Hot
In what was slated to become the second-largest corporate debut in Australian history, artificial intelligence infrastructure darling Firmus Technologies abruptly withdrew its planned US$5.5 billion (A$7.1 billion) initial public offering on the Australian Securities Exchange (ASX) on Friday morning. The company—closely tied to hardware partner and equity backer Nvidia, alongside private equity giants Blackstone, Coatue Management, and Jane Street—officially cited “market volatility and prevailing macroeconomic conditions” for shelving the listing.
Yet behind the polite corporate boilerplate lies an undeniable inflection point for the broader artificial intelligence economy: institutional investors are staging an open revolt against dizzying multiples for data center capacity that exists primarily on pitch decks. Firmus’s listing would have valued the company at an eye-watering A$43.7 billion ($30.6 billion)—trailing only the historic 1997 privatization of telecom monopoly Telstra on the Australian bourse. Instead, institutional bookrunners faced cratering demand and sharp price pushback, forcing the board to pull the pin rather than accept a humiliating valuation haircut.
The A$44 Billion Mirage: How A Crypto-Mining Pivot Tripled Its Paper Worth in Eight Weeks
The speed at which Firmus attempted to scale Wall Street’s AI hype cycle into public equity was breathtaking. Founded in 2019 by Oliver Curtis, Tim Rosenfield, and Jonathan Levee as a specialized immersion-cooled Bitcoin mining venture, Firmus completed an aggressive corporate pivot into so-called “AI factories” as generative AI demand took flight. Leveraging proprietary liquid immersion cooling technology designed to handle extreme heat loads from dense GPU clusters, the firm attracted strategic backing from Nvidia, which supplied thousands of cutting-edge Blackwell and Hopper chips.
However, fund managers performing pre-IPO due diligence balked at the math. In August 2026, Firmus was valued in private funding rounds at roughly A$10.5 billion. By late September, leadership sought an audacious target of A$44 billion at A$11.00 per share—effectively demanding a 300% markup within 60 days without a corresponding expansion in physical revenues.
| Operational & Financial Metrics | Firmus IPO Ambition | Grounded Reality (Q3 2026) |
|---|---|---|
| Implied Equity Valuation | A$43.7 Billion (~$30.6B USD) | Withdrawn after severe institutional pushback |
| Active Operational Footprint | Projected 5 Pan-Asian Hyperscale Hubs | 2 Data Centers (Melbourne & Singapore) |
| Operational Capacity Realization | 100% Fully Monetized Cloud Compute | ~5% of Claimed Capacity Currently Online |
| Core Hardware Dependency | Proprietary Liquid Immersion Platform | Heavily concentrated on Nvidia GPU delivery |
| Primary Capital Source | Public Equity (ASX Public Float) | Reverted to Private Credit & Sovereign Funds |
“Fund managers simply refused to underwrite an asset where 95% of the projected cash flow relies on building future facilities that haven’t even broken ground,” noted one Sydney-based equities analyst who evaluated the order book. “The market is drawing a hard line between companies selling real chips today and speculative landlords hoping to house them tomorrow.”
The Hardware Depreciation Trap: The Looming Threat of Obsolete Silicon
A core vulnerability driving investor skepticism was the brutal depreciation cycle inherent to frontier artificial intelligence computing. Unlike traditional commercial real estate or utility infrastructure—where assets retain value over 30-year spans—AI data centers run on silicon that depreciates precipitously every 18 to 24 months as next-generation accelerators render older architectures economically unviable for frontier training runs.
(Responsive / Native Ad)
Firmus’s business model required continuous, multi-billion-dollar capital expenditure to purchase Nvidia’s latest server racks. In an environment where tier-one cloud providers like Microsoft, Amazon Web Services, and Google are expanding their own proprietary custom silicon (such as Maia, Trainium, and TPU clusters) to lessen their reliance on third-party GPU colocation, pure-play infrastructure landlords risk getting caught in a margin squeeze between staggering hardware amortizations and falling spot prices for rented compute tokens.
The Retreat to Private Credit: Why the AI Capex Beast Can No Longer Rely on Public Markets
Rather than restructuring the float at a realistic discount, Firmus announced it will immediately pivot toward private debt and sovereign capital. This retreat into private credit mirrors a broader structural shift across the technology sector. As public equity markets demand GAAP profitability, positive free cash flow, and clear payback horizons, capital-intensive AI startups are increasingly forced into private credit arrangements, where pension funds and sovereign wealth entities accept illiquid, high-yield structured debt.
The collapse of the Firmus float sends a direct warning to upcoming venture-backed AI infrastructure floats waiting in London, Tokyo, and New York. The era of claiming a multi-billion-dollar multiple simply by mentioning liquid cooling and Nvidia allocation appears to be closing. For public market investors, the question is no longer how many GPUs an enterprise can promise to rack, but whether the software models running inside them can ever generate the revenue required to pay for the electric bill.
