Australian AI Infrastructure Firm Firmus Raises $2 Billion From Nvidia, Coatue for Asia-Pacific Data Centre Push

Firmus, an Australian AI infrastructure company most people outside the data centre industry have never heard of, just pulled in $2 billion from Nvidia, Coatue Management, Blackstone-affiliated vehicles and Jane Street — pushing its valuation past $10.5 billion in a single stroke. The fully subscribed equity round funds the next phase of "Project Southgate," a sprawling AI data centre buildout that starts in Tasmania and Sydney and is designed to spread across the Asia-Pacific region over the coming years.

The round is notable less for its size — nine-figure and ten-figure AI infrastructure raises have become almost routine in 2026 — than for what it confirms about where the capital is actually going. It is not going into new AI labs, new consumer products, or new hiring pipelines. It is going into concrete, steel, transformers and racks of Nvidia silicon, in a country better known for iron ore and gas exports than for AI compute. Firmus has now raised more than $3 billion in equity over the past twelve months alone, on top of a roughly $10 billion Blackstone-led debt package secured in February. Its valuation has nearly doubled since April, when it stood at $5.5 billion.

Who Actually Wrote the Checks

The investor list reads like a snapshot of where institutional capital believes the AI infrastructure trade is headed next: not into another foundation model lab, but into the physical plumbing underneath every model that gets trained or served. Nvidia and Coatue Management, both existing backers, returned as follow-on investors — a signal that early conviction on Firmus has only strengthened. The new money came from Blackstone Tactical Opportunities and other Blackstone-affiliated vehicles, alongside Jane Street, the quantitative trading firm whose appearance on an AI data centre cap table underlines just how mainstream this asset class has become for firms that historically stuck to markets, not megawatts.

A Valuation That Nearly Doubled in Four Months

What makes the round striking is the pace of re-rating. Firmus was valued at $5.5 billion in April 2026. By August, four months later, it had cleared $10.5 billion — a near-doubling driven almost entirely by the credibility of its build pipeline rather than by revenue growth in any conventional sense. That is the AI infrastructure market in 2026 in miniature: valuations are increasingly a function of gigawatts under contract and land secured, not customers served.

Milestone	Date	Detail
Blackstone-led debt package	February 2026	Approximately $10 billion in project-level financing
Equity valuation	April 2026	$5.5 billion
Nvidia deepened partnership	June 2026	Agreement covering Nvidia infrastructure purchases and cloud services
Latest equity round	August 2026	$2 billion raised; valuation exceeds $10.5 billion

Project Southgate: The Physical Buildout

Strip away the funding headline and Project Southgate is, at its core, a construction programme — one measured in megawatts, transformers and grid connections rather than in lines of code. Firmus builds around Nvidia's DSX AI Factory reference architecture, positioning itself as an "AI factory platform" company: it does not train models or sell consumer AI products, it builds and operates the physical facilities that let other companies do so.

The Australian Core

The immediate build-out centres on Australia, where Firmus is expanding a campus in Tasmania anchored by a 90MW facility in Launceston, with further capacity growth planned for the existing Tasmanian site. From there, the company has signalled expansion into Melbourne, South Australia and New South Wales — effectively stitching together a national AI compute backbone in a country whose electricity grid was never designed with gigawatt-scale AI training clusters in mind.

The Asia-Pacific Reach

Beyond Australia, Firmus is taking its first major step offshore with a 360MW Nvidia DSX AI Factory campus in Batam, Indonesia, developed in partnership with Singapore-based DayOne and targeted for operation in the first quarter of 2027. Batam's position just across the strait from Singapore — a market increasingly constrained on data centre power and land — makes it a logical release valve for regional AI compute demand, echoing a broader pattern of hyperscale capacity migrating to wherever power and permits are easiest to secure.

The Capital-to-Headcount Problem

Here is the number that does not appear in any press release: how many permanent jobs $2 billion of fresh AI infrastructure capital actually creates. It is a fraction of what the dollar figure might suggest, and that gap is the real story behind every one of these mega-rounds.

Where the Money Actually Goes

Capital raised for hyperscale AI data centres overwhelmingly funds:

• Nvidia GPUs and networking hardware: Frequently the single largest line item, often exceeding the cost of the building itself
• Grid connections and power infrastructure: Substations, transformers and, increasingly, dedicated generation to feed facilities the existing grid cannot support
• Land acquisition and construction: Site works, cooling systems and shell buildings engineered around power density rather than office footprint
• Financing costs: Debt service on the billions in project-level leverage layered beneath the equity, such as Firmus's own ~$10 billion Blackstone-arranged facility

What is conspicuously absent from that list is a large, ongoing wage bill. Once a facility is commissioned, hyperscale AI data centres run with lean permanent staff — typically technicians monitoring power, cooling and hardware health, plus a small security and facilities crew. The construction phase is genuinely labour-intensive and temporarily employs far more people, but that workforce moves on to the next site once the concrete cures and the racks are populated. The operational headcount that remains behind is a rounding error next to the billions deployed to get there.

What This Means for the Asia-Pacific Labour Market

For Australia and Indonesia, the calculus is straightforward and somewhat uncomfortable:

1. Construction booms are real but temporary: Tasmania, Melbourne and Batam will all see genuine construction employment spikes as Project Southgate sites break ground, but these are project-length jobs, not careers
2. Permanent operations headcount stays small: The steady-state jobs — data centre technicians, electrical engineers, security staff — number in the dozens to low hundreds per site, not the thousands that a $2 billion manufacturing investment might once have generated
3. The economic value captured is disproportionately in power, land and tax revenue: State and national governments benefit through electricity sales, land leases and corporate tax, rather than through broad-based wage growth
4. Regional competition intensifies, not for workers, but for grid capacity: Singapore's power constraints pushed this specific project toward Batam; the next constrained market will push the next project somewhere else — the site selection criterion is megawatts available, not workforce availability

This is not a criticism unique to Firmus — every hyperscaler and AI infrastructure developer from Amazon to Microsoft to Meta shares the same capital-intensity profile. But Firmus's rapid valuation growth, from $5.5 billion to $10.5 billion in four months on the back of build pipeline rather than revenue, makes the pattern unusually visible. The company is, in effect, a pure-play bet on the proposition that AI compute demand will keep outstripping supply for years to come — a bet institutional investors are now making at a scale that dwarfs almost any comparable investment in human capital in the same region and the same period.

What Comes Next

Project Southgate's next twelve months will be a useful test case for how quickly capital can be converted into operating megawatts in a market as power-constrained as Australia's. The Launceston facility and the wider Tasmanian campus expansion are the most immediate proof points, while the Batam campus in Indonesia — Firmus's first venture outside Australia — will show whether the company's build model travels beyond its home grid.

For HumansAreObsolete readers, the takeaway is less about Firmus specifically and more about the shape of the AI economy it exemplifies: enormous, fast-moving pools of capital chasing physical infrastructure that requires comparatively few humans to run once it is built. The jobs argument for AI, when it is made at all in rounds like this one, tends to be about the temporary construction surge rather than the permanent operational workforce — and investors clearly do not need it to be about jobs at all. They need it to be about megawatts, uptime and Nvidia allocation. On all three counts, Firmus just told the market it is exactly where it wants to be.