By Brad Ferris · 8 August 2026
On 7 August the Australian Financial Review reported that Firmus had closed a $2.9 billion raise to accelerate its AI factory roll-out, with co-CEO Oliver Curtis saying the money lets the company push ahead with data centre construction across Australia and the Asia Pacific. It is the largest private AI infrastructure raise this market has seen.
The more interesting detail is where that money lands. Three days earlier, the AFR reported that Maas Group had put a further $300 million into Firmus for a 3.2 per cent stake and picked up $855 million of new electrical contracts tied to the build. Maas is a regional construction and materials business. The single largest disclosed Australian beneficiary of this AI capital, so far, gets paid to run cable and pour concrete.
That is worth sitting with, because it inverts how most operators have been told to think about AI. The conversation for two years has been about what AI does inside your business. The money moving through Australia right now is buying buildings, switchgear and power.
The raise is real, and the delivery is early. When shareholders approved the $2.9 billion at $230 a share on 31 July, the AFR noted that only one of the company's five Australian projects had planning approval: an 84 megawatt facility in Launceston, under construction, due to start operating in 2027. The company is working towards a listing valued around $15.5 billion.
So this is a funded pipeline, not an operating one. Announcements will keep outrunning delivery for at least a couple of years, and any operator reading a capacity announcement as though it were capacity should discount it accordingly.
The reason to watch this closely has less to do with silicon than with electricity. On 5 August the AFR reported that Energy Minister Chris Bowen had signalled the Commonwealth would not accept a "race to the bottom" among states competing to attract data centres at the cost of household bills, reliability or emissions, setting up a national cabinet fight over green data centre rules.
Whichever way that lands, it decides two things that matter to businesses with no AI project at all: where this construction happens, and what happens to industrial and commercial power prices in the regions that host it.
If you run a business of 20 to 300 people, this cycle reaches you through the order book and through the energy bill, well before it reaches you through a chatbot.
On the demand side, the named sites are public. Bell Bay in Tasmania, Tailem Bend in South Australia, the western Sydney corridor. Electrical, civil, HVAC, water, security, fencing, traffic management, labour hire, freight and site services are all being contracted now, and they are being contracted to businesses that can evidence capacity, safety systems and a credible schedule. That is a quoting and scheduling problem, not a technology one, and it has a closing window.
On the cost side, if you are an energy-intensive operation in one of the host regions, the outcome of the Commonwealth and state argument is a line in your P&L in a future contract year. Knowing when your electricity contract renews, and on what assumption, is cheap diligence.
There is a second number in circulation this week. EY-Parthenon modelling says AI could add between $95 billion and $116 billion to the Australian economy by 2036, on a multifactor productivity uplift of 2.0 to 2.4 per cent, supporting tens of thousands of additional jobs.
Treat that as what it is: a modelled potential, not an observed outcome. And note that the two numbers in this article do different work. The $2.9 billion is capital that has already moved. The $116 billion is a projection that only materialises if thousands of individual businesses change how work gets done.
Data centres produce compute capacity. Capacity is not productivity. Nobody builds a facility in Launceston that makes your quoting faster, your invoices cleaner or your service desk cheaper. That gap stays exactly where it has always been, inside your own four walls, and the national build does not close a millimetre of it for you.
Check your demand-side exposure. The projects and their tier-one contractors are public. Spend an hour working out whether any of this touches your pipeline, and if it does, whether your capacity story would survive a procurement conversation.
Know your energy reset. Find the renewal date on your electricity contract and the assumption behind the current rate. If you are in a host region, that assumption is now carrying policy risk.
Pick one workflow and price it. Choose a single repeatable process, measure what it costs today in hours and dollars, and you will have the only baseline that lets you judge whether an AI investment paid for itself. This is the work no infrastructure build does on your behalf, and it is the work that turns a projection into a return.
The capital cycle is real, it is Australian, and it is happening whether or not you have an AI strategy. The productivity is a separate job, and it is still yours.
Where does your business stand? The free AI Scorecard takes three minutes and shows you. If you want a straight steer from a person, book an AI Opportunity Call.