Firmus IPO Seeks $5 Billion as Draft Forecast Shows a $77 Million Loss

Firmus is preparing a $5 billion IPO while forecasting a $77 million after-tax loss for the first half of fiscal 2027. That contrast turns the Firmus IPO into a direct test of how public investors value unfinished AI infrastructure.

The figures come from a draft prospectus circulated to prospective investors, according to two people who spoke with Reuters. The document reportedly describes Firmus as historically loss-making and provides no forecast beyond the current half.

Firmus is not asking investors to finance a conventional data center operator with a mature portfolio. It is asking them to fund an accelerated buildout across Australia and the wider Asia-Pacific region. Only two of its seven identified facilities are operational, while five remain under development.

That creates the central conflict behind the proposed listing. Firmus says customer commitments and AI computing demand justify building at unusual speed. Public investors must decide whether those future contracts outweigh the losses, construction risks, and capital requirements visible today.

What the Firmus IPO Draft Reportedly Reveals

The reported loss matters because Firmus wants investors to value infrastructure that is still moving from contracts and construction plans into operating assets.

Firmus expects a pro forma loss after tax of $77 million for the first half of the financial year ending June 30, 2027. Pro forma figures adjust reported accounts to reflect specified assumptions or transactions, so investors will need the final prospectus to understand the calculation.

Reuters reported that two people familiar with the draft prospectus supplied the forecast. Those sources said the document was circulated to prospective investors during the week of September 21.

The draft reportedly provides no forecasts after the first half of fiscal 2027. That limited window leaves investors without a formal public earnings path covering most of Firmus’s planned development schedule.

The company declined to comment on the draft. Its public offer document had not been released when the figures emerged, preventing outside readers from checking its assumptions, accounting adjustments, or risk disclosures.

The current schedule would begin institutional bookbuilding on October 6. Bookbuilding is the process through which banks collect investor orders and use that demand to set an offering price.

The prospectus is expected to become public on October 8, followed by an Australian Securities Exchange debut on October 22. Those dates give investors a short period to assess a company with facilities, contracts, suppliers, and financing arrangements across several countries.

Firmus intends to raise approximately US$5 billion, according to the term sheet described by Reuters. Dealogic data cited in the report would make the transaction Australia’s second-largest IPO, behind Telstra’s US$10 billion listing in 1997.

Local reports have suggested a post-listing valuation as high as US$60 billion. That figure remains unconfirmed because the final pricing range, share count, and prospectus have not yet been published.

The distinction is important. The US$5 billion represents proposed new capital, while the reported US$60 billion represents the potential value assigned to the whole company after listing.

Firmus reportedly plans to use the IPO proceeds for additional capital expenditure. That purpose aligns with its business model, which requires land, power connections, cooling systems, buildings, networking equipment, and large numbers of AI accelerators before revenue can scale.

The draft prospectus reportedly attributes historical losses to the cost of developing the business and preparing it to sign substantial customer agreements. That explanation is plausible for an infrastructure developer, but it is not enough to establish future returns.

Investors still need to see how much capital each facility requires, when contracted revenue begins, and which party carries delays or equipment costs. They also need to know whether customer commitments contain minimum payments, cancellation rights, or performance conditions.

The final prospectus should turn those questions into measurable disclosures. Until then, the $77 million forecast is not merely a headline loss. It is the clearest available indicator of the financial gap between Firmus’s operating position and its proposed valuation.

Why Firmus Needs Public Capital Now

Firmus is trying to finance a development schedule that has expanded faster than its operating footprint.

The company currently has two operational data centers, located in Australia and Singapore, according to Reuters. Five more facilities are under development across the Asia-Pacific region, with most still in early stages.

That ratio explains the timing of the Firmus IPO. Operating sites can generate revenue, but projects under construction consume cash before they can serve customers.

Firmus has already raised substantial private capital. In April 2026, it announced a US$505 million strategic equity investment led by Coatue, with participation from Nvidia subject to closing conditions.

An additional US$2 billion equity investment was announced in August. Investors reportedly included Blackstone, Jane Street, Coatue, and Nvidia, placing Firmus among the region’s most heavily financed private AI infrastructure companies.

The company has also described a US$10 billion debt facility led by Blackstone and Coatue. Debt can accelerate construction, but it introduces interest obligations and financing conditions that equity investors must examine carefully.

A public offering would add another funding source without requiring Firmus to finance its entire expansion through debt. It would also give early shareholders a publicly traded valuation for their holdings.

Firmus’s largest public initiative is Project Southgate, a planned network of AI-focused facilities across Australia. The company calls these facilities AI factories because they are designed around dense accelerator clusters rather than general-purpose colocation.

The first Southgate developments included sites in Melbourne and Tasmania. Firmus later expanded the planned network to Sydney, Canberra, Perth, and other locations.

In its October 2025 Southgate expansion announcement, Firmus described a target of 1.6 gigawatts of operational capacity through 2028. The company also cited a possible total investment of A$73.3 billion across the broader program.

Those are development targets, not completed assets. Each site still depends on construction, power delivery, equipment installation, customer acceptance, and reliable operation.

The equipment burden is especially large because Firmus is building around Nvidia accelerators. These systems require more than access to chips. They need high-density power distribution, liquid cooling, fast networking, and software capable of managing large clusters.

Firmus says its modular HyperCube design can reduce energy consumption and construction costs compared with traditional data centers. Those remain company claims that investors must evaluate against actual operating data.

The financing requirement also reflects how quickly accelerator generations change. A facility designed for one Nvidia platform can face different power, cooling, and networking requirements when newer systems arrive.

Firmus must therefore build quickly enough to serve contracted customers while avoiding infrastructure that becomes commercially less attractive before achieving acceptable utilization. That timing risk becomes more important when several facilities are developed simultaneously.

The IPO does not remove that risk. It would transfer a portion of it from private investors and lenders to public shareholders.

That shift is the real reason the timing matters. Firmus has reached a scale at which future growth requires public-market-sized capital, yet much of the infrastructure supporting its valuation remains unfinished.

Contracts Must Catch Up With Construction

The primary contest is between contracted future demand and the present cost of delivering the capacity behind it.

Firmus has announced customers and partners that give its expansion greater credibility than a purely speculative data center plan. Reuters identified Nvidia, Meta, and OpenAI among its customers.

In September 2026, Firmus said it had surpassed 900 megawatts of contracted capacity. It also announced OpenAI as an anchor customer for two planned facilities in Malaysia.

Contracted capacity can provide a valuable demand signal. It does not automatically equal recognized revenue, collected cash, or profit.

Investors must determine what “contracted” means in each agreement. A binding minimum payment offers more protection than an expression of demand, a flexible reservation, or a contract dependent on construction milestones.

Firmus has disclosed several arrangements that illustrate the scale it wants to reach. In March, it announced a multi-year agreement with an unnamed global technology company covering approximately 18,400 Nvidia GB300 GPUs at its Melbourne facility.

The company described that agreement as a multibillion-dollar commitment. It also called it the second large-scale customer secured for Project Southgate.

In June, Firmus announced a planned 360-megawatt campus in Batam, Indonesia, covering up to 170,000 Nvidia accelerators. Its Batam partnership runs through 2034 and includes a revenue-sharing arrangement with Nvidia.

Firmus said customer commitments could generate between US$25 billion and US$30 billion during the partnership’s first six years. That range is management’s expectation, not independently verified revenue.

The arrangement is notable because Nvidia occupies several positions in the Firmus story. It supplies the accelerators, has invested in the company, supports its infrastructure design, and can earn a share of cloud revenue from supported capacity.

That alignment can help Firmus secure hardware and customers. It can also make the economics harder to evaluate because equipment purchases, equity ownership, and revenue sharing interact.

The draft prospectus reportedly estimates that Firmus’s developed data centers will produce US$5 billion in combined annual earnings within five years. Reuters attributed that figure to a third person familiar with the plans.

The word “developed” does considerable work in that forecast. It assumes the facilities reach completion, receive power, install equipment, pass customer testing, and achieve enough utilization to support the projected earnings.

The reported $77 million loss and the five-year earnings estimate are therefore not contradictory. They describe opposite ends of the same development plan.

The critical question is the path between them. Investors need annual or project-level information showing how construction spending becomes available capacity, contracted capacity becomes billable usage, and billable usage produces cash.

The absence of forecasts beyond the first half of fiscal 2027 makes that bridge harder to inspect. A five-year destination offers less guidance when the intervening milestones remain unclear.

Firmus can strengthen its case by disclosing contracted revenue duration, customer concentration, required capital, commissioning dates, and expected utilization. Without those details, headline capacity numbers can make the business appear more mature than it is.

The IPO is therefore a wager on execution rather than a simple bet on AI demand. Demand can remain high while an individual operator still experiences delays, cost overruns, financing pressure, or weak margins.

Firmus Faces a More Demanding Public Comparison

Public investors can compare Firmus with established data center operators that already disclose operating, financing, and construction results.

Australia is not entering the data center investment cycle for the first time. The market includes established operators such as NEXTDC and AirTrunk, alongside newer AI-focused infrastructure businesses.

NEXTDC provides one useful benchmark because it is already listed on the ASX. Investors receive regular information about utilization, development spending, revenue, debt, and commissioning progress.

The company reported A$1.285 billion of capital expenditure for the half-year ending December 31, 2025. Spending covered developments in Sydney, Melbourne, Kuala Lumpur, and other expansion projects.

In April 2026, NEXTDC said contracted utilization had reached 667 megawatts. Its forward order book reached 544 megawatts, while the company raised its annual capital expenditure guidance.

Those figures show that heavy spending is not unusual in this sector. The difference is that an established public operator provides recurring disclosures that let investors compare capital deployment with operating progress.

Firmus will face the same expectation after listing. Announcements about total planned capacity will matter less than quarterly evidence showing delivered megawatts, customer acceptance, utilization, and cash generation.

AirTrunk provides a different reference point. A Blackstone-led consortium acquired the Asia-Pacific data center operator at an enterprise value exceeding A$24 billion in 2024.

The AirTrunk acquisition covered an established regional platform with operations across Australia, Japan, Malaysia, Hong Kong, and Singapore. Its valuation reflected operating assets and a substantial development pipeline.

A reported Firmus valuation of up to US$60 billion would place the company above that transaction value. Direct comparisons remain imperfect because currencies, ownership structures, growth rates, and business models differ.

Firmus also sells access to accelerator-based computing rather than relying only on traditional colocation. That model can produce more revenue per megawatt, but it exposes the operator more directly to hardware cycles and cloud-service margins.

Traditional data center operators often lease powered space under long contracts. A neocloud, meaning a cloud provider focused on AI accelerators, can bundle infrastructure with GPUs, networking, and managed computing services.

That added layer creates opportunity and risk. Firmus can capture more value if demand stays high and customers use the installed systems extensively.

It also carries more technology exposure. Accelerator prices, performance, availability, financing terms, and resale values can change faster than the underlying buildings.

Blackstone’s role adds another dimension. It backed the AirTrunk acquisition and has also supported Firmus through financing and investment.

That does not make the two businesses equivalent. It shows that sophisticated capital providers are willing to fund several approaches to Asia-Pacific data center growth.

Public shareholders will still demand their own evidence. The reputation of private backers cannot replace audited accounts, contractual disclosures, or project-level delivery.

Firmus’s customer list gives it a compelling market story. NEXTDC and AirTrunk give investors alternatives with longer operating records.

That comparison raises the standard for the Firmus IPO. The company must show why its faster, accelerator-centered model deserves a valuation based on future earnings before most planned facilities enter service.

The $77 Million Loss Is Only One Risk

The headline loss is measurable, but the larger uncertainties concern capital intensity, delivery timing, customer concentration, and forecast visibility.

A loss during rapid construction does not establish that Firmus has a weak business. Infrastructure companies often spend heavily before new assets produce revenue.

The loss also does not prove that future facilities will generate acceptable returns. That depends on the economics of individual projects and the contractual protection attached to customer demand.

The first concern is construction risk. Firmus is developing several facilities across different jurisdictions, each with its own approvals, contractors, supply chain, and energy requirements.

A delay in one project can postpone customer revenue while financing and staffing costs continue. Simultaneous projects can magnify the effect because management must solve several delivery problems at once.

The second concern is power. AI data centers require large, dependable electricity supplies, and grid connections can take years to secure.

Firmus has tied its Australian strategy to renewable generation and liquid-cooled systems. Those features can improve efficiency, but they do not eliminate transmission constraints or the need for firm power.

Its Tasmania plan initially described 90 megawatts of capacity during 2026, with another 300 megawatts subject to final approvals. The larger 1.6-gigawatt Southgate target extends across several regions and depends on extensive new infrastructure.

The third concern is customer concentration. Large AI facilities often depend on a small number of customers because few buyers can commit to tens of thousands of advanced accelerators.

A large contract can support project financing and reduce demand risk. It can also give the customer bargaining power and make the operator vulnerable to one buyer’s strategy or financial condition.

Investors need to know how Firmus allocates capacity among Nvidia, Meta, OpenAI, and other customers. They also need the duration, renewal terms, payment security, and termination provisions attached to major agreements.

The fourth concern is equipment economics. New accelerator generations can increase performance while changing power and cooling requirements.

Firmus says its Batam campus will support Grace Blackwell, Vera Rubin, and Vera systems through 2027 and 2028. Managing several generations can broaden customer options, but it complicates procurement and asset planning.

The fifth concern is forecast visibility. The reported draft ends its formal financial outlook after the first half of fiscal 2027, while the company’s most ambitious earnings claim extends five years.

That gap leaves investors with a near-term loss and a distant destination. The final prospectus needs enough intermediate information to test whether the destination is credible.

ABC News raised a related caution before the draft details emerged. Its IPO review noted that valuation reports were based partly on roadshow material and market conjecture because no public prospectus had been lodged.

That verification gap should narrow on October 8. Investors will then be able to inspect audited financial history, material contracts, debt obligations, related-party transactions, and formal risk factors.

The $77 million loss will remain important, but its meaning will depend on those disclosures. A planned loss attached to fully financed, protected contracts differs from a loss supporting projects whose capital or demand remains conditional.

Until the prospectus arrives, both interpretations remain possible. Firmus has assembled major customers, investors, and suppliers, but the public evidence does not yet reveal the complete economics.

Three Signals Will Decide the Firmus IPO Story

The prospectus, institutional order book, and project delivery record will determine whether Firmus can convert AI demand into a credible public-market business.

The first signal is the public prospectus expected on October 8. It should disclose the assumptions behind the $77 million pro forma loss and reconcile those figures with the company’s historical accounts.

Investors should look beyond total contracted capacity. The most useful disclosures will cover recognized revenue, operating cash flow, capital expenditure commitments, available liquidity, debt conditions, and the timing of customer payments.

Contract terms deserve close attention. Firm commitments, deposits, minimum usage requirements, and termination penalties would strengthen the connection between planned facilities and future cash.

Conditional reservations or agreements tied to construction milestones would leave more risk with Firmus. The prospectus should also show whether the company must buy accelerators before receiving customer payments.

The second signal is the institutional bookbuild scheduled to begin October 6. Strong demand at the proposed valuation would show that major investors accept the gap between near-term losses and projected earnings.

Weak demand, a lower valuation, or a smaller offering would suggest that investors want a greater margin of safety. Such changes would not invalidate the business, but they would reprice its execution risk.

The final allocation will matter too. Long-term infrastructure funds can bring a different investment horizon from short-term buyers seeking exposure to an active AI trade.

The third signal is operational delivery during the months after listing. Firmus needs to commission planned capacity on schedule and show that customers are using it.

The Melbourne facility offers an early test because Firmus announced a large deployment there and linked it to a multi-year customer agreement. Progress in Tasmania and Malaysia will show whether the company can repeat delivery across markets.

The most useful operating indicators will include energized capacity, installed accelerators, accepted customer clusters, utilization, and revenue-producing megawatts. Announced capacity alone cannot measure execution.

Firmus should also report whether construction and equipment spending remain within its original budgets. A project can enter service and still produce weak returns if costs rise faster than contracted revenue.

The company’s public-market challenge is straightforward even if its infrastructure is technically complex. It must demonstrate that signed demand creates durable cash before financing costs consume the expected returns.

That test matters beyond Firmus. A successful listing would give Australian investors a large, publicly traded vehicle focused directly on AI computing infrastructure.

It could also influence financing for other regional developers. A disappointing offering would make investors more selective about capacity claims that extend years beyond currently operating assets.

The Firmus IPO is therefore not simply a referendum on whether artificial intelligence needs more computing power. That demand is already visible across cloud providers, model developers, and accelerator suppliers.

The harder question is who captures the returns after paying for chips, buildings, power, cooling, financing, and revenue-sharing agreements. Firmus wants public investors to believe its contracts and architecture answer that question.

The final prospectus must supply the missing evidence. Readers should watch whether its contract protections, cash requirements, and delivery timetable support the valuation being discussed.

If those disclosures create a credible bridge from the reported loss to operating cash, the IPO case becomes stronger. If the bridge depends mainly on distant capacity and earnings targets, the $77 million loss will look less like a temporary construction cost and more like an early warning.

When the prospectus becomes public, focus on three questions: How much additional capital will Firmus need, when do binding customer payments begin, and how much capacity will actually operate during fiscal 2027? Those answers will determine whether the Firmus IPO finances a durable AI infrastructure platform or asks public shareholders to absorb too much development risk.