National Computational Merit Allocation Scheme (NCMAS)

The National Computational Merit Allocation Scheme (NCMAS) provides access to Australia’s national high‑performance computing (HPC) facilities through competitive peer review grounded in research and computational merit. Within this framework, the Scientific Advisory Committee (SAC) has an overriding responsibility: to allocate resources at levels that make the proposed meritorious research feasible, rather than dilute allocations across more projects at the cost of viability.

The 2026 round occurred under an unprecedented surge in demand. Across NCI and Pawsey, the call attracted roughly 245 proposals, and aggregate demand substantially exceeded the NCMAS allocation envelope. The pressure was especially acute in areas of rapidly escalating computational intensity—including large‑scale simulation, data‑driven discovery, and AI‑enabled workflows.

This round made visible a structural trend that has developed over several years: requests for HPC resources are growing far faster than Australia’s publicly funded national compute capacity. Between the NCMAS 2023 and 2026 rounds, demand rose by ~68%, while the resource envelope expanded by only ~6%. Specifically with respect to the NCMAS 2026 round, the annual demand grew by ~24%, while the resource envelope shrank by 4%. This widening gap between demand for HPC and available resource envelope to NCMAS is the central context for the 2026 outcomes, which reflects an increasingly urgent national vulnerability: Australia’s publicly funded research HPC is not scaling with the computational and simulation requirements of modern science and engineering and is not fit for purpose to take advantage of the opportunities that HPC and simulations bring to Australian Science, Engineering and the Economy.

Evidence snapshot (NCMAS requests vs available capacity of NCMAS)

Note:   The NCMAS 2023 round was already oversubscribed by 1.70×, with NCI Gadi
leading with an oversubscription of 2.34×, which by the current NCMAS 2026 round has ballooned to an oversubscription of 3.56×.

Key changes since the NCMAS 2023 round:

• Total requests increased by 68% while total available capacity increased by 6% (oversubscription increased by 58%).
• Gadi requests increased by 62% vs Gadi capacity +6%.
• Setonix‑GPU requests increased by 123% while Setonix‑GPU capacity remained flat over the same period.
• By 2026, aggregate requested service units reached 2,242,817 kSU against an available envelope of 835,000 kSU (≈ 2.69x oversubscribed).

NOTE

This preamble is intended to provide context for the 2026 outcomes and to support clear communication with applicants and stakeholders. It does not replace the published NCMAS guidelines, Terms of Reference, or the outcome communication.

This mismatch has immediate consequences for a merit‑based process. As oversubscription intensifies, the SAC's ability to "fund the tail" diminishes. Allocation thresholds rise sharply; proposals competitive in prior years become unfundable; and partial allocations—already common across the last four NCMAS rounds—can push otherwise meritorious projects below the minimum level needed to remain feasible or internationally competitive. These outcomes are disruptive not only to research teams but also to national science programs and publicly funded grants that hinge on access to scalable HPC.

The SAC acknowledges community concerns regarding the 2026 outcomes—particularly perceptions of: fairness, transparency, disciplinary balance, and the concentration of resources. The committee emphasises, however, that its assessment and allocation process has not changed. It remains grounded in expert domain‑specific assessment of research and computational merit, informed—but not dictated — by facility technical advice. Yet in a year of severe oversubscription, even a consistent process may produce community‑level outcomes that appear abrupt, because proposals cluster tightly in score while the envelope is exhausted earlier.

Critically, constrained national capacity not only reduces the number of funded projects, but it also increases the likelihood that funded programs become delayed, de‑scoped, or rendered unfeasible and uncompetitive internationally. This risk is most acute in computationally and simulation-intensive national priority areas—including climate and environmental science, renewable energy and critical materials, transport, advanced manufacturing, medical and health research, and AI‑enabled research.

Against this backdrop, the SAC considers it essential to be candid: the 2026 outcomes are not the result of any procedural change or shift away from merit. They are the outcome of a structural shortfall in national HPC investment. Without sustained growth in HPC capacity—and the workforce and data infrastructure required to support it—future NCMAS rounds will remain volatile, with increasing risks to research feasibility and community confidence in the allocation process.

This structural challenge is not unique to NCMAS. The Australian Academy of Science has recently highlighted the national risk created by the absence of a long‑term HPC strategy and the urgent need for major investment, including at least one exascale‑class capability to sustain sovereign research capacity. The Academy’s February 2024 brief ("The future computing needs of the Australian science sector") provides detailed evidence of the scale and urgency of this issue and is available here:

https://www.science.org.au/sites/default/files/Publication/document/brief-1-hpcd-in-australia.pdf