Nvidia-Backed Firmus Grid Cuts IPO Price as Investors Balk at Deal
Firmus Grid pursued an A$43.7 billion valuation in one of Australia's biggest planned listings. This week, with investors balking at A$11 a share, reports said the price was cut to A$8.25 and bankers were trying to close the deal.
Oliver Curtis, who co-founded Firmus Grid, set out to raise as much as A$7.1 billion for the Australian AI data center developer. This week that plan hit a wall: the company closed the books on its IPO after investors balked at the price, according to Bloomberg. At A$11 a share, the offering would have valued Firmus Grid at roughly A$43.7 billion ($30.3 billion), making it the second-largest listing in Australian history behind Telstra's 1997 debut. Instead, The Australian and the Australian Financial Review reported the offer price was cut as low as A$8.25 as bookbuilding wrapped, with bankers trying to get the deal across the line at all.
The math investors pushed back on is not complicated. Firmus is losing money, and it told prospective buyers to expect a A$77 million loss in the first half of fiscal 2027, a figure first reported by Reuters and echoed across Australian financial press. More importantly, only about 5% of the data center capacity Firmus has sold to customers is actually built and running. Compare that to NextDC, the ASX-listed data center operator that trades at a fraction of Firmus's multiple, which runs roughly a quarter of its sold capacity live today. Two of Firmus's facilities are operational. The rest exist on paper, in planning approvals, or mid-construction.
Nvidia's name is all over this deal, and that is precisely why it matters as a signal beyond Australia. Nvidia, CDC Data Centres and Firmus struck a partnership worth up to A$73.3 billion last October to build AI factories across the country, but CDC and Firmus have since ended the planned Project Southgate rollout after only an initial slice of capacity was delivered. Nvidia remains tied to Firmus as a chip supplier and shareholder, and Firmus has separately described Nvidia-linked revenue-sharing arrangements for new AI factory capacity. Coatue led a $505 million funding round into Firmus earlier this year, with Nvidia participation subject to closing conditions. That is serious company for a young startup. But a chip maker's partnership and a venture round are bets on a thesis playing out over years. An IPO book is a bet due in weeks, priced against actual revenue, and Firmus doesn't have much of that to show yet.
Bloomberg also reported a stock overhang weighing on the deal: more than half of existing shares were set to be freely tradable when Firmus lists, while escrow agreements would restrict 42.4% of the register, including part of the holdings of founders, family members and other shareholders. Separate reporting said roughly half of the IPO book was likely to go to existing investors, potentially letting backers such as Nvidia and Blackstone top up rather than be diluted. That structure can spook investors who want their money building capacity, not just reinforcing the old cap table.
There's a human story sitting underneath the spreadsheet, and it's hard to ignore. Curtis was convicted of insider trading in 2016 and served roughly a year in prison after a scheme that netted $1.4 million in illegal profits trading CFDs on confidential takeover information passed along by a friend. He co-founded Firmus in 2019, not long after his release. A A$43.7 billion valuation would have made him an AI billionaire on paper inside a decade of walking out of prison. The Nightly covered that turnaround in detail, and it's the kind of redemption arc that makes for a great headline. It just doesn't make a loss-making company with 5% utilization worth $30 billion on its own.
What this tells you about the AI infrastructure trade
Frankly, this is the test AI infrastructure financing has been waiting for. Chip demand stories and data center buildouts have carried premium valuations all year on the promise that capacity will eventually fill up and investors just need to be patient. Samsung projected roughly $80 billion in quarterly operating profit this month on memory shortages tied to AI demand, and SK Hynix is reportedly already lining up bankers for a Solidigm IPO that could value that business near $150 billion. Those numbers kept the broader AI capex story looking unstoppable. Firmus is the first major test of whether public equity investors will pay up front for that promise when a company can't yet show the revenue to back it.
The answer, so far, is no. Public markets are pickier than private ones, and a bookbuild forces every investor to put a number on the gap between what Firmus says it will become and what it actually is today. That gap, on the evidence of this week, was worth roughly A$15 billion of the original ask.
Firmus had been targeting an October 23 ASX debut. Whether it gets there at A$8.25 a share, lower still, delayed or not at all will say more about where the AI infrastructure boom actually stands than any ministerial speech about the sector's future.
Also read: • Samsung's quarterly profit jumped ninefold to roughly $80 billion on AI chips • SK Hynix's Solidigm picks banks for a US IPO that could value it at $150 billion
This article is posted in Technology News, check it out for more related stories.