Firmus Pulls $31 Billion IPO as AI Investors Choose Caution Over FOMO

Firmus Technologies wanted a lot, and the market said no.

The Nvidia- and Blackstone-backed data center developer scrapped its initial public offering on Friday after seeking a A$44 billion ($31 billion) valuation. That figure was triple what investors assigned the company just two months earlier in an August funding round, and 23 times its worth a year ago. At more than $5 billion, the planned raise would have been the largest IPO in Australia in nearly three decades. Instead, it never priced.

The math simply didn't hold up for buyers. Co-CEOs Tim Rosenfield and Oliver Curtis walked into their roadshow confident. They had signed contracts with hyperscale clients including OpenAI. Existing shareholders had committed to buying roughly half the offering. By most conventional measures, that's a strong setup for a deal.

It wasn't enough. What started as fear of missing out curdled into something closer to fear of getting screwed.

Why the Deal Fell Apart

Timing didn't help. Geopolitical tension has rattled global markets broadly: oil jumped 5% in a single session on Thursday. AI stocks specifically have been under pressure for months. CoreWeave, a rival neocloud operator, has shed 41% of its value since peaking in May. Both Anthropic and OpenAI have separately pushed back their own IPO timelines, suggesting the caution extends well beyond one Australian issuer.

But macro headwinds were known quantities before Firmus launched its pitch. The deal's real problems were structural.

Firmus has just 42 megawatts of operating capacity against a stated goal of 1 gigawatt, a gap that invites skepticism about execution. The company also pivoted in recent months from a primarily domestic Australian focus to targeting Asian markets more broadly. Following where clients lead makes business sense. But the shift raised fresh questions about whether Firmus could actually hit the targets underpinning its valuation ask.

Its own Australian joint-venture partner didn't do the company any favors. CDC Data Centres CEO Greg Boorer said publicly on Tuesday that the venture between the two firms was essentially dead. That's not the kind of commentary that inspires confidence in institutional buyers mid-roadshow.

Firmus also leaned on unconventional valuation metrics to justify its pricing, a tactic that tends to draw scrutiny rather than deflect it. Then there was the structure: current investors, who own almost 60% of the company, would have been free to sell their stakes the moment trading began. That's a setup practically engineered to pressure the stock lower on day one, and sophisticated buyers noticed.

Supporters of the deal had pitched Firmus as the BHP of data center providers, a reference to the Australian mining giant's ability to extract iron ore more cheaply than competitors. It's a flattering comparison and a deliberate one. BHP is Australia's largest listed company and among its biggest exporters and taxpayers. Firmus, with 42 megawatts online and a two-month-old pivot to a new region, is not yet in that conversation.

What This Means for AI Infrastructure Bets

The Firmus collapse is a data point, not an isolated anomaly. Taken together with CoreWeave's 41% drawdown and the delayed IPO plans at Anthropic and OpenAI, it suggests investors are starting to separate AI infrastructure companies with durable contracts and proven build-out capacity from those selling a growth story on thinner evidence.

Signed hyperscaler contracts and committed existing-shareholder demand used to be enough to get a deal done. They weren't sufficient here. Buyers wanted proof of execution at scale, and Firmus couldn't yet show it. A tripling of valuation in two months, absent a commensurate jump in operating capacity, is a hard sell even in a sector still attracting enormous capital.

The lock-up structure deserves particular attention from anyone evaluating future neocloud or AI infrastructure listings. Allowing nearly 60% of existing ownership to sell immediately at IPO is an aggressive term that shifts risk onto new shareholders right out of the gate. In a market still willing to chase AI exposure, that might have been tolerated. In a market that's grown more selective, it becomes a red flag worth walking away from.

The Bottom Line

Firmus didn't fail because AI infrastructure demand evaporated. Hyperscalers are still signing contracts, and capital is still flowing into data center buildouts globally. Firmus failed because its pricing outran its proof points, and its deal terms asked buyers to absorb risk that sellers weren't willing to retain themselves.

For investors watching the broader AI infrastructure trade, the lesson is straightforward: valuation multiples detached from operating capacity are getting tested now, not rewarded. Contracts alone no longer clear the bar. Track records, transparent metrics, and sensible lock-up structures matter more than they did a year ago, when the sector was riding pure momentum.

FOMO built this bubble in neocloud valuations. What killed the Firmus deal was something closer to due diligence reasserting itself. That's a healthy development for the sector's long-term credibility, even if it's a painful one for this particular issuer. Expect more scrutiny, not less, as additional AI infrastructure names approach public markets in the months ahead.