether that option matures before the financed hardware does.
References
- ABR 2026: Australian Business Register active-entity search for ResetData.
- AEMO 2025: Australian Energy Market Operator, 2025 Electricity Statement of Opportunities.
- ASX 2026: ASX company and market page for Centuria Capital Group (CNI), 13 August 2026.
- Centuria 2022: Centuria Capital Group FY2022 Annual Report.
- Centuria 2023: Centuria Capital Group FY2023 Annual Report.
- Centuria 2024: Centuria Capital Group FY2024 Annual Report.
- Centuria 2025: Centuria Capital Group FY2025 Annual Report.
- Centuria 2026a: Appendix 2A confirming CNI's quoted securities, 14 July 2026.
- Centuria 2026b: Centuria and ResetData advance AI Factory deployment, 13 August 2026.
- Centuria 2026c: A$300 million equity-raising investor presentation, 22 June 2026.
- Centuria 2026d: ResetData briefing, 28 May 2026.
- Centuria 2026e: HY2026 financial report, 25 February 2026.
- Charter Hall 2025: Charter Hall Group FY2025 Annual Report.
- Oxford Economics 2025: Data Centre Energy Demand Final Report prepared for AEMO.
- W.Media 2026: Report on Centuria, ResetData, CDC capacity and Macquarie GPU finance, 13 August 2026.