Why Did Investors Walk Away from Firmus?

You may have seen the headlines: Firmus, the Nvidia-backed AI data centre developer, has pulled what would have been one of the largest floats in ASX history. The company had sought to raise around $7 billion at a market value of roughly $44 billion, a listing expected to be the biggest since Telstra in 1997. 

On the morning of Friday 9 October, just days after its books opened, Firmus withdrew its application to list. It is a striking turn for one of the most talked-about companies in Australian tech, and it offers some timely reminders for anyone weighing up a new listing or simply trying to make sense of the AI investment story. 

What happened 

Founded in Tasmania in 2019, Firmus has rapidly grown into a builder of “AI factories”: purpose-built data centres that house Nvidia chips and rent that computing power to customers. OpenAI and Meta have both signed on, and Nvidia itself is a major shareholder. The company has plans for five data centre sites across Australia, Malaysia and Indonesia, with a total build cost of around $51 billion. 

The Initial Public Offering (IPO) was marketed at $11 per share. Within days, demand from local super funds and fund managers proved too thin to support that price. Midweek, CDC Data Centres confirmed that its planned partnership with Firmus had been abandoned, and reports emerged that the price would need to be cut. Rather than accept lower terms, the board withdrew the offer, citing market volatility and saying a reduced price would not reflect the business’s long-term prospects. Firmus will now seek capital from private markets. 

Why investors walked away 

The concerns raised by fund managers were less about AI itself and more about the specifics of this offer: 

• Valuation had moved very fast. Firmus was valued at $1.85 billion in September 2025. Barely a year later, it was asking public investors to pay roughly 24 times that.
• Revenue was still largely in the future. Contracted revenue runs into the billions, but is only collected once facilities are operating. Revenue last year was around $50 million.
• Very little was built. Firmus had around 46 MW of capacity operating, against 865 MW under development. Analysts noted that only about 5% of the capacity it had sold was built and running.
• Early backers could sell immediately. Escrow, which normally stops major holders selling straight after listing, applied to only about 42% of shares.
• Late surprises. The CDC exit and a new Meta agreement arrived mid-bookbuild, shortly before investors had to commit. 

Lessons for investors 

A great theme is not the same as a great price. Few would dispute that demand for AI computing is real. But owning a business in a growing industry only rewards you if the price you pay leaves room for things to go wrong. The investors who stepped back weren’t rejecting AI; they were rejecting the terms. 

Hype is not demand. Early reports said the book was well covered. Within days, that confidence had evaporated. Headline enthusiasm, big-name backers and media coverage can create a sense that missing out is the bigger risk. Disciplined investors judge the offer on its merits, not the noise around it. 

Read the fine print on who is selling. Escrow arrangements matter. When a large share of the register can sell from day one, early backers who bought at a fraction of the price may be the first to take profits, and new investors can be left holding the stock. 

Contracts are promises, not profits. Contracted revenue is encouraging, but it depends on facilities being built on time and on budget, financed at reasonable cost, and customers continuing to need the capacity. A plan that requires tens of billions of dollars in future funding carries execution and financing risk that a prospectus cannot remove. 

Where this leaves investors 

Firmus may yet prove its doubters wrong. It has world-class partners and customers, and it may well return to market once more of its capacity is built. But a withdrawn float is a useful reminder that markets still do their job: when the price runs ahead of the evidence, investors can and do say no. 

If you’d like to understand your portfolio’s exposure to AI and technology, or how we approach new listings, please don’t hesitate to contact us. You may also enjoy Alexi’s earlier piece, Late-Stage Artificial Problem?