Firmus’ bankers at Morgan Stanley told clients interested in buying into the Australian data centre start-up in late September that the firm was named for a Latin adjective meaning “strong, stable, steadfast or secure”.
But Firmus’ plans for an almost $44 billion stockmarket listing in Australia turned out to be anything but.
A week that began with the Firmus’ bankers crowing that they had interest from buyers “well in excess of the offer size” spiralled into price cuts by mid-week and ended with the company’s board pulling its planned listing on Friday.
Now the company, led by chief executive Oliver Curtis, will be hoping its backers embody another meaning of “firmus” that Morgan Stanley provided in the research document seen by this masthead: “steadfast, loyal, or unwavering”. But in an industry where momentum and hype are crucial to staying on the right side of the wild swings in share prices that have characterised the AI age to date, that looks far from guaranteed.
Firmus had been presented to the market as an Australian way to benefit from the surging demand for artificial intelligence data centres amid the debates about copyright, noise and jobs. What had once been a bitcoin mining company now planned to roll out so-called AI factories – which differ from normal data centres because they are designed to run or train artificial intelligence tools rather than storing information – in four countries. It inked agreements with AI giants OpenAI, Meta, and Nvidia. “Signing customer contracts of this magnitude, in this time frame, is the ‘slowly, then suddenly’ part of the Firmus story,” wrote Morgan Stanley.
The bankers pushing Firmus’ plans to sell about $US5.5 billion ($7.9 billion) in stock at the public listing used those contracts to forecast extraordinary growth. The company that recorded $50 million in revenue last financial year would deliver more than $US6 billion every six months in 2029, Bank of America claimed. That could make the company worth between $81 billion and $99 billion, the bank concluded. At $44 billion, more than 800 times its last full year of revenue, Firmus was being sold as cheap.
The Australian market was hungry for a listing like Firmus. Macquarie, which was not involved in the float, told clients as late as Thursday that “Firmus’ anticipated late-October listing could drive a rebound” in the moribund market for new share offerings. Several top flight Australian fund managers had a more direct reason for their enthusiasm for the listing: Ellerston Capital, Regal Partners and Wilson Asset Management all hold stakes in the company.
Each was clearly comfortable with the unusual background of its leadership. Curtis, the chief executive, was the son of wealthy mining entrepreneur Nick Curtis and had attended the exclusive Sydney private school Riverview. He had the “the best car in the year to drive to school,” the 2003 graduating yearbook noted. Marriage to influencer and public relations consultant Roxy Jacenko followed, and then a hard left turn: the younger Curtis was sentenced to two years jail. He had been taking insider tips from his best friend, fellow Riverview alumni and housemate John Hartman, and trading on the information to make $1.4 million for the pair.
In her 2016 sentencing decision, NSW Supreme Court Justice Lucy McCallum said it was “troubling that, unlike Mr Hartman, Mr Curtis has not embraced responsibility for his offending”. A year later, Curtis walked out of Cooma Correctional Centre on good behaviour. Two years after that, he founded Firmus alongside Tim Rosenfield, his cousin and a former lingerie entrepreneur, and Jonathan Levee, who had been married to Curtis’ sister Erin. None had a background in technology. All were poised to have their shares, plus those of the elder Curtis, who had been Firmus’ chairman, collectively valued at more than $10 billion, had the IPO gone to plan.
(This year, Curtis said he now “owns” the crime he committed, saying he had been “stupid”.)
One investor, who had listened to Curtis pitch his company and spoke on condition of anonymity to preserve relationships, said he was convincing. Curtis may not have a technology background, but he had a background in finance and was fluent in the numbers.
“He paints a very optimistic picture that is easy for investors to fall in love with,” the investor said.
Firmus talks of its HyperCubes, large metal structures that house its AI chips, as a point of difference from rival firms. But it wasn’t the cubes, made by a division of an industrial air conditioning company on the outskirts of Canberra that Firmus bought last month, that impressed the investor. It was the contracts. Yet as Bank of America and Morgan Stanley made clear to their clients, the deals signed by Meta, OpenAI and Nvidia to support Firmus are not ironclad guarantees.
Meta has agreed to buy the computing power from Firmus’ Australian sites, Morgan Stanley told clients, while OpenAI had made a similar agreement for the company’s South-East Asia facilities. But both of those contracts cover only a relatively small proportion of the computing capacity Firmus is hoping to build. The rest is “an option to acquire additional compute capacity”, Morgan Stanley noted. In other words: the tech giants have first dibs at whatever computing power Firmus generates. They are under no obligation to buy. And most of that computing capacity does not exist yet. Just four per cent of Firmus’ planned revenue comes from the two facilities it currently operates, and one of those is inside a data centre in Melbourne owned by a rival called CDC. To build those data centres, Morgan Stanley told clients Firmus would rely on debt to cover 80-90 per cent of the cost. The bank was contacted for comment; Bank of America declined to comment, citing its usual practice on deals.
In the end, Firmus’ contracts were not the “slowly, then suddenly” inflection point that the banks had cast them as. That was the unravelling of the float. Despite missives from the bankers presenting the float as an all-but-done-deal on Monday, investors told confidants that they were alarmed that more than half of the shares in the company could be sold on the first day of trading. They were worried about the high price of the shares when so much about the company’s future was uncertain.
A hedge fund, Plato Investment Management, claimed to have identified 30 “red flags” in the company. It noted that Curtis had pitched AI token production as akin to mining. “It is actually a very good analogy,” Plato declared. “Large amounts of debt and equity are used to finance expensive equipment and infrastructure today to produce a commodity several years into the future. But the metaphor cuts both ways. Commodity industries are inherently cyclical.” Shortage turns to surplus as speculators rush in, crashing prices, the firm concluded.
As the blows rained down, Firmus’ bankers pulled every lever. They cut the price of shares from $11 to $8.25, and industry sources speculated that Firmus would have gone even lower, had they thought it would save the listing. They discussed a smaller share sale. They mulled changing the terms of the deal that let existing shareholders sell out so quickly. But once panicked talk made its way to the public domain, it was clear nothing would save Firmus’ plans.
In a tightly worded statement, Firmus conceded defeat on Friday morning, blaming “recent market volatility and prevailing market conditions”. More funding from the likes of Nvidia and then, perhaps, a Nasdaq listing could be an option, sources close to the company have suggested.
As bankers licked their wounds, Anne-Marie Bastian was celebrating. The president of the Environmental Awareness Association, a small community group formed to fight Firmus’ planned facility near Long Reach in northern Tasmania, rattled off a list of concerns about data centres: water use, pollution and noise among them.
The float’s flop, she said on Friday, “was a bit of good news for the morning.”
The Business Briefing newsletter delivers major stories, exclusive coverage and expert opinion. Sign up to get it every weekday morning.
About the Author
