Marcus Today straight talk

Australia’s SpaceX brought back down to earth. The hype is coming out of the price. Talk started at $50bn, the offer was set at $11 ($43.9bn), and on Wednesday night the float was repriced, with fund managers now expecting about $9 a share – roughly $37bn. The numbers add up on paper: US$67.8bn of contracts and a US$5.8bn EBIT target. But they rest on 865MW still to be built, roughly US$30bn of debt at a time AI borrowing costs are rising, and revenue last year of just US$50.8m. At $9 the price is close to our central estimate, so there is little margin for error. At first impressions, our fair value is 4–32% below the original $11 offer price, and roughly in line at $9. Please note: Firmus is valued relative to global peers and there are differing opinions as to how fairly valued these peers currently are. Expect the volatility in share price to be high post-IPO.

In brief

Firmus is raising $7.1bn (US$5bn), up to $7.9bn with the greenshoe – the second-largest Australian IPO after Telstra. The offer was set at $11 a share but is being repriced, with about $9 expected. More than half the book is going to existing backers, including Nvidia (7.2%) and Blackstone (6.7%). It has 46MW running and 865MW in five projects under way in Indonesia, Malaysia and Tasmania, contracted to Meta, OpenAI and Fireworks AI. Investors were told the sites would generate US$5.8bn of EBIT (earnings before interest and tax) once built. The institutional bookbuild closes at 9am on Thursday 8 October.

We estimate fair value at $30–42bn, with a central estimate of about $36bn. At the original $11 ($43.9bn) the float was priced above the top of our range. At the expected $9 (about $37bn) it sits inside our range and about 4% above our central estimate of $8.67 a share, with the probability-weighted value at $9.46. The disclosures reconcile on paper, but they depend on 865MW being built on time and about US$30bn of new debt as AI borrowing costs climb.

Key reasons

What would change the view?

A final price below $9, a prospectus confirming the take-or-pay terms and the cost of the debt, and Batam and Launceston starting on time in early 2027 would make the stock more attractive. Further repricing on weak demand, rising debt costs, delays in Batam or Tasmania, or heavy early selling would make it less so.

Firmus is a seven-year-old Sydney-based builder and operator of ‘AI factories’ – liquid-cooled data centres packed with Nvidia GPUs that it rents to AI labs and hyperscalers on multi-year, mostly take-or-pay contracts. It started in Launceston, Tasmania, running Bitcoin mining on cheap hydro power, then pivoted to AI compute. Today it runs 46MW in Melbourne and Singapore and has 865MW in five projects under way. Revenue last financial year was US$50.8m. Its private valuation has gone from $1.85bn in September 2025 to US$10.5bn (about $15bn) in August 2026. The IPO was set at $11 ($43.9bn) and is being repriced to about $9 (about $37bn). Beyond today’s 911MW it has 2.9GW planned in Australia and an agreement with Nvidia to buy up to 10GW of GPUs over five years.

Key shareholders (disclosed or reported): Nvidia (7.2%), Blackstone (6.7%), Coatue, Jane Street, Maas Group (ASX: MGH), Regal Funds Management, Ellerston Capital, Archibald Capital and the founders, who hold about 15% between them, with relatives holding a further 8%. Maas Group is the only listed Australian way to own a piece of Firmus before the float: it holds a stake worth $400m at cost and an $855m contract to supply Power Cube electrical units.

Repriced to about $9, the offer moves inside our fair value range

Equity value in A$bn. The original $11 offer was above our range; the expected $9 is close to our central estimate.

Even at $9, the valuation has risen about 20-fold in 13 months

Post-money valuation in A$bn at each raise. The IPO bar uses the expected $9 offer.

Investors bought in at a US$10.5bn valuation in August. At $11 the IPO valued Firmus at US$30.7bn – about 2.9 times that. After offshore demand came in lighter than expected, the float was repriced on 7 October, with about $9 (US$26bn) now expected.

What we are watching: the final price, how much of the book existing holders take up, the cost of the next debt package, and how the shares trade in the first week against a free float of more than half the company.

Firmus has 911MW contracted, but only 46MW is running. In our view it is a contract book, not yet an earnings stream.

How it makes money

Firmus fits out the data hall, buys the Nvidia systems and sells GPU capacity on multi-year contracts, mostly take-or-pay – customers pay for reserved capacity whether they use it or not. This is the ‘neocloud’ model used by CoreWeave and Nebius. The three Asian projects sit in leased third-party data centres; the Tasmanian sites are its own. Server budgets across the disclosed sites total about US$36bn.

Customers

Meta – uses the Melbourne site, has contracted capacity at Batam and signed a further South-East Asia deal in early October. OpenAI – dedicated capacity at two Kuala Lumpur AI factories on Nvidia’s Vera Rubin platform. Fireworks AI – part of the Batam capacity. Asia accounts for about 86% of contracted revenue.

At about $9, Firmus’s equity alone is worth about US$29m per contracted MW – a modest premium to CoreWeave (US$26m) and IREN (US$21m). Add the roughly US$30bn of debt it is expected to need and the figure is about US$62m per MW, close to Nebius, for a company with 46MW running.

The numbers now reconcile – on paper. Investors were told the sites would generate US$5.8bn of EBIT once built. Revenue last financial year was US$50.8m, and the draft prospectus reportedly points to a pro forma loss of about US$77m in the first half. We checked the target from the bottom up.

Consensus on the listed peers supports caution: all three are forecast to lose money next year. In our view 12x the US$5.8bn target, less the debt, is worth about $57bn once every site is built – but that is two years and 865MW away, and every 1% on the cost of US$30bn of debt is about US$300m a year off pre-tax profit.

Per-share values assume about 4.15bn shares at $9 (the $7.9bn raise unchanged). At $9 our base case implies about 4% downside and the probability-weighted value about 5% upside. At the original $11, the same values were $9.02 and $9.84 a share – 18% and 11% below the offer.

What could hold the price up after listing

A lower entry price, strategic holders taking more than half the book, likely fast entry to the ASX 200 and strong Australian demand for AI exposure. CoreWeave priced below its range in March 2025 at US$40 and now trades at US$88.45. Against that, about 58% of the company can trade from day one. Early trading could run well ahead of fundamentals, in either direction.

General information only. This content does not take into account your objectives, financial situation or needs. Consider whether it is appropriate for you and read the relevant PDS or prospectus before acting. Forecasts are possibilities, not guarantees.