Almost a year ago, Firmus and CDC announced a plan that sounded like it would remake Australian digital infrastructure. The program was said to involve up to AU$73.3bn of investment, with CDC providing the underlying data centre infrastructure. Last week it came apart. The two have stopped working together after deploying just 42MW of capacity, roughly 2.5% of the 1.6GW headline.
What actually happened
The wording matters here. According to the AFR, Firmus will continue to occupy a CDC data centre in Melbourne but will no longer expand to the 150MW initially planned for the site. So this is a stalled expansion rather than an eviction. Firmus’s existing customers in Melbourne (including META) are not affected.
The two sides tell it differently. Firmus’s CEO says the companies “mutually agreed earlier this year” not to proceed, and that the decision doesn’t affect its development plans or contracted customer capacity. CDC’s Greg Boorer puts it more pointedly. He said Firmus had “made other choices regarding doing their own data centre developments”, which was different from what CDC had envisaged.
The strategic read
This looks like a strategy divergence more than a collapse. CDC is a colocation and wholesale operator that builds for contracted demand. Firmus wants to own the whole stack, from land and power through to GPUs and customers. Firmus’s own Australian sites, about 430MW in northern Tasmania and 2.7GW planned in South Australia, sit outside the CDC agreement. Meanwhile its next big builds are offshore. Batam and Malaysia carry its next two years of builds.
For Melbourne, that has consequences. Melbourne was meant to be the anchor of Firmus’s southern-states presence. If further Australian capacity comes from Firmus’s own sites in Tasmania and South Australia, Victoria’s role as a hub for this particular player shrinks. That is a loss for the local ecosystem of power, fibre, cross-connects and interconnection. Pure-play AI campuses still need diverse terrestrial and subsea connectivity, and for those of us in the cable world, where the capacity lands matters as much as who builds it.
The IPO problem
The timing is awkward. Firmus is set to list on the ASX on October 23, and the AFR reports it has cut its price from A$11 to A$9 because of weak demand from local, international and retail investors. The original ask was ambitious: a A$43.7 billion valuation for a company founded in 2019 that still expects to post a loss. The debt also worries investors. Morningstar estimates the debt burden would represent about six times Firmus’s forecast US$5 billion of operating earnings for 2028.
The CDC split doesn’t create these concerns, but it feeds them. Fund managers had already questioned whether Firmus can execute fast enough to justify its valuation. Delivering 2.5% of a flagship program in twelve months is the kind of data point those questions feed on.
Likely outcomes
- The IPO proceeds at a lower price (most likely). The reported cut to A$9 suggests the bankers are adjusting rather than pulling the deal. Strategic insiders including Nvidia, Blackstone and Coatue were reported to be taking about half the allocation, which gives the book a floor. Expect a smaller raise and a lower valuation than the headline, with early trading sensitive to sentiment.
- A delay. If the book doesn’t firm up at A$9, a postponement is plausible. It would carry reputational cost, and it would be read as a signal about the Australian appetite for AI infrastructure paper generally.
- A narrative reset after listing. Firmus can argue that the CDC exit is a tidy-up, because its contracted capacity is with Meta and OpenAI in Indonesia and Malaysia and its Australian land is its own. The question investors will keep asking is whether this is still “sovereign Australian AI” or an Asian build-out with an Australian listing. Firmus has signed agreements with Meta and OpenAI to provide capacity in Indonesia and Malaysia respectively.
- Flow-on for CDC and the sector. CDC isn’t unscathed. An independent valuation lowered CDC’s value by $78 million in the September quarter, citing higher interest costs. It still reports a growing pipeline and rising contracted capacity. Other operators will take note of how a very large “headline” AI program shrank in practice. Capital is available for AI infrastructure but it is more selective about who gets it.
The bigger lesson
Headline gigawatt numbers and dollar figures are announcements, and deployed megawatts are delivery. The next 12 months will show how many of Australia’s big AI data centre plans end up as live, powered, connected capacity. Power access, grid timelines, fibre diversity and customer contracts will decide that more than press releases.
A caveat on balance: Firmus’s position is that its contracted customer capacity is intact and its strategy unchanged, and with Nvidia, Blackstone and Coatue involved it has credible backing. The bearish reading rests on valuation and debt rather than on any demonstrated failure of demand.
