Firmus Could Not Sell Its IPO at A$11. The Target Valued It at Nearly Three Times Its August Round.

Firmus, an Australian AI data-centre developer backed by Nvidia and Blackstone, closed the institutional bookbuild for its IPO on 8 October 2026 without clearing its marketed price of A$11 a share. Bloomberg reported that the company plans to postpone the listing and may consider a private funding round instead; other reports described the deal as in limbo. Several figures in the early coverage conflict, including the size of the cut, the amount being raised and the valuation, so this article uses only numbers we could tie to a named report and flags where reports disagree. The most useful comparison is one the coverage mostly leaves implicit: the IPO target against what investors paid two months earlier.

What happened, and where reports disagree

Data Center Dynamics reported on 7 October that Firmus had hoped to sell shares at A$11 and, according to the Australian Financial Review, cut the price to A$9 that day. It said the planned listing date on the ASX was 23 October and the prospectus was due on 12 October. On 8 October, AI Weekly, citing Bloomberg, reported that books closed after the deal failed to attract adequate demand at the A$11 marketing price. Investors flagged aggressive valuation for a company with two data centres, according to that report, and about 58% of shares were expected to be freely tradable on listing, which could flood first-day trading. Bloomberg, as relayed by CryptoBriefing, also said foreign investor interest declined during the process, and a portfolio manager, Jun Bei Liu, called the offering polarising, citing a gap between early international interest and firm commitments.

The disagreements matter. The raise is described variously as about A$5.5 billion, at least A$5 billion, or A$7 billion including over-allotment, and some reports leave the currency unstated. The valuation is given as about A$43.7 billion (Data Center Dynamics, equal to US$30.3 billion in its conversion, and Business News Australia) in some places and as nearly A$30 billion in another, which looks like a currency mix-up. We use A$43.7 billion and treat ‘nearly A$30 billion’ as unreliable. The price cut is A$9 per the AFR, an 18% reduction from A$11; Business News Australia relays a suggestion the price could fall as far as A$8, which would be a 27% cut. A 25% cut to A$8.25, which appeared in one summary we were given, was not in any source we could read, so we do not use it. Whether the deal was pulled, postponed or merely stalled also varies by outlet; Bloomberg says postpone, not cancel.

The valuation step-up

The comparison most readers will want is this. Data Center Dynamics reports that in August 2026 Firmus raised US$2 billion of equity at a post-money valuation above US$10.5 billion. Business News Australia lists the investors as Blackstone, Coatue, Nvidia and Jane Street. The IPO target of about A$43.7 billion, or US$30.3 billion in Data Center Dynamics’ conversion, is about 2.9 times the August figure. That step-up in roughly two months is the cleanest explanation for why the book would not clear: institutional buyers were being asked to pay nearly three times what sophisticated private investors had just paid. CryptoBriefing notes that the A$ and US$ figures make a direct comparison difficult, and that is right at the margin, but the order of magnitude survives any reasonable exchange rate.

A second data point sits beside it. Maas Group Holdings, an ASX-listed company, told investors it has paid a total of A$410 million for a 3.2% fully diluted interest in Firmus, including a A$300 million top-up in August, according to Business News Australia. Dividing 410 by 0.032 implies about A$12.8 billion for the whole company. That is an arithmetic illustration only, because the purchases happened at different dates and prices, but it sits close to Business News Australia’s own conversion of the August round to about A$15 billion. Against that, A$43.7 billion is roughly three times.

Built, sold and promised

The second quantity investors weigh is delivered capacity. Data Center Dynamics reports that Firmus’s collaboration with CDC Data Centres on Project Southgate could have delivered 1.6 gigawatts but delivered only 43 megawatts, and that reports earlier in the week said the collaboration had ended, with Firmus reportedly continuing the build without CDC. The article’s own URL says 42 megawatts, so the figure is slightly inconsistent. Separately, Reuters Breakingviews, summarised by TechStartups, said only 42 megawatts of a planned 1 gigawatt was operational. The two denominators differ, one a collaboration’s potential and one the company’s planned total, so the delivered share is somewhere between about 2.7% (43 of 1,600) and about 4.2% (42 of 1,000). Either way, the figure is a small single-digit percentage of what was promised.

Why does that gap matter for price? Contracted capacity is a promise. Operating capacity generates revenue and proves the company can build, power and cool a site on schedule. Public investors typically apply a heavy discount to the former, because it depends on construction timelines, power connections, equipment deliveries and the customer staying committed. Data Center Dynamics lists Meta (capacity in Indonesia and Malaysia) and OpenAI (capacity in Malaysia) as customer agreements, and CryptoBriefing says the facilities are being built to serve those two. Bloomberg’s list of investor concerns includes the company’s track record, which is another way of saying the delivered figure.

The Maas Group connection

Maas Group’s position is the most instructive part of the story, because it shows how an IPO stall travels. Its shares fell as much as 30% on 8 October, from a close of A$6.39 the day before to a low of A$4.47, and the ASX sent it a query asking it to explain. Maas told the ASX it knew of no undisclosed information explaining the trading and said speculation about the stalled IPO had influenced investor sentiment. It is also a supplier: Business News Australia reports that in August a Maas subsidiary received an estimated A$855 million order from Firmus for modular power and high-voltage equipment, delivered over 18 months, plus a separate A$200 million contract for a 100-megawatt AI factory in Launceston. At its 24 September AGM Maas had cited customer and counterparty concentration among its risks. When one company is both an investor and a supplier, a failed financing can hurt it twice: once on the value of the stake and once on the pipeline of orders.

What it signals, and what it does not

The wider debate is whether AI infrastructure can still raise public equity at the prices private rounds set. Reuters, via TechStartups, reported that Morgan Stanley estimates AI infrastructure could need about $1.5 trillion in external financing by 2028, which is why a failed A$5 billion-plus offering is read as a signal. One deal, in one market, is a thin base for that conclusion. Data Center Dynamics notes that other neoclouds, Nscale in the UK and Lambda in the US, are planning public listings, and the Bloomberg-sourced reports say Firmus is postponing rather than cancelling and weighing a private round. The company’s stated reasons are not in the sources we read, since Data Center Dynamics said it contacted Firmus for comment and the price cut rests on the AFR report.

A fair reading is therefore narrower than the headlines. Investors declined to pay a price that assumed a three-fold step-up and rapid delivery, from a company with two data centres and a small fraction of its promised capacity operating. That is a statement about this price and this track record. It is not evidence that demand for AI capacity has gone, and it does not show that a more modestly priced listing, or one from a company with more capacity online, would fail. The window may be narrower, not shut.

What the evidence does not establish

We have not seen the prospectus, Firmus’s own statement or the full Bloomberg and AFR reports. Claims that Firmus has a US$10 billion Blackstone-led debt package, that its built capacity is about 5% of sold capacity compared with 25% at NextDC, and that it partners with DayOne on an Indonesian site appeared in a research summary but not in any source we could read, so we have left them out. The valuation, raise size and cut are reported inconsistently, and the delivered-capacity ratio depends on which denominator is correct. This analysis is current as of 11 October 2026.