Firmus IPO Pulled: What It Means for AI Compute Buyers

The Firmus IPO is off. On October 9, 2026, the Nvidia-backed Australian AI data centre operator withdrew its application to list on the ASX, Reuters reported, just three days after it began taking orders from big investors.

The float would have raised about US$5 billion and valued Firmus at around US$30.6 billion. Instead, orders faded as investors questioned the price, the debt and a key partner's exit. This article covers what happened, why it matters for the AI trade, and what teams that rent GPU capacity should check before they sign.

Key Takeaways

• Firmus pulled its roughly US$5 billion ASX float on October 9, 2026, after orders that bankers said covered the deal melted away in about three days.
• The A$11 offer price implied an equity value of about US$30.6 billion, nearly triple the US$10.5 billion valuation of its August funding round.
• Investors balked at about US$30 billion of debt, a short building record, loose escrow terms and news that CDC Data Centres had stepped back from a 1.6 gigawatt plan.
• Firmus now plans to raise money privately, and a source told Reuters a Nasdaq listing would follow.
• For teams that rent GPUs, the lesson is simple: price a provider on capacity that's running today, not capacity on a slide.

What Happened to the Firmus IPO?

The deal fell apart in a single week. Here's the sequence as Reuters, the ABC and Forbes Australia reported it:

• Tuesday, October 6: bankers opened the book at A$11 a share. A term sheet said indicative offers already covered the deal.
• Wednesday, October 7: investors began pulling orders after CDC Data Centres CEO Greg Boorer told a podcast that a plan to build 1.6 gigawatts of AI factories with Firmus was no longer underway.
• Thursday, October 8: the banks stopped taking bids, and reports said the price could be cut to between A$8 and A$9.
• Friday, October 9: Firmus withdrew its listing application that morning.

The listing had been set for October 23. Firmus considered a lower price, one person involved told Reuters, but chose a private round instead.

How Big Was the Firmus IPO Meant to Be?

Very big for Australia. Reuters said it would have been the second-largest new share sale in the country's history and the fourth-largest IPO in the world so far this year, behind SpaceX, CXMT and Cerebras Systems, citing Dealogic data.

Reported numbers are mixed in Australian and US dollars, so here they are side by side:

Measure	Reported figure	Source
Offer size	About US$5 billion, roughly A$7 billion	Reuters
Offer price	A$11 a share	Reuters
Equity value at offer	About US$30.6 billion, around A$43.7 billion	Reuters, InDaily
August 2026 round	US$10.5 billion	Reuters
Debt	About US$30 billion	Analysts for the lead managers, via Reuters
Enterprise value	About US$60 billion	Reuters
Earnings target	US$5 billion a year within five years	Draft prospectus, via Reuters

The full prospectus was never released. A draft went to big investors, which is why most of these figures come from people who saw it rather than from a public filing.

Why Did Investors Walk Away From the Firmus IPO?

Four worries stacked up in the same few days. Each one alone might have been fine, but together they cut demand fast.

• Price: at A$11, Firmus would have been worth about three times what investors paid in August. By our math, that's a jump from US$10.5 billion to US$30.6 billion in roughly two months, before most of its data centres exist.
• Debt: analysts for the banks put it at about US$30 billion. Against the US$5 billion earnings target, that's six times earnings the company hasn't made yet, by our math.
• Stock supply: on Tuesday, investors were told the escrow terms would let existing holders sell more than half the shares from day one. That kind of overhang can sink a new stock's early trading.
• CDC: a major partner said a 1.6 gigawatt build plan was no longer underway, which we cover next.

Blackwattle portfolio manager Joseph Koh, whose firm looked at the deal but didn't bid, told Reuters the price assumed "near flawless execution." The market, he said, wasn't ready to take that leap yet.

What Was the CDC Data Centres Partnership?

CDC Data Centres is one of Australia's biggest operators. Under a plan called Project Southgate, it and Firmus aimed to build 1.6 gigawatts of AI computing capacity across several Australian cities by 2028, a deal Reuters valued at A$73 billion.

Only a small part got built. Data Center Dynamics reported that the partnership delivered 42 megawatts in Melbourne and that the site won't grow to the planned 150 megawatts.

By our math, 42 megawatts is under 3% of the 1.6 gigawatt goal. Boorer said CDC had chosen to focus on its own developments, according to business outlet Rampart. Firmus said the decision doesn't affect its plans, contracted customer capacity or strategy.

How Much of Firmus Is Built Today?

Less than the valuation implied. Reuters says Firmus has two leased data centres online, in Melbourne and Singapore, and plans five more across Asia-Pacific.

Forbes Australia put operating capacity at about 46 megawatts, with 865 megawatts under development. Firmus said in September it had passed 900 megawatts of contracted capacity, according to reports at the time.

The customer list is strong on paper, according to Reuters and other reports:

• OpenAI signed in September as the anchor customer for two planned sites in Malaysia.
• Meta has signed capacity in the region, including at one planned Batam site, reports say.
• Nvidia is a partner on a 360 megawatt campus on Indonesia's Batam island, built with DayOne.

But contracts pay when buildings are finished. Forbes Australia reported revenue of about $50 million last year, with the "billions" in contracted revenue collected only once sites go live.

Who Backs Firmus?

The cap table is a who's who of AI money. Reuters lists Nvidia, Coatue Management, Blackstone and Jane Street among its backers, and Blackstone and Coatue led a US$10 billion debt facility earlier this year.

Firmus was founded in 2019 by Oliver Curtis and Tim Rosenfield. It began as a crypto miner in Tasmania before turning to AI, according to Forbes Australia.

Wilson Asset Management has held shares since last year. Its lead portfolio manager, Oscar Oberg, told Reuters the result was "disappointing given that three days ago we thought it was all going well."

What Does the Firmus IPO Say About the AI Trade?

It says buyers are getting choosier, not that the boom is over. Ten Cap co-founder Jun Bei Liu told Reuters it was "an important reality check," but not "the beginning of the end of the AI trade."

Other signals point the same way. US-listed chip stocks, up more than 80% this year, fell 3.4% on October 8 after media reports that OpenAI's September revenue run rate was almost US$50 billion, below what it had signalled earlier, Reuters said.

Yet money is still flowing. Reuters notes that Nvidia and SpaceX are raising tens of billions in debt, and that Anthropic is seeking to raise as much as US$100 billion in an IPO. The message is not "no AI," it's "show the math."

Morningstar's Lochlan Halloway summed up the model to the ABC: neoclouds borrow against customer contracts to buy chips, then use the rent to repay the loans. That works when sites open on time and customers pay. It strains when either slips.

How Does the Firmus IPO Compare With CoreWeave's?

CoreWeave is the closest match, and its float is a useful guide. When the US neocloud listed in March 2025, it also met cautious demand, but it chose a different fix.

CoreWeave cut its deal instead of pulling it. It priced at US$40 a share, below its planned range, and raised about US$1.5 billion, less than it first aimed for. The stock then rose sharply over the following months as AI demand held up.

Factor	CoreWeave, March 2025	Firmus, October 2026
Response to weak demand	Cut the price and size, then listed	Pulled the float, chose a private round
Running capacity at listing	Dozens of US and European sites	Two leased sites, about 46 megawatts
Big-name customers	Microsoft and OpenAI	OpenAI and Meta, mostly for sites still being built
Main investor worry	Customer concentration and debt	Valuation, debt, delivery and escrow terms

The difference that stands out is how much was already running. CoreWeave had revenue flowing from live sites, while most of what investors were asked to price in the Firmus IPO was still being built.

What Is a Neocloud, and Why Does It Matter?

A neocloud is a cloud company built mainly to rent GPUs for AI. CoreWeave is the best-known example, and Firmus is one of a newer wave building in Asia-Pacific.

They matter because they fill a gap. When AWS, Google Cloud and Azure run short of the newest chips, neoclouds often have them sooner, sometimes at lower prices.

The trade-off is risk. Compared with the big three clouds, neoclouds tend to have:

• More debt, often borrowed against customer contracts.
• Fewer sites, so one late build can hold up many customers.
• Fewer large customers, so one loss can strain the whole platform.

What Should AI Compute Buyers Ask After the Firmus IPO?

Most teams don't buy shares in data centre firms. They rent capacity from them, and the questions investors asked of Firmus this week are the same ones a buyer should ask. Here's a checklist for picking a GPU provider:

Question	Why it matters	Good answer
Which site will run my workload, and is it live today?	Planned capacity can slip or vanish	A named, running site, or a firm date with credits if it's late
How is the build financed?	Heavy debt raises the odds of delays and price changes	Funding already in place for the capacity you're buying
What happens if delivery is late?	Your launch plan depends on it	Service credits, a right to exit and no early payment for unbuilt space
How long is the commitment?	Long terms lock you in if things change	Terms that match your own planning horizon, often one year or less
Can I move my models and data out fast?	Exit is your best protection	Standard tools, no lock-in formats and fair data transfer fees
Who are the other large customers?	One big loss can strain the whole platform	A spread of customers, not one or two giants

None of this means you should avoid neoclouds. It means you should buy from them the way you'd buy from any young supplier with a big build plan: with your eyes open and an exit ready.

How Should Teams Spread GPU Risk?

Don't put every workload in one place. A simple split works for most teams, and it costs less than people expect.

• Steady inference: keep it on a hyperscaler or a provider with a long record, where uptime matters most.
• Training and big batch jobs: use a neocloud or spot capacity, where a delay hurts less and the price gap is widest.
• Overflow: keep a second provider set up and tested, even if you only use it a few days a year. Startup credit programs like the Cloudflare startup program can cover a backup cheaply.
• Model APIs: where you can, route through a gateway that can switch between models, so an outage at one lab doesn't stop your product.

The goal isn't to spend more. It's to make sure that if one supplier slips, you lose a few days, not a quarter.

Will the Firmus IPO Change GPU Prices?

Not right away. One pulled float doesn't add or remove chips, and the hyperscalers are still spending heavily on new capacity.

Over time, though, tighter funding could slow some planned builds. If fewer speculative data centres get financed, new supply may arrive more slowly than the pipeline suggests.

What that could mean for your budget:

• Newest chips: prices may stay firm into 2027 if planned supply arrives late.
• Older GPUs: more of them are already built and running, so there's less upward pressure.
• Planning: budget with a price range, not a single number.

What Comes Next for Firmus?

A private round first. Firmus told shareholders it would "pursue capital from private markets and consider alternative international public market options."

A person involved in the deal told Reuters a Nasdaq listing would follow. Firmus declined to confirm that, and Bloomberg reported it was exploring raising up to US$3 billion from existing investors.

Watch three things next:

• The private round: does it price well below the A$11 a share the Firmus IPO asked for?
• Delivery: do the Batam and Malaysia sites open on time for OpenAI and Meta?
• Partners: does a new builder step in for the capacity CDC won't deliver?

What Does This Mean for Australia's Market?

It's a setback. Reuters called the collapse a blow to an exchange that's been dealing with a shrinking number of listed companies and a weak pipeline of new floats.

There may be a policy echo too. The Australian Financial Review said questions should be asked about how fast the orders disappeared, and suggested the corporate regulator take a closer look. That's a call for scrutiny, not a finding of wrongdoing.

Australia still wants AI data centres, and Firmus still has its backers. But the next big float in the sector will likely need:

• A public prospectus before the book opens, not a draft for a few big investors.
• Tighter escrow, so early holders can't flood the market on day one.
• A price that leaves room for new buyers to win.

What We Take From the Firmus IPO

Price capacity on what's built, not what's planned. That was the market's message in the Firmus IPO, and it's good advice for anyone signing a GPU contract this quarter.

The Firmus IPO also shows how quickly confidence can shift in AI infrastructure. Bankers said the book was covered on Tuesday, and by Friday the deal was gone. Your compute plan should be able to absorb that kind of swing.

If you're planning AI workloads and want a second opinion on providers, contracts or a backup plan, that's the work we do. Our breakdown of NVIDIA's Vera Rubin NVL72 covers the hardware side, our guide to cloud cost guardrails covers the budget side, and our multi-cloud FinOps guide covers running across more than one provider.