In the compressed arc of a single year, Firmus Technologies rose from a $1.85 billion startup to a near-$44 billion IPO candidate — and then, weeks before its scheduled ASX debut, the edifice began to give way beneath the weight of its own ambition. The Australian AI datacentre company, backed by Nvidia and Blackstone, now faces the humbling arithmetic of investor skepticism: a loss-making business whose contracted revenue rests almost entirely on facilities not yet built cannot indefinitely sustain a valuation built on euphoria. What was to be Australia's largest public offering in a generati
Firmus's $44bn valuation crumbles weeks before ASX debut
The whole thing was getting fanciful
So Firmus went from being worth less than $2 billion to $44 billion in about a year. How does that even happen?
It's a combination of genuine excitement about AI infrastructure and successive rounds of funding from the same core investors—Nvidia, Blackstone, others—each round pushing the valuation higher. The company has real contracts with Meta, OpenAI, Nvidia. But the valuation kept accelerating faster than the business could actually deliver.
But we should be careful here. We know the valuation went up. We know investors are now skeptical. But do we actually know why the valuation went up so fast? Was it pure hype, or were there genuine business developments that justified each jump?
That's the thing—the source material doesn't really explain the reasoning behind each valuation increase. We know it happened, but not the mechanics of why.
What about the actual risk? Is Firmus in trouble, or is this just a repricing?
The real risk is that 97 percent of their contracted revenue depends on facilities that don't exist yet. They have two small operational sites. Everything else is planned or under construction.
That 97 percent figure comes from one analyst, Armina Rosenberg. It's a specific claim, but it's one person's assessment. We don't have Firmus's own breakdown of their revenue pipeline.
And if those facilities don't get built on time, or if AI spending slows down, the whole valuation collapses.
Exactly. The company is betting that construction stays on schedule, financing stays available, and hyperscaler spending doesn't moderate. If any of those assumptions breaks, the math falls apart.
We should note that Firmus was asked for comment but hasn't responded in this reporting. So we're hearing one side of the story—the skeptical side.
What happens now?
Either they slash the share price dramatically and go ahead with the October 23 listing, or they pull the offering entirely. Either way, it's a major reversal from where they were a few weeks ago.
Le Pouls
- A valuation that tripled in the final weeks before listing — from $15 billion to nearly $44 billion — has collapsed under the weight of investor resistance, with insiders describing the figure as 'fanciful' for a company losing hundreds of millions of dollars.
- Firmus quietly withdrew from a parliamentary AI inquiry mid-negotiations, signalling the depth of the crisis as its team scrambled to find a price the market would actually bear.
- The structural fragility is stark: 97% of contracted revenue depends on datacentres that do not yet exist, making the entire investment thesis contingent on a chain of construction, financing, and renewal assumptions all holding simultaneously.
- Analysts have mapped the company's trajectory onto the classical arc of a market bubble — displacement, boom, euphoria — warning that the underlying business may have real value, but that investors were being asked to pay an unsustainable premium for it.
- With the October 23 listing date now in serious doubt, Firmus must choose between a dramatic repricing that salvages the float or a full withdrawal back into private markets — either outcome a sharp reversal from the confidence of weeks prior.
In the compressed arc of a single year, Firmus Technologies rose from a $1.85 billion startup to a near-$44 billion IPO candidate — and then, weeks before its scheduled ASX debut, the edifice began to give way beneath the weight of its own ambition. The Australian AI datacentre company, backed by Nvidia and Blackstone, now faces the humbling arithmetic of investor skepticism: a loss-making business whose contracted revenue rests almost entirely on facilities not yet built cannot indefinitely sustain a valuation built on euphoria. What was to be Australia's largest public offering in a generation has become, instead, a parable about the distance between a compelling story and a defensible price.
Firmus Technologies was supposed to mark a turning point for Australian capital markets — the country's largest IPO since Telstra went public nearly three decades ago. The AI datacentre company had secured a late-October ASX listing at $11 per share, with Nvidia, Blackstone, Jane Street, and Coatue among its backers. Then, weeks before the scheduled debut, the structure began to crack.
Multiple sources told Guardian Australia that Firmus is now cutting its valuation in search of a price investors will accept — or may abandon the offering entirely. The company's withdrawal from a parliamentary AI inquiry on Thursday came amid what insiders described as frantic negotiations to salvage the float. Just over a year ago, Firmus was valued at $1.85 billion. Eight weeks before the planned listing, that figure had reached $15 billion. Days before the withdrawal, it had nearly tripled again to almost $44 billion. Now that valuation is being 'heavily unwound.' One investment manager put it plainly: 'The whole thing was getting fanciful.'
The core tension is not whether AI will continue driving massive infrastructure spending — it almost certainly will. The question is whether Firmus, still in its startup phase, is worth anywhere near what investors were being asked to pay. The company operates two small facilities, with seven contracted and four planned datacentres in the pipeline. But roughly 97% of its contracted revenue depends on sites not yet built. As Minotaur Capital's Armina Rosenberg noted, reaching the offer price requires delivery, financing, and renewals all going to plan — a great many moving parts.
Firmus builds liquid-cooled 'AI factories' packed with Nvidia GPUs, with Meta, OpenAI, and Nvidia among its contracted customers. Its expansion focus on Asia rather than Australia or the US sidesteps local community resistance to new datacentres, but introduces its own risks: power constraints, long construction timelines, and unfamiliar regulatory environments. Morningstar analyst Lochlan Halloway, drawing on economist Charles Kindleberger's bubble framework, concluded that sentiment around Firmus had entered the euphoric phase — not a verdict on the company's worth, but a warning that investors were being asked to pay a price resting on assumptions that may not hold.
Firmus now faces a binary choice: reprice dramatically and proceed, or withdraw and return to private fundraising. Either path marks a sobering reversal from the confidence that surrounded the deal just weeks ago.
Firmus Technologies was supposed to be Australia's biggest initial public offering since Telstra went public nearly three decades ago. The AI datacentre company had lined up a late-October listing on the ASX, priced at $11 per share, and was backed by some of the world's most powerful investors: Nvidia, Blackstone, Jane Street, and Coatue. Then, just weeks before the scheduled debut, the entire structure began to crack.
Multiple sources told Guardian Australia that Firmus is now frantically cutting its valuation to find a price investors will actually accept—or may abandon the offering entirely. The company withdrew from a scheduled parliamentary inquiry into artificial intelligence on Thursday amid what insiders described as harried negotiations to salvage the float. The momentum that had carried Firmus to a near-$44 billion valuation had evaporated almost as quickly as it had built up.
The scale of the reversal is staggering. Just over a year ago, when Nvidia and other early backers took equity stakes, Firmus was valued at $1.85 billion. Eight weeks before the planned listing, that figure had climbed to $15 billion. Days before the withdrawal, it had nearly tripled again to almost $44 billion. Now that valuation is being "heavily unwound," in the language of those tracking the situation. One investment manager briefed on the float captured the sentiment plainly: "The whole thing was getting fanciful. It's a business that's losing hundreds of millions of dollars, and yet its valuation kept going up nonstop every couple of months."
The core problem is not whether artificial intelligence will continue to drive massive spending—it almost certainly will. The problem is whether Firmus, a company still in its startup phase, is worth anywhere near what investors were being asked to pay. Firmus operates two small facilities and has seven contracted and four planned datacentres in the pipeline. But here is the catch: roughly 97 percent of its contracted revenue depends on sites that have not yet been built. Armina Rosenberg, co-founder of Minotaur Capital, laid out the dependency plainly: "You only get near the offer price if delivery, financing and renewals all go to plan." That is a lot of moving parts that need to align perfectly.
The company's appeal is straightforward enough. It is building and operating liquid-cooled "AI factories" packed with Nvidia graphics processing units—the chips that power modern artificial intelligence systems. Meta, OpenAI, and Nvidia have all signed contracts to use Firmus's facilities. But Firmus has chosen to focus its expansion in Asia rather than Australia or the United States, where local communities are increasingly pushing back against new datacentre developments. That geographic choice comes with its own obstacles: power constraints, long construction timelines, and the inherent risks of building in unfamiliar regulatory environments.
There is also a deeper vulnerability. The current stratospheric spending on AI infrastructure by the so-called hyperscalers—Microsoft, Google, Meta—cannot continue at this pace indefinitely. When that spending eventually moderates, third-party datacentres like Firmus could be among the first to feel the impact. Morningstar analyst Lochlan Halloway drew on economist Charles P. Kindleberger's framework for understanding market bubbles—displacement, boom, euphoria, distress, and revulsion—and concluded that sentiment around Firmus had entered the euphoric phase. That does not mean the company is worthless or that its underlying business is hollow, Halloway cautioned. Rather, the risk is that investors were being asked to pay an unsustainable price for an asset whose value depends on a chain of assumptions that may not hold.
As of now, Firmus faces a binary choice. The company can radically reduce the share price—and with it, the overall valuation—and proceed with the listing. Or it can withdraw the offering altogether and return to private fundraising. Either path represents a dramatic reversal from the confidence that surrounded the deal just weeks earlier. The October 23 listing date that was supposed to mark a milestone for Australian capital markets is now in serious doubt.
Citations marquantes
It's a business that's losing hundreds of millions of dollars, and yet its valuation kept going up nonstop every couple of months.— Investment manager briefed on the float
You only get near the offer price if delivery, financing and renewals all go to plan.— Armina Rosenberg, Minotaur Capital co-founder