Firmus secures A$500M to scale AI infrastructure across Australia
Australian AI company Firmus will raise A$500 million (about US$324.9 million) to speed up Project Southgate, its build-out of AI infrastructure across the country. The round is backed by existing investors Nvidia and Ellerston Capital. It's the second raise in two months, following a A$330 million equity placement in September.
The focus is simple: more capacity, faster timelines, and a clearer path to large-scale training and inference inside Australia. For engineering teams in APAC, this points to more local compute and better access to accelerators over the next few years.
What Project Southgate aims to deliver
Project Southgate is Firmus's plan to deploy data centres and infrastructure for AI training and inference, in collaboration with CDC Data Centres and Nvidia. The initiative targets up to 1.6 gigawatts of capacity through 2028.
Proceeds from the raise will go into site development, infrastructure deployment, and energy agreements for AI factories in selected Australian locations. As Co-CEO Tim Rosenfeld put it: "With demand for AI infrastructure accelerating, this funding ensures we can meet it quickly, cost-effectively, and in line with Australia's renewable-energy future."
Why this matters for builders and teams
- Closer compute in APAC: Lower latency options for training and high-throughput inference within Australian borders.
- Data residency: Easier compliance for workloads that must stay onshore.
- Capacity outlook: Additional GPU clusters and high-bandwidth networking should ease waitlists and scheduling pain over time.
- Sustainability targets: Expect energy deals tied to renewables, impacting procurement and reporting for ESG-conscious teams.
Technical signals to watch
- Sites and timelines: Which regions go live first, and when reservation windows open.
- Accelerator mix: Specific GPU generations and memory footprints, plus multi-GPU topology and partitioning options.
- Interconnect choices: InfiniBand vs. Ethernet, cluster sizes, and bandwidth guarantees for distributed training.
- Thermals and density: Liquid cooling standards, rack densities, and PUE targets that affect workload tuning.
- Network egress and peering: Paths to major clouds and local carriers for hybrid setups.
Market context
Global demand for compute keeps climbing, and infrastructure is catching up. More capital is flowing to data centre build-outs, energy procurement, and grid integrations. For dev, ML, and platform teams, this means planning for regional placement, queue times, and multi-site failover-especially as Australian capacity comes online.
The funding timeline
This raise follows Firmus's A$330 million placement in September, which also included participation from Nvidia and a cornerstone investment by Ellerston Capital. Firmus and Nvidia did not provide additional comment on the new round, and Ellerston did not respond to a request for comment.
Actionable next steps for teams
- Map training and inference workloads that could move to Australia for latency, residency, or cost reasons.
- Prepare cluster requirements: GPU type, memory per GPU, interconnect needs, and expected job sizes.
- Align with sustainability goals: Track carbon intensity and renewable sourcing in vendor assessments.
- Set up a procurement watchlist for RFPs, reservations, and early access programs.
$1 = 1.5389 Australian dollars
If you're planning skills development around AI infrastructure and MLOps, see our curated programs by role at Complete AI Training.
Your membership also unlocks:
