Firmus Wanted Three Times Its August Valuation And The Market Said No

In this storyFirmus

Firmus, the Nvidia-backed Australian AI data centre operator, has withdrawn a planned initial public offering of about $5 billion, citing market volatility and conditions.

The offering, priced at A$11 a share, would have valued the company at roughly $30.6 billion and been the second-largest new share sale in Australian history. Reuters reported that it met lukewarm demand.

Firmus said its board concluded the terms did not adequately reflect the strength of the business or its long-term growth outlook, and that proceeding was not in shareholders' interests. It will pursue private capital and consider alternative public and private options.

The valuation sought represented a sharp step up. In August the company raised $2 billion from Nvidia, Blackstone, Coatue Management and Jane Street at a valuation above $10.5 billion, taking total equity raised over the preceding year past $3 billion.

Firmus designs and operates modular AI data centres using proprietary energy and cooling technology. Last month it announced agreements with Meta to provide GPU computing capacity at facilities in South-East Asia, built on Nvidia's platform, to support Meta's AI research, model development and training.

The withdrawal came in the same week that OpenAI told investors its annualised revenue was about $50 billion rather than the $68 billion widely reported, a disclosure that sent Nvidia, Oracle and CoreWeave shares lower on 8 October.

Two months and roughly three times is the whole of this story.

In August, a group including Nvidia, Blackstone, Coatue and Jane Street put $2 billion into Firmus at a valuation above $10.5 billion. In October the company asked public investors for about $30.6 billion. The business did not change materially in between; it signed a Meta agreement, which is meaningful but not a tripling.

What was being tested was whether public markets would accept private AI valuations and then extend them. They would not. Firmus's own statement, that the terms did not reflect the strength of the business, is the standard formulation and it describes a disagreement about price rather than a disagreement about the company.

The timing makes it harder to read as a one-off. The book was being built in the same week that OpenAI's revenue was restated downward by roughly a third on a definitional point, and that Nvidia, Oracle, AMD, Broadcom and CoreWeave all fell. An AI infrastructure issuer asking for a tripled valuation into that week was asking at the worst possible moment.

Which leaves the genuinely open question. A listing that fails on timing can be brought back in a calmer quarter at a similar price. A listing that fails because the valuation was never supportable comes back lower or not at all, and from outside it is not yet possible to tell which this was.

The private round will answer it.

The private-market fallback is the part that matters, and it is not obviously available on the same terms.

Firmus says it will raise privately instead, which is what a company says when a listing fails. But the investors who marked it at $10.5 billion in August are the ones being asked to fund the next round, and they now have a public data point telling them what the open market thought of three times that figure. A private round after a withdrawn IPO usually clears below the sought price, and sometimes below the previous one.

That matters beyond Firmus because the same structure sits under much of the AI infrastructure build. Private valuations in this category have been set largely by strategic and crossover investors rather than by public markets, and until an offering is tested, there is no independent check on them. Firmus is the first substantial test in this cycle, and it did not clear.

For Indian readers the relevant exposure is the capacity being built here on comparable assumptions. Nxtra raised $1 billion and Neysa $600 million this year, and SB Energy filed in the United States while disclosing substantial dependence on OpenAI. None of those are Firmus, and India's data centre demand has a domestic component Australia's does not.

The common factor is that all of them are financed against expected AI demand rather than contracted revenue, and the cost of that financing has just gone up. Micron's memory shortage makes the same capacity more expensive to build at the same time.

For daily, sharp analysis of the biggest moves in the Indian business and startup ecosystem, follow StartupFox.