Skip to main content

Compute

Compute refers to the processing power, particularly GPUs and supercomputers, needed to train and run AI and other systems. Australia's two largest public systems, Gadi and Setonix, deliver around 80 petaflops of peak performance, while a rise in new onshore GPU clusters and AI factories from commercial operators and universities is increasing capacity. Demand for compute time sits at around 3 times supply.

Total Systems

19

tracked clusters
Total Accelerators

11,181

GPUs / accelerators up
Sovereign Compute

74.2 PF

AU-controlled PFLOPS up
Offshore (in AU)

86.0 PF

commercial, offshore-owned stable

Systems by Peak PFLOPS

Sorted by peak performance. Hyperscaler counts are estimated — fleet sizes are not publicly disclosed.

AU-controlled Non-AU controlled
What is a PFLOP? A petaFLOP (PFLOP) is 10¹⁵ floating-point operations per second — the standard yardstick for supercomputer performance. One NVIDIA H100 GPU delivers ~0.034 PFLOPS of double-precision compute. Training a GPT-4-class model requires roughly 25,000 A100-equivalents running for 90 days — about 800 PFLOPS sustained. Australia's largest sovereign cluster (Pawsey Setonix) reaches 50 PFLOPS peak.

Accelerator Count by System

Number of GPUs / accelerators per cluster

Compute Context

Australia's sovereign compute is dominated by two public research systems: NCI Gadi (ANU/Canberra) and Pawsey Setonix (Perth). Both are accessible to Australian researchers via competitive merit allocation.

Commercial GPU cloud from hyperscalers (AWS, Azure, GCP) is available in Australian regions but remains offshore-owned and subject to export controls. Accelerator counts for these are estimated — the providers do not publish regional fleet sizes.

Australia has no sovereign frontier-scale GPU cluster (100k+ H100-class GPUs) needed to train large foundation models domestically.

Key gap: No Australian sovereign cluster exceeds 4,000 H100-class GPUs. The US NAIRR has 50,000+. Training a GPT-4-class model requires ~25,000 A100s for ~90 days.

All Compute Systems

SystemOperatorOwnershipAccelerator# AcceleratorsPeak PFLOPSStatusAccess
Pawsey SetonixPawsey Supercomputing Research CentreAU GovernmentAMD EPYC + AMD Instinct MI250X GPUs75750OperationalRestricted
AWS Sydney GPU Cloud (ap-southeast-2)Amazon Web ServicesOS PrivateNVIDIA A100 40GB / H100 SXM (P4d + P5 fleet)307240OperationalCommercial
Azure Australia East GPU CloudMicrosoft AzureOS PrivateNVIDIA A100 80GB / H100 (ND A100 v4 + ND H100 v5)204826OperationalCommercial
Google Cloud Australia (australia-southeast1)Google CloudOS PrivateNVIDIA A100 40GB / H100 (A2 + A3 series)153620OperationalCommercial
NCI GadiNational Computational Infrastructure (ANU)AU GovernmentIntel Xeon Cascade Lake + NVIDIA V100 GPUs24729.88OperationalGovernment Only
National AI Centre Compute PlatformNational AI Centre / CSIROAU GovernmentNVIDIA H100 SXM5 80GB5126.55OperationalRestricted
NCI Nirin GPU ClusterNational Computational Infrastructure (ANU)AU GovernmentNVIDIA A100 SXM4 80GB2002.56OperationalGovernment Only
CSIRO PetrichorCSIRO Data61AU GovernmentNVIDIA A100 80GB GPUs1281.64OperationalRestricted
UQ BunyaUniversity of QueenslandAU GovernmentAMD EPYC + NVIDIA A100 GPUs961.23OperationalRestricted
Pawsey MagnusPawsey Supercomputing Research CentreAU GovernmentIntel Xeon Phi Knights Landing01.07DecommissionedRestricted
DST AI Compute PlatformDefence Science and Technology Group (DST)AU GovernmentNVIDIA A100 / H100 GPUs640.82OperationalGovernment Only
Spartan HPCUniversity of Melbourne / Research Computing ServicesAU GovernmentIntel Xeon + NVIDIA A100 / V100 GPUs560.35OperationalRestricted
University of Adelaide PhoenixUniversity of AdelaideAU GovernmentIntel Xeon + NVIDIA A100 GPUs480.31OperationalRestricted
UNSW KatanaUniversity of New South WalesAU GovernmentIntel Xeon + NVIDIA A100 / V100 / K80 GPUs680.28OperationalRestricted
Monash MonARCHMonash UniversityAU GovernmentIntel Xeon + NVIDIA Tesla V100 GPUs440.2OperationalRestricted
QUT HPCQueensland University of TechnologyAU GovernmentIntel Xeon + NVIDIA A100 GPUs320.16OperationalRestricted
RMIT PunimRMIT UniversityAU GovernmentIntel Xeon + NVIDIA Tesla V100 GPUs200.1OperationalRestricted
Griffith Gowonda HPCGriffith UniversityAU GovernmentIntel Xeon + NVIDIA T4 GPUs160.08OperationalRestricted
Deakin DRAGO HPCDeakin UniversityAU GovernmentIntel Xeon + NVIDIA V100 GPUs120.06OperationalRestricted