Nvidia-backed data centre operator Firmus cancelled its $5 billion initial public offering on the Australian Securities Exchange after failing to draw enough demand at the valuation it wanted. The company confirmed the decision Friday following two days of discussions with bankers. The float, which had targeted roughly $5.5 billion in proceeds, would have been the largest new listing on the ASX in nearly three decades and one of the biggest tech IPOs anywhere in the world this year.
The withdrawal lands on a market already on edge about AI-linked valuations. Firmus had cut its offering price by about 25% during the book-build, stepping back from a valuation around $31 billion, and still could not get the book filled. Reuters first reported the shelving on October 8, citing people familiar with the process, and the Australian Financial Review confirmed the final decision Friday morning Sydney time.
Investors who looked at the deal flagged a specific problem: the company operates only two data centres and has no completed track record of building AI-grade facilities at scale. At the asking valuation, they were effectively buying a development pipeline, not an operating business. Several large Australian institutional investors told bankers they found the pricing aggressive even after the discount, according to people briefed on the process.
What the float would have been
At $5 billion, Firmus would have ranked as the fourth-largest IPO globally so far this year, behind SpaceX, CXMT Corp and Cerebras Systems, according to Dealogic data. It would also have been the second-largest IPO in ASX history, behind only Telstra’s roughly $10 billion float in 1997, a deal that continues to serve as the benchmark Australian issuers measure themselves against.
| Deal | Size | Status |
|---|---|---|
| Firmus IPO | ~$5.5 billion targeted | Cancelled Oct. 9 |
| Telstra 1997 | ~$10 billion | ASX’s largest ever |
| SpaceX (2026) | Largest global YTD | Completed |
| Cerebras (2026) | Third-largest YTD | Completed |
| ASX ranking | Second-largest | Would have been |
The IPO had been closely watched partly because of who was behind it. Nvidia, the chipmaker whose GPUs underpin most large AI data centre buildouts, backed Firmus, and the company pitched itself as a specialist in high-density AI compute facilities with liquid cooling, positioned for the wave of demand coming from hyperscalers and sovereign AI projects across the region.
Market reaction
Maas Group, the construction and materials company that owns 3.2% of Firmus, saw its shares drop about 30% on October 8 as reports spread that the float might be cut or pulled. Maas had been expected to crystallise value from the listing, and the collapse erased most of that expectation in a single session.
“At $5 billion, the closely-watched IPO would have ranked as the fourth-largest public offering globally so far this year, behind SpaceX, CXMT Corp and Cerebras Systems, according to Dealogic data.” – Reuters, October 9
Brokers involved in the deal are now trading blame over who pushed the valuation too high and who misread the demand book, according to the Australian Financial Review, which described a “blame game” with bankers at the centre. Some investors argued the offering was brought to market too quickly relative to the company’s operating track record. Others argue the AI financing window itself has narrowed, pointing to a string of AI-linked deals and financings worldwide that have repriced lower in recent weeks, from chip stocks pulling back on disappointing revenue reports to private rounds renegotiating terms.
Context: a tougher window for AI deals
Firmus’s failure is the most visible Asia-Pacific example of a broader shift. Global markets have spent the past week repricing AI-linked assets, with chip stocks falling after an OpenAI revenue report came in below the $70 billion annualized figure many traders had assumed, and industrial companies tied to data centre construction pulling back on expansion plans. Public floats depend on a level of price agreement that private rounds do not require, and the Firmus book failed to find that agreement despite a brand-name backer and a genuinely scarce asset class. The comparison investors kept making on the syndicate calls was between an operating company with proven cash flow and a development-stage operator asking for a similar valuation multiple.
The valuations embedded in AI-linked IPOs this year have, in several cases, assumed multi-year revenue ramp-ups the companies have not yet achieved. That works when the buyer is a single sovereign fund with a long horizon. It is harder when the offering needs broad institutional participation to fill a $5.5 billion book, and when every large institution has its own AI exposure committee reviewing what it already holds.
For the ASX, the loss is more than a single deal. Australia’s equity market has struggled for years with listings that fail to reach the exchange, and the local pool of public companies has shrunk as take-privates and cross-listings pulled names offshore. A float of this size was expected to bring a flow of passive and index money that ripples into other tech names, and bankers in Sydney now face the task of explaining to prospective issuers that the reception is thinner than their pitch decks assumed. Several had queued mid-stage AI and resources technology deals for a post-Firmus window, and at least a few are now rethinking their own timelines, according to people on the sell side.
For Nvidia’s ecosystem, the episode is a caution more than a verdict. The chipmaker’s backing did not solve the pricing question, and the deal will likely return with a lower valuation, a longer operating track record, or both before it comes back to market. Data centre demand itself is real and still growing, with hyperscaler capex guidance continuing to rise. Whatrepriced here is not the demand thesis. It is the price at which a minority of that thesis is worth owning in public markets, and for now, at least one investment committee in Sydney has made that judgment explicit.