Firmus Grid pulls $5 billion IPO after asking 2.9 times its valuation of two months earlier

A target of $30.4 billion ^[3] over two operating data centers ^[5] comes to $15.2 billion per operating site ^[21]. CoreWeave's $37 billion across 51 facilities ^[17] is about $725 million a site ^[22], so Firmus was priced at roughly 21 times that figure ^[23]. The comparison is crude (sites differ in size, and Firmus was also priced on five more in development across Australia, Malaysia and Indonesia ^[5]), but it shows what the bookrunners asked investors to underwrite: a pipeline, at 2.9 times the valuation of a round two months old ^[24].

UniSuper did not buy into the offering, its chief investment officer John Pearce said ^[7]. "It all comes down to valuation: We think that Firmus indeed has a compelling story. It just doesn't have a compelling valuation," Pearce said ^[6]. A wider version came from Bloomberg's sources. "Investors still believe in AI," said Maxence Visseau, chief investment officer at Arkevium Capital. "What they won't do is pay any price for companies that spend huge amounts on data centers, depend on a few big customers, and promise profits years from now." ^[15]

Bloomberg, citing discussions with at least 10 investors and advisers, added a supply worry: about 58% of shares would be free to trade from day one, and existing holders might sell soon after listing ^[20]. A holder who came in at the $10.5 billion valuation would sit on a paper gain of about 190% at $30.4 billion ^[25]. Analysts also flagged Firmus's cost of debt after bond yields surged ^[10]. My Bui, an economist at AMP, said: "In a higher-yield environment, investors are clearly demanding greater compensation for risk, including in equities, which places a cap on lofty valuations." ^[8]

The two accounts differ on size and status. Mint reports a $5 billion raise that was withdrawn ^[1]; Bloomberg describes a $5.5 billion listing whose books closed Thursday before Firmus indicated a price or structure ^{[26] [13]}.

We think the sources support a narrower claim than thinning appetite for neoclouds, or rather the sharper version of it: one bookbuild failed on price, free float and debt cost at a company with two operating sites. CoreWeave fell nearly 8% on Thursday on worries about profitability ^[18], and a sector verdict would need more than one session of that. Toby Hearst, associate investment strategist at Yardeni Research, said: "The demand for compute still far outstrips supply. As long as the AI data center build-out continues, neoclouds should keep filling the gap the hyperscalers can't build fast enough to cover." ^[19] Nvidia holds a direct equity stake in Firmus and is also its customer and hardware supplier ^[9]; neither account says investors cited that tie in staying out.

The view is wrong if CoreWeave keeps falling after Thursday, if Firmus's private investors pay close to $30 billion (the failure would then sit with the float and the structure), or if Anthropic's book is weak, since that would show the discount reaches past infrastructure owners.