Oct 9, 2026 · 25 min
Firmus pulls IPO as AI infrastructure costs face scrutiny
Theory of Thing: A Firm No for Firmus
Firmus’s failed float exposes the funding, dilution, execution, and depreciation risks investors must weigh in the AI infrastructure boom.
- 1Firmus withdrew its proposed multibillion-dollar IPO after weak demand and unresolved questions about funding future capital expenditure.
- 2The hosts challenge GPU depreciation assumptions and compare them with practices at Microsoft and Google.
- 3Markets softened modestly while long-duration bonds, central-bank gold buying, and geopolitical oil risks shaped the outlook.
Don't miss
The hosts question whether Firmus can justify its GPU depreciation assumptions, turning the IPO discussion into a broader test of AI infrastructure economics.
The brief
Heath Moss and James Whelan examine Firmus’s abandoned multibillion-dollar float, arguing that weak investor demand reflected an expensive offer with too many unanswered funding questions.
The bullish case rests on contracted AI-factory capacity and prominent backing, but limited operating sites, construction delays, expected losses, and future dilution complicate the story.
The sharpest challenge concerns GPU depreciation: the hosts question whether Firmus’s assumptions reflect equipment lives realistically and compare the issue with Microsoft and Google.
The market report finds softer equities, potentially attractive long-duration bonds, continued central-bank gold buying, and range-bound oil exposed to shipping and geopolitical risks.
Small-cap updates on Renascore Resources and Lake Resources close the substantive discussion after Firmus’s trading halt brings the episode’s concerns into focus.
What was said on this episode
27 statements · 13 positive · 10 negative · 4 neutral
Firmus’s proposed valuation appeared excessive relative to its assets.
“a probably an evaluation people found a little bit spicy especially for some company that maybe didn't really have as much stuff as they thought they might have had”
Listen at 1:14
Firmus’s IPO was heavily valued and primarily based on promises.
“What was basically an IPO, a very heavily valued IPO built on basically a promise. That's it.”
Listen at 2:09
Firmus’s prospectus lacked detail on funding its capital expenditure.
“The prospectus was light on detail about how they were going to fund the colossal amount of capex.”
Listen at 2:46
Firmus would need to fund $50–70 billion of capex over three to five years.
“It's just like, you know, how are you going to fund this? Outside of the IPO, how are you going to fund – I think it was like $50 to $70 billion of capex over the next – five years or three to five years.”
Listen at 3:57
Firmus IPO investors would likely face dilution from future fundraising.
“So IPO listers would have been diluted out of their holdings over the next few years.”
Listen at 4:12
Firmus’s operating data centers were small and located outside Australia.
“the couple of data centers they had going were tiny in terms of the operation and not in Australia”
Listen at 4:21
Firmus was seeking an unattractive valuation versus listed Australian data-center peers.
“I couldn't believe what they were asking for and valuations have asked for when, how cheaply you can buy those other listed entities already.”
Listen at 4:43
Firmus’s bullish case involved more than 900 megawatts of contracted capacity.
“The bull case was 900. Megawatt plus?”
Listen at 4:59
Firmus had contracted capacity across seven AI factories in four countries.
“Contracted against seven AI factories in Australia, Singapore, Indonesia, Malaysia.”
Listen at 5:05
Firmus had two operating sites, five under construction, and expected a $77 million first-half loss.
“The bare case being that there are only two sites operating. Five still being built. Expect a $77 million Aussie first half loss.”
Listen at 5:23
GPUs depreciate rapidly.
“GPUs do depreciate fast.”
Listen at 5:46
Google still uses some GPUs that are eight to nine years old.
“Google have got GPUs at the moment that are eight, nine years old that they're still using”
Listen at 7:08
CoreWeave extended the useful life of many GPUs by three years.
“CoreWeave. have just extended the life of a lot of their GPUs for another three years.”
Listen at 7:14
Five to six years is currently a justifiable GPU depreciation period.
“for now, I think five, six years is a thing.”
Listen at 7:33
Megaport depreciates GPUs over two to three years, matching associated contracts.
“they're depreciating their GPUs over two to three years in line with contracts associated with those GPUs.”
Listen at 7:51
MGH held approximately 3% of Firmus and entered a trading halt.
“MGH, which holds about 3% of it, has just been into a trading hold as well.”
Listen at 8:25
OpenAI’s lower reported revenue does not indicate company slowdown.
“I don't think it is the company slowing down or anything.”
Listen at 13:18
Long-term bond yields appear near a top, potentially supporting bonds.
“they all look very topy to me. So we could see some bond support here”
Listen at 17:26
Structural support remains for gold, making current levels attractive for accumulation.
“The structural tailwinds behind gold are still there. So happy sort of accumulator of gold at these levels.”
Listen at 18:21
Shipping costs and insurance are currently keeping oil prices elevated.
“The only thing keeping oil prices high now are shipping costs and insurance, really.”
Listen at 19:30
Dallaroo Metals reported high-grade results from maiden drilling at its Bonduku Gold Project.
“Dallaroo Metals. Ooh, Dallaroo's a good one. They've got stuff in Côte d'Ivoire. High-grade hits in maiden drilling at the Bonduku Gold Project in Côte d'Ivoire.”
Listen at 20:44
Iltani Resources raised $4.6 million to accelerate drilling at its Orient project.
“Iltani Resources, ILT. raised $4.6 million to speed up drilling at its Orient Silver Zinc Indium project.”
Listen at 21:58
Cyclopharm’s Technegas was named a preferred lung-imaging ventilation agent in China.
“Cyclopharm, CYC. Technogas, named a preferred ventilation agent for lung imaging in China”
Listen at 22:25
Dimeryx reached its first patient-completion milestone in the DMX 200 program.
“Dimeryx, first patient completion milestone in its DMX 200 program.”
Listen at 23:02
Playside Studios’ Dumb Ways to Build launch exceeded internal forecasts.
“Playside Studios, dumb ways to build, launch, beat, internal forecasts.”
Listen at 23:07
Adveritas’s TrafficGuard fraud detection now covers ChatGPT advertising.
“Adveritas, traffic guard fraud detection now covers chat GPT advertising.”
Listen at 23:13
Sea Forest signed a sea-feed distribution deal targeting 480,000 cattle.
“Seaforest is sea feed distribution deal with Taze Australia, targeting 480,000 cattle.”
Listen at 23:20
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.