Capability 1.1.2.2.1
Public Sector & Public Interest AI Training Infrastructure
Definition
High Performance Computing (HPC) systems optimised for AI training, owned and operated by government, universities or research agencies. These systems may combine traditional Central Processing Unit (CPU) based HPC with AI accelerator enhanced architecture.
Examples
- National laboratories, university HPC centres, scientific agencies' supercomputers (e.g. CSIRO's Virga).
Indicators
- Number and scale of publicly owned AI training-capable HPC clusters
- Availability and cluster sizes of accelerators
- Number of H100-equivalent accelerators
- Top500 /Top100 world rankings
- Compute hours accessible to public interest research
National AI Plan
PLAN COMMITMENT
Limited commitment
See general comment at 1.1.2.1.1 (Cloud Training Compute Infrastructure as a Service (public cloud)). Given the high public interest in fostering this capability, clarification and review of the government's current commitment is recommended.
Stocktake
Institutional research computing capabilities in Australia: 2024. Kitaeff et al. arXiv preprint. 22 September 2025. https://arxiv.org/abs/2509.17351
National Research Infrastructure Roadmap 2026._ Department of Education_. Forthcoming. Accessed April 2026. https://www.education.gov.au/national-research-infrastructure/consultations/help-shape-future-research-infrastructure-planning
National Computational Merit Allocation Scheme 2026. National Computational Infrastructure. https://my.nci.org.au/mancini/ncmas/2026/
Top500 List. June 2025. https://top500.org/
Made in Australia: Our AI opportunity. Australian Academy of Technological Sciences & Engineering. 22 August 2025. https://www.atse.org.au/what-we-do/strategic-advice/made-in-australia-our-ai-opportunity/
Is the Stocktake missing evidence?
Share additional information for potential future Assessments