Two months ago, Firmus Grid was worth US$10.5 billion to the people funding it. This week it asked public investors for US$30.4 billion. They said no. The Nvidia-backed cloud-computing company pulled what would have been Australia's biggest IPO in almost 30 years. John Pearce, who runs investments at the retirement fund Unisuper, put it simply. Firmus has a compelling story and “just doesn't have a compelling valuation.”
Firmus rents out computing power built on Nvidia (NVDA) chips to big tech customers, OpenAI and Meta Platforms (META) among them. The Journal's Stuart Condie reports it wanted to raise $5 billion at a $30 billion valuation. The exact target was A$43.7 billion, or US$30.4 billion. Only two months earlier, a US$2 billion private round that Blackstone and Coatue took part in had valued the company at US$10.5 billion. Nearly triple, by our arithmetic. Many analysts and investors thought that too rich. Firmus runs two data centers today. Five more are in development in Australia, Malaysia and Indonesia. Pearce's fund sat out the deal. Firmus says it will look for more private money instead.
Two things made the price harder to swallow. First, bond yields. My Bui, an economist at the Australian wealth manager AMP, said investors in a higher-yield world demand more pay for risk, in stocks too. That, Bui said, puts a cap on lofty valuations. Second, the chips themselves. Nobody knows how long they stay useful. Better ones keep arriving, so operators keep buying while writing down the value of the older ones. The paper groups Firmus with Nasdaq-listed CoreWeave (CRWV) as neoclouds, companies that rent computing power to AI firms. CoreWeave runs 51 facilities and carries a $37 billion market value. Its shares fell nearly 8% Thursday on worries about AI companies' profits, then rose less than 1% Friday. Plenty of analysts stay upbeat. Toby Hearst of Yardeni Research says demand for computing still far exceeds supply.
Meanwhile, in London
In London, Airtel Money got a cool reception of its own. The African mobile-money business sold 270 million shares at 1.96 pounds, about 10% of the company, and existing holders did the selling. That valued it at 5.29 billion pounds, about $7 billion. In September the target had been $8 billion to $9 billion. The shares opened higher and closed down 6% at 184 pence, about $2.44. This is a real business. It earned $676 million in the fiscal year that ended in March, before interest, taxes and the accounting charges for aging assets. It has 53 million monthly users across 13 markets in sub-Saharan Africa. London badly wanted the listing after a run of takeovers and moves to New York. Buyers who paid the reduced price were still down 6% by the close.
Our read
Investments and risk (IN04). A good business at the wrong price is a bad investment. Higher bond yields lower the right price for profits that are years away. That's the whole lesson of this week in IPOs. The discipline came from a pension fund's investment chief saying no. That is the job. Now the uncomfortable part. CoreWeave, the story's U.S. comparison, sits in one of our model books at a 16% weight. It is the only neocloud in our books. Held at weight, nothing added, and we won't defend the valuation. A pulled IPO for a similar company is that position's risk written out in the paper. Nvidia is held in 31 books and Meta in 20, both at weight.
For a household, two rules. A first-day pop is never part of the deal, and here the sellers were existing holders cashing out at a cut price. If a new issue reaches you, ask who is selling and why. Then ask what the business earned last year, before you ask what it might earn years from now. You pick the price you'd pay on a dry day. Then the rain can't rush you.