NextFin

Bankers in Failed Firmus IPO Probed by Australian Regulator

Summarized by NextFin AI
  • ASIC is making targeted inquiries into the joint lead managers of Firmus Grid Ltd.'s collapsed A$7.1 billion IPO, stopping short of a formal investigation but signaling regulatory scrutiny of AI listing salesmanship.
  • Firmus withdrew its float on Oct. 9 after bookbuilding failed to clear the marketed A$11 per share price, with price talk sliding to around A$8.25 before the company pivoted to a Nasdaq listing and private fundraising.
  • The IPO valued Firmus at roughly US$30 billion, nearly 600 times its US$51 million annual revenue, despite only 46 of 912 megawatts of data-centre capacity being built, exposing the AI infrastructure valuation bubble.
  • The collapse tests how regulators police AI infrastructure deals, with ASIC examining disclosure adequacy, representation accuracy, and bookbuilding conduct amid concerns that Australia's public markets may lack depth for capital-intensive AI listings.

NextFin News - Australia's corporate watchdog is seeking information from the investment banks behind Firmus Grid Ltd.'s collapsed initial public offering, a move that stops short of a formal investigation but puts the salesmanship behind one of the world's most hyped artificial-intelligence listings under regulatory scrutiny. ASIC chair Sarah Court said in an Oct. 28 letter to Liberal Senator Dave Sharma that the regulator "is making targeted inquiries" of the joint lead managers on the IPO, adding that "these inquiries are ongoing and ASIC will consider the information obtained and whether any further consideration is warranted."

The distinction matters. The Australian Securities and Investments Commission is not yet running a formal probe into the A$7.1 billion float that was withdrawn on Oct. 9 after bookbuilding failed to clear its A$11 a share price. But the fact that the chair's office is engaging directly with the banks - Bank of America Corp., JPMorgan Chase & Co., Morgan Stanley and Morgans Financial Ltd. - signals that Canberra is treating the Firmus failure as more than a routine market misfire. The episode has become the first serious test of how regulators will police the AI infrastructure gold rush, where soaring valuations rest on capacity that mostly exists only on paper.

The Collapse in Five Days

Firmus began the week as the marquee listing on the Australian Securities Exchange, an AI data-centre operator backed by Nvidia Corp. that aimed to raise as much as A$7.9 billion including an over-allotment option. By the time bidding from large investors closed on Oct. 8, the banks were working on a price near A$8.25 a share, roughly a quarter below the marketed A$11, according to a report the company did not confirm. On Oct. 9 the company shelved the float entirely, saying it would instead pursue a Nasdaq listing and private fundraising.

The sequence is important because of what preceded it. Firmus had moved the close of its bookbuild forward from Friday to Thursday, a step bookrunners typically take only when early indications of demand exceed the size of the offer. Some investors who received those indications later found themselves holding nothing but a withdrawal notice. That gap between the marketed narrative and the final outcome is precisely what Sharma flagged in his letter: he asked ASIC to examine "the adequacy of disclosure, the accuracy of representations made to prospective investors, and the conduct of the bookbuilding process."

Sharma, who sits on the parliamentary committee that oversees the regulator, also pointed to two specific flashpoints: bankers' claims that there was indicative demand for the raise at A$11 a share, and the previously undisclosed termination of Firmus' deal with CDC Data Centres. A separate line of questioning concerned whether Maas Group, the construction company that invested A$410 million for a 3.2 per cent stake in Firmus and tied more than A$1 billion of electrical work to the company, kept its own shareholders properly informed.

Why the ASX Cared So Much

The scale of what failed is part of why the fallout has reached the chair's office. At roughly US$5 billion, the Firmus float would have ranked as the fourth-largest public offering globally in 2026, behind only SpaceX, CXMT Corp and Cerebras Systems, according to Dealogic data. Domestically, it would have been the second-largest IPO in the history of the Australian Securities Exchange, behind only Telstra Corp's A$10 billion privatisation in 1997. A listing that size was meant to do more than raise capital: it was supposed to widen an index dominated by the Big Four banks and miners, and to signal that Australia could still host a marquee technology offering.

Instead, the withdrawal lands on a market that has been steadily shrinking. The ASX has lost listed companies for years, particularly in the infrastructure sector, as private capital takes them out. Over the same period, Hong Kong saw 118 companies - mainly in technology - raise around US$50 billion through IPOs, according to exchange data. The contrast matters because it frames what regulators are really weighing: not just whether one bookbuild was mishandled, but whether Australia's public markets can remain credible venues for capital-intensive AI infrastructure at all. When the single biggest listing attempt in a generation collapses, the questions about process are inseparable from the questions about market depth.

What ASIC Can and Cannot Do

The regulator's posture so far is deliberately narrow. At a hearing of the Joint Select Committee on Artificial Intelligence on Oct. 20, ASIC enforcement head Chris Savundra said: "There isn't a formal investigation on foot at this point, but we are undertaking enquiries, in the usual course." ASIC chief executive Scott Gregson added that the agency plays a limited role in new listings but regularly assesses them afterwards, and that its focus is ensuring investors have not been given "misleading or deceptive information" during the process.

"There isn't a formal investigation on foot at this point, but we are undertaking enquiries, in the usual course."

That framing reflects a real constraint. Australian securities law gives ASIC broad information-gathering powers, and the misleading-or-deceptive conduct regime under the Corporations Act and the ASIC Act applies to conduct in relation to financial products such as shares. But building an enforcement case requires evidence that disclosure documents or representations were materially misleading - and here the defence is baked into the process itself. Lawyers have noted that no retail investors were directly harmed by the withdrawal, and prospective institutional investors were told only of "indications" of interest rather than firm commitments. An indication is not an order, and a bookbuild that fails to clear its price range is not, on its own, unlawful.

Even so, the targeted inquiries carry weight because of whom they target. By going to the joint lead managers rather than only to Firmus, ASIC is testing the conduct of the intermediaries who priced, marketed and distributed the deal. If the regulator finds that banks represented demand levels they could not substantiate, or failed to pass material information - such as the CDC termination - through to investors, the exposure shifts from the issuer to the bookrunners. That is a higher-stakes question for Wall Street than for a company that has already walked away from the listing. ASIC's 2026 enforcement priorities have signalled a continued uplift in enforcement action, with recent Federal Court penalties for fundraising disclosure breaches and misleading conduct running into the millions of dollars - a backdrop that makes the banks' responses to the inquiries anything but routine.

The AI Valuation Machine Meets Reality

Firmus is the clearest example yet of a new market archetype: the AI infrastructure story valued on a pipeline rather than a track record. The company reported revenue of US$51 million in its 2026 financial year, yet the IPO priced the equity at roughly US$30 billion - nearly 600 times annual sales. Its development pipeline stood at 912 megawatts, of which only 46 megawatts had been built. The proceeds from the float were meant to fund purchases of graphics processing units for its first data-centre project in Batam, Indonesia, developed with DayOne Data Centers under an eight-year partnership with Nvidia.

That valuation gap is where the cyclical and the structural parts of this story separate. The cycle is straightforward: after two years of AI euphoria, investors are becoming more discriminating about who actually owns the scarce inputs - power, land, grid connections, chips - and who merely owns a narrative about them. Firmus priced its base offer at A$11 a share, implying an equity value of roughly A$43.7 billion. When the books failed to clear that level, the price talk slid to around A$8.25 before the deal was pulled. That is a mean-reverting move: the market is repricing AI infrastructure from "anything with an AI label" toward "assets with contracted power and customers."

The structural shift runs deeper and will not revert on its own. Regulators, auditors and exchanges are now being forced to build new muscles for assessing AI-capex claims, because the traditional disclosure playbook was written for companies with revenue histories, not for companies whose principal asset is a promise to buy chips. ASIC's limited statutory role in listings means the immediate Firmus inquiry may produce little more than information. But the pressure it reflects - from lawmakers, from investors who received indications that did not convert, and from a market that has begun to distinguish between AI haves and have-nots - is durable. Expect more questions, more hearings, and more scrutiny of the gap between marketed demand and booked demand across the AI complex.

The Counter-Thesis: This Is Just a Failed Bookbuild

The strongest case against reading too much into the ASIC move is the simplest one: bookbuilds fail all the time, and a failed bookbuild is not evidence of misconduct. Under this view, Firmus simply priced too aggressively for a market that was cooling, the banks adjusted the price when real orders arrived, and the company chose to withdraw rather than list at a discount. No retail investor lost money. The "indications" that were circulated were, by definition, non-binding. On this reading, the targeted inquiries are routine housekeeping - the regulator confirming that process was followed - and the appropriate conclusion is that nothing happened beyond an ordinary market repricing.

There is force in that argument, and it is why a formal investigation has not opened. But it underestimates two things. First, the speed of the reversal - from an early close on claimed excess demand to a withdrawn deal within days - is unusual even for aggressive tech floats, and it is the pattern, not the outcome, that draws regulatory attention. Second, the AI infrastructure market is large enough and systemically important enough that a pattern of over-marketed deals would matter even if each individual case is defensible. If banks can repeatedly claim demand that does not materialise, the price-discovery function of the IPO market degrades for everyone.

The sceptics were not hard to find before the collapse. Jun Ben Liu, portfolio manager at Ten Cap, told the Australian Broadcasting Corporation that she was steering clear of the Firmus IPO because of the "unprecedented lack of detail" and enormous structural risks, noting that 97 per cent of the data-centre capacity Firmus had promised had yet to be built. "They might deliver cheaper earnings in two years if they build what they promise, but that's speculative," she said. That is the crux of the structural argument: the market was asked to underwrite a balance sheet whose main assets were contracts to acquire assets that did not yet exist.

The falsifying signal is concrete: if ASIC's inquiries close without findings and the major banks face no consequences, while AI infrastructure IPOs resume pricing at 500-to-600 times revenue with pipelines that remain largely unbuilt, then this was a one-off repricing and the regulatory episode is a footnote. If, instead, future AI listings carry more conservative pricing, more explicit risk disclosure around unbuilt capacity, and regulators in other jurisdictions open their own reviews of bookbuild conduct, then Firmus marks the end of the AI IPO free-for-all.

What Comes Next

In the short term, the inquiry itself is the story. ASIC will collect information from the four joint lead managers, review it, and decide whether further action is warranted. The banks will respond through counsel, and the process is likely to be measured in months rather than weeks. Firmus, meanwhile, is pursuing a private fundraising - reports have put the target at between US$2 billion and US$3 billion - and a Nasdaq listing, a path that would place it under US disclosure rules and Securities and Exchange Commission oversight rather than ASIC's.

Over the medium term, the outcome will shape how AI infrastructure companies access public markets. A base case is that ASIC closes the inquiries without enforcement but that the episode leaves a lasting mark on pricing discipline: bookrunners will be more cautious about circulating demand indications they cannot back, and issuers will price closer to what the books will actually clear. An upside case for the banks is that the inquiries confirm clean conduct, restoring confidence in the IPO process and allowing AI listings to resume with tighter disclosure. A downside case is that ASIC finds evidence of misleading representations, opening the door to enforcement action and a broader chilling effect on AI-related listings in Australia.

For investors, the lesson is narrower than the headlines suggest. The regulatory question is about process and disclosure, not about whether AI data centres are a sound long-term investment. The market question - which AI infrastructure companies own real assets versus real narratives - is separate, and it is already being answered in the price.

Firmus did not fail because regulators arrived; it failed because the price asked investors to pay for a pipeline before the pipes were laid. The inquiry is the bill for the sales pitch, not the cause of the collapse.

Explore more exclusive insights at nextfin.ai.

Search, ask or analyze anything...