Firmus Technologies withdraws A$44 billion ASX IPO after investor pushback

Firmus priced its IPO at A$11 a share, implying a valuation near A$44 billion, even though only about 5% of its sold capacity was actually operational versus roughly a quarter for comparable neocloud operators (cloud providers built specifically to rent out AI chip capacity to customers like OpenAI and Meta), and the company was projected to lose A$77 million in the first half of its financial year. University of Adelaide finance lecturer Marta Khomyn put it plainly: at that price, 'investors are betting that everything goes right.' ^[1]

That math only looks worse set against Firmus's own recent history. The company raised private capital at roughly A$1.85 billion in September 2025 - about 13 months, and a roughly 23-fold markup, before it asked public markets for A$44 billion. When institutional demand faltered, underwriters tried to rescue the deal by testing successively lower prices, reportedly cutting toward A$8.25 and then as low as A$5.50, before concluding no price would clear the market and withdrawing the offer altogether. ^[2]

Nvidia's relationship with Firmus is more than a customer deploying its chips: the chipmaker holds a 7.2% equity stake worth roughly US$2 billion in commitments while simultaneously being Firmus's main hardware supplier, a structure the Reserve Bank of Australia has separately flagged as a 'circular AI financing' risk across the sector. ^[3]Morningstar analyst Lochlan Holloway laid out why that matters operationally: Firmus and its neocloud peers are 'heavily geared,' borrowing against customer contracts to buy chips and then using rental income to repay the debt - a model that works smoothly only as long as contracted revenue keeps arriving on schedule. ^[4]

That leverage is exactly why Firmus needed outside capital so badly in the first place. In August 2026, barely two months before the IPO attempt, Nvidia joined Coatue Management, Blackstone, and Jane Street in a US$2 billion private funding round for Firmus, with Maas Group also chipping in A$300 million to lift its stake to 3.2%. ^[5]Betting the IPO on a valuation many multiples above that round is what ultimately made institutional buyers suspicious that the business's financing needs, not its growth story, were driving the timeline.

As recently as October 1, Firmus was still framing its ambitions around a roughly A$30 billion valuation target. ^[6]Within days, reports surfaced that more than half the IPO stock would be available for existing investors to sell with little to no lock-up, an overhang structure that made prospective buyers nervous about being diluted by insider selling almost immediately after listing. ^[7]By October 6, Bloomberg was reporting that this overhang alone was souring institutional appetite even before any news event moved the price. ^[8]

The moment that tipped hesitation into active order-pulling arrived on October 8, when CDC Data Centres chief executive Greg Boorer said on a podcast that a planned 1.6 gigawatt joint AI-factory development with Firmus was 'no longer underway.' That single comment, landing in the middle of the bookbuild, triggered a wave of investors withdrawing orders within hours. ^[9]Bankers scrambled to save the raise by weighing a postponement and a private-round alternative that same day, ^[10]while a separate report framed the cratering demand as a broader warning signal for AI infrastructure funding generally. ^[11]By October 9, the board concluded that proceeding was no longer in shareholders' interests and formally withdrew the listing application, citing 'recent market volatility and prevailing market conditions.' ^[12]

The clearest casualty of Firmus's collapse wasn't Firmus itself - it was Maas Group, the ASX-listed construction and materials company that holds a 3.2% Firmus stake and whose subsidiary JLE Group carries roughly A$1.1 billion in FY26-FY27 work orders tied to Firmus's data center build-out, of which A$373 million has already been paid. ^[13]When the IPO wobbled, Maas entered a trading halt and its shares fell somewhere between 22% and 28%, wiping out roughly A$517 million in market value in a single trading window - a far steeper proportional hit than anything Firmus itself absorbed, since Firmus never had to mark its own valuation down in a public market. ^[14]

That asymmetry is the real lesson for anyone assessing AI infrastructure exposure: the parties most exposed to a neocloud's valuation story are often not the neocloud itself but its contractors, construction partners, and minority shareholders, who have no direct say over pricing decisions yet absorb the market's verdict in full.

Not everyone read the collapse as bad news. Fund manager Jun Bei Liu argued the original ask was simply indefensible - 'almost three times what the valuation was when they raised money just a few months ago,' a reference point distinct from the 23-fold markup over the September 2025 private round cited earlier, since her comparison was to Firmus's own characterization of its most recent raise - and that the market's rejection was the system working as intended. ^[3]Wealth Within analyst Filip Tortevski framed the deeper structural problem: AI infrastructure demands that 'billions must be spent on chips, power and data centres before the expected cash arrives,' and investors are growing unwilling to fund that gap at any price. ^[15]Other reporting suggested the failed float could chill Australia's broader pipeline of AI and neocloud listings, as public investors apply more scrutiny to infrastructure valuations than private markets have so far. ^[11]

The reaction across social media echoed that split verdict. On X, posts mocked Firmus's valuation methodology for leaning on inflated future projections rather than today's numbers, and juxtaposed the company's heavy losses, mounting debt, and modest installed capacity against the A$44 billion ask; VanEck's head of digital assets research went further, calling the outcome 'good riddance' and arguing that the lead underwriters, not the market, were the ones who misjudged the deal. Reddit discussion in Australian finance communities ran in a similar register, dominated by skepticism over the valuation-versus-fundamentals gap and unease about lock-up terms seen as letting early backers cash out with few restrictions, though a minority pushed back that AI infrastructure investment still holds long-term value for Australia regardless of Firmus's own execution stumbles. YouTube coverage of the collapse floated SpaceX's 2026 direct-offering IPO model as a possible alternative path for similarly leveraged, pre-revenue infrastructure plays.

Notably, none of this skepticism has touched Firmus's actual customer relationships: OpenAI's multi-year deal covering two Malaysian sites still pushed Firmus past 900 megawatts of contracted capacity, ^[16]and Meta separately committed to Nvidia GB300 NVL72 capacity at Firmus's Melbourne site and across Southeast Asia. ^[17]The gap between enthusiastic private backers and skeptical public investors is itself the story - neocloud operators have typically priced at 20-30x forward revenue in private rounds versus the 10-15x multiples public markets apply to listed data center operators, ^[18]and Firmus's abortive listing, which would have been Australia's second-largest IPO ever (after Telstra's 1997 float) and the fourth-largest globally in 2026 at its original ask, ^[19]is the clearest evidence yet that those two pricing worlds haven't reconciled.