The listing was meant to test how much Australian investors would pay for the country’s AI-infrastructure boom. It ended before it really began. Now the company at the center of it is raising money the quieter way.
Firmus, an Australian data-center operator backed by Nvidia, is in talks with existing investors to raise between $2 billion and $3 billion in a private round, according to people familiar with the matter. The move comes after the company shelved an initial public offering that was expected to rank among Australia’s largest listings.
The retreat is a familiar one in the business of building data centers for artificial intelligence. Companies like Firmus, which the industry calls neoclouds, need money continuously to keep their construction pipelines moving. When public markets balk, they turn back to the investors who already know them.
That is exactly what Firmus is doing. Its existing backers had planned to subscribe to roughly half of the money the IPO was meant to raise, which could help the private round come together more quickly, the people said. The financing structure has not been settled and could combine equity and debt. Whether Nvidia itself will take part is unclear, they said.
Firmus’s story tracks the arc of the AI build-out. Founded in 2019 by Oliver Curtis, the company began in cryptocurrency and high-performance computing before repositioning as a builder of “AI factories” — the data centers that house the chips training and running AI models. Its backers include Nvidia, Blackstone, Coatue Management and Jane Street.
The company’s main effort, called Project Southgate, aims to scale to 1.6 gigawatts of capacity through 2028 across several Australian sites, with Nvidia supplying the DSX platform that its AI factories run on. In April the company raised $505 million in a round led by Coatue, and it has previously secured about $2 billion in investor commitments.
The IPO was meant to be the next step, and it became a cautionary tale. Reports described insufficient demand for the marketed share price, with investors uneasy about how much of the promised capacity had actually been built. By one account, most of it remained unbuilt — a gap between the sales pitch and the balance sheet that made the valuation hard to defend.
The company closed the books on the listing early and later withdrew its application to list on the Australian Securities Exchange, citing market conditions. It said it would look at other ways to fund its plans.
The math behind the retreat is the same math that now governs the whole sector. Building a data center is capital-intensive long before it is revenue-generating, and the AI build-out has pushed that intensity to an extreme. Neoclouds, which lack the balance sheets of the hyperscalers, must keep raising — through equity, debt, or whatever the market will accept.
That dependence is the sector’s exposed nerve. When appetite for AI risk cools, the companies furthest from revenue feel it first. Firmus’s shelved IPO, and its turn back to private money, is one early data point in that cooling.
Analysts have begun to ask the same question across the industry: how much of the promised capacity is real, and how much is still a rendering. The answer matters because the money is already largely spent or committed, and a project that stalls partway is harder to finance than one that never started.
Australia has become one of the more active fronts in this build-out. Nvidia has struck partnerships with a cluster of local data-center firms, betting that the country’s land and power can anchor AI compute for the Asia-Pacific region. Firmus is one of the most prominent names in that group, which is part of why its listing carried so much weight.
For Firmus, the private round is a chance to buy time. Raising from investors who already hold a stake is usually faster than pitching a skeptical public market, and it keeps ownership in fewer hands while the company tries to convert its construction pipeline into operating revenue.
The deal is not done. The final size and terms could change, and the mix between equity and debt is still being worked out, the people familiar with the talks said. But the direction is clear: the public markets were not ready for Firmus, so Firmus is going back to the investors who already are.
A completed raise of that size would keep Firmus on schedule for the build-out its business depends on, buying time to convert construction into revenue and, in time, to try the public markets again.
The episode will be read two ways. To believers, it is a hiccup on the way to a larger listing once the market warms. To skeptics, it is a sign that the AI-infrastructure boom has begun to price risk more honestly — and that the companies closest to the construction are the first to feel it.
