The IPO has become a test of AI-infrastructure enthusiasm
Firmus Technologies entered October as one of Australia's most closely watched floats. The company sits at the centre of the AI infrastructure boom: data centres, GPU capacity and hyperscale computing. But the latest reporting suggests investors are pushing back on the price. Reuters Breakingviews says the offer has been repriced by about 25%, taking the proposed valuation from roughly A$44 billion to about A$33 billion, while the planned raise has been reduced to around A$5.9 billion.
Why the valuation is so hard to judge
The core problem is a timing mismatch. Firmus is being valued on what it expects to build and earn in the future, while only a small fraction of the planned capacity is operating today. Reuters says around 42 megawatts are currently operational against a pipeline approaching 1 gigawatt. That does not mean the pipeline has no value, but it means investors must price construction, financing and customer-delivery risk years before all of the assets are producing cash.
| Key figure | What it tells investors |
|---|---|
| Reported revised valuation | About A$33 billion after a roughly 25% cut |
| Reported revised raise | About A$5.9 billion |
| Operating capacity | Around 42 MW |
| Pipeline | Approaching 1 GW |
| Scheduled ASX trading | October 23, subject to the offer process |
| Major backers reported | Nvidia, Blackstone, Coatue and others |
AI demand can be real while an IPO is still overpriced
It is possible for two things to be true at once: demand for AI compute can remain exceptionally strong, and investors can still decide a particular company is priced too aggressively. Data centres are capital-intensive businesses. They require land, grid connections, power contracts, cooling systems, construction, chips and debt financing before customers generate recurring revenue. A high-growth market does not eliminate the cost of building the infrastructure.
Firmus has changed strategy as it expands
Recent reporting says Firmus has shifted away from parts of its earlier Australian partnership model and is pursuing more capacity in Malaysia and Indonesia for large AI customers. International expansion can increase the addressable market, but it adds execution complexity: different power markets, construction environments, regulations, currencies and financing structures. Investors need to know whether each planned site has committed power, contracted customers and a realistic commissioning timetable.
Debt matters as much as the headline equity value
A data-centre operator can grow rapidly while also carrying a heavy financing burden. That is why investors should look beyond the equity-market valuation to enterprise value, debt maturities, interest costs and capital still required to finish projects. When global bond yields are high, the cost of financing new infrastructure can materially change project economics even if demand for GPU capacity remains strong.
Customer commitments need to be read carefully
Firmus has been linked to major AI names and reports a large contracted pipeline. The critical question is how much of that pipeline represents binding, funded and near-term revenue versus capacity that depends on sites being built first. Long-term AI infrastructure contracts can be valuable, but they may also contain conditions tied to delivery dates, power availability or performance.
Why only 42 MW operating capacity matters
Operating capacity provides evidence that a company can build, energise, cool and serve customers. A pipeline is a plan. The gap between roughly 42 MW in operation and a target near 1 GW is therefore the central execution challenge. Every additional site must move through planning, procurement, grid connection, construction, commissioning and customer acceptance. Delays at any of those stages can push revenue further into the future.
What retail investors should look for in the final documents
- The final offer price and number of shares issued.
- Net debt and the maturity schedule after the IPO.
- Exactly how IPO proceeds will be allocated between new projects, debt and existing holders.
- Operating versus contracted versus proposed megawatts.
- Customer concentration and termination clauses.
- Power and grid-connection agreements for each major site.
- Construction budgets, contingencies and commissioning dates.
- Related-party transactions and governance arrangements.
Does the repricing mean the AI bubble is bursting?
Not by itself. One difficult IPO can reflect company-specific valuation and execution concerns rather than a collapse in AI spending. But it is an important signal that public-market investors are becoming more selective. In 2025 and early 2026, many AI infrastructure deals were rewarded mainly for scale and speed. The Firmus debate suggests investors now want stronger proof of cash flow, operating assets and financing discipline.
The October 23 listing date is the next key milestone
Firmus has been expected to begin trading on the ASX on October 23. Until the final prospectus and offer terms are lodged, reported valuation figures should be treated as evolving deal terms rather than fixed facts. If the offer proceeds at a lower price, the first days of trading will provide a useful measure of how much demand exists outside the pre-IPO investor group.
Why this matters beyond one company
Australia is competing to become a significant AI and data-centre market, supported by renewable-energy potential, land and proximity to Asia. A successful major listing could help fund more domestic infrastructure. A weak listing could make future projects more expensive by raising the return investors demand. Firmus is therefore becoming a broader test of how public markets value the enormous capital requirements behind the AI boom.
Bottom line
The reported 25% repricing is a reminder that AI enthusiasm does not remove financial gravity. Firmus has ambitious plans and heavyweight backers, but investors are being asked to value a business whose future capacity is far larger than what operates today. The final prospectus should be judged on power, construction, customer contracts, debt and cash flow — not simply on the size of the AI market.