Australian AI &
Machine Learning Software

The Australian AI software market in 2026 is effectively a hyperscaler market. Microsoft committed AUD 5 billion through 2026 for GPU capacity across Sydney and Melbourne, deploying 20,000 H100-class GPUs to support Azure OpenAI workloads.[Mordor] AWS added EC2 P5 (H100) instances in Sydney. Google Cloud launched Vertex AI in Melbourne in 2024 for low-latency model inference.[Mordor] Across global AI/ML operationalisation software, the top five vendors — led by Microsoft Azure, AWS, and Google Cloud — hold 37% of the market, with AWS SageMaker and Google Vertex AI cited as the dominant managed platform choices.[Mordor]

Local Australian AI software companies — including Complexica, Flamingo AI, and Faethm — do not appear in any competitive landscape, funding data, or contract disclosure in the research available for 2025–2026. This absence is itself meaningful: it suggests these firms are operating at a scale and in verticals that do not surface in market-level analysis, or have been absorbed into larger platforms. NEXTDC, Australia's largest data centre colocation provider, posted 34% revenue growth in 2024 driven by AI demand — but NEXTDC is infrastructure, not software.[Mordor] The colocation layer is booming precisely because hyperscalers are leasing space from local operators while controlling the software stack above it.

The mechanism behind hyperscaler dominance is straightforward: GPU supply constraints — 12 to 18 month lead times and 15–25% cost increases — mean only entities with the balance sheet to pre-order at scale can build the compute required to credibly offer enterprise AI services.[Mordor] Australian enterprises using AI are almost universally running it on hyperscaler infrastructure. That dependency makes switching to a local software vendor structurally difficult — the data, the models, and the compute are already inside a single hyperscaler environment.

The largest identified Australian-linked AI funding round of the 2023–2026 period was Firmus Technologies' AU$330 million raise in September 2025, valuing the company at AU$1.85 billion (approximately USD 1.2 billion).[TechCrunch] Firmus, originally a Singapore-based cooling technology provider for Bitcoin mining, is building an energy-efficient AI data centre — an 'AI factory' — in Tasmania, with Nvidia among the investors. Nscale, a data centre business spun out of Australian firm Arkon Energy in 2023, raised USD 433 million in a SAFE round in October 2025 with Nvidia participation, though Nscale's operations are in the UK and Norway.[TechCrunch]

Nvidia's repeated appearance across large AI infrastructure rounds — also including OpenAI's USD 6.6 billion raise, Cohere's USD 500 million Series D, and Together AI's USD 305 million Series B — reveals a consistent strategic logic: Nvidia is investing in the entities that will consume the most GPUs.[TechCrunch] This is not venture philanthropy. It is demand creation for Nvidia's core hardware business. For Australian observers, the implication is that capital flowing into local AI infrastructure is partly upstream supply-chain investment, not a pure bet on Australian software demand.

No Australian-headquartered AI software company has disclosed a funding round above USD 100 million in the period covered. No acquisitions of Australian AI software firms by global players appear in the available research. This data gap is itself informative: either Australian AI software companies are bootstrapped, subscale, or acquired quietly — or the venture and private equity community has not yet identified an Australian AI software firm worth a headline round. The absence of named local software investment is the most important fact the capital flow data reveals.

Professional services leads sector adoption at 79%, followed by financial services and large enterprises broadly at 82% — though the enterprise figure spans all large organisations, not a single vertical.[OpenAI Australia] The public sector trails on headline adoption (70% of workers reporting daily AI use as of February 2026) but has moved fastest in policy terms: the Australian Public Service launched its AI Plan in November 2025, Services Australia published an Automation and AI Strategy covering 2025–27 in May 2025, and the Digital Transformation Agency released responsible AI use guidelines in September 2024.[APS AI Plan][Services Australia]

The 70% public sector figure comes from an Appian survey of 500 workers in February 2026 — a Tier 3 source with commercial incentive to show high AI adoption.[Appian] It should be read as directionally indicative, not as a precise measurement. What is not in dispute is the policy momentum: Canberra has moved from no formal AI governance to a National AI Plan (December 2025), an APS AI Plan, and the first mandatory regulatory instrument (the Privacy Act amendments) within roughly 18 months. That pace of policy movement typically precedes a wave of procurement activity.

No named enterprise buyer — Commonwealth Bank, Telstra, Westpac, or any Australian government agency — has publicly disclosed an AI software contract with a vendor name and dollar value in the research available. Commonwealth Bank published an AI adoption report in February 2026 and launched an AI video series in December 2025, but these are communications products, not contract disclosures.[CBA] The result is a paradox: adoption is high, budgets are moving, but the market for AI software in Australia is almost entirely opaque from the outside. This opacity benefits incumbents — primarily hyperscalers — whose enterprise relationships are already in place.

The Privacy and Other Legislation Amendment Act 2024 received royal assent on December 10, 2024, and its automated decision-making (ADM) transparency provisions take effect on December 10, 2026.[OAIC] Any AI software vendor whose product processes personal information to make or substantially support decisions that significantly affect individuals — loan approvals, HR screening, insurance pricing, customer service routing — must disclose in their privacy policy the types of personal information used, the kinds of fully automated decisions made, and the kinds of decisions where AI substantially assists. Maximum penalties for serious breaches rose from AUD 2.2 million to AUD 50 million — a 23-times increase — and the OAIC gained new powers to issue infringement notices up to AUD 333,000 per breach without court involvement.[OAIC]

The OAIC began a compliance sweep of privacy policies in 2026 and published guidance in October 2024 emphasising privacy by design, AI data accuracy (explicitly including hallucination risk), and consent for sensitive data used in AI training.[OAIC] For AI vendors, this means the compliance burden is not theoretical — it is eight months away from this report's publication date and the regulator is already examining the market. The practical effect is that vendors without clear ADM disclosure frameworks embedded in their privacy policies are running a material legal risk in their largest contracts.

Voluntary instruments — the Department of Industry's Voluntary AI Safety Standard (August 2024) and the Digital Transformation Agency's responsible AI policy for government (September 2024) — establish expectations without enforcement teeth.[Industry.gov.au] A risk-based legislative framework for high-risk AI applications is under active consideration by the government but has not been enacted as of April 2026. The National AI Plan published in December 2025 signals intent to build a 'competitive, productive, and resilient' AI-enabled economy but does not add enforcement mechanisms.[Industry.gov.au] The regulatory picture therefore has one hard constraint (December 2026 privacy deadline) and a growing body of soft guidance that may harden into legislation before 2028.

The most structurally distorting feature of the Australian AI software market is that supplier power and competitive rivalry are inverse to what a healthy software market would look like. GPU supply is constrained — 12 to 18 month lead times and 15–25% cost inflation — meaning only hyperscalers can reliably deliver enterprise-grade AI compute.[Mordor] That hardware scarcity translates directly into software platform lock-in: enterprises running Azure OpenAI, AWS SageMaker, or Google Vertex AI are not simply choosing a software vendor. They are choosing a compute environment, a data residency location, a compliance framework, and a pricing structure — all in one decision.

Buyer power is theoretically high — 82% enterprise adoption suggests a mature, informed buying community — but in practice, switching costs suppress it.[OpenAI Australia] Once an enterprise has fine-tuned models, built data pipelines, and signed enterprise agreements with a hyperscaler, moving to a competing platform or a local alternative requires re-training models, re-engineering integrations, and renegotiating pricing. For large Australian banks and insurers, 82% of which require sovereign cloud for AI workloads, the constraint narrows further — only hyperscalers with Australian data centre regions qualify.[OpenAI Australia] New entrants face the same GPU scarcity problem as incumbents, only without the existing enterprise relationships to justify large compute pre-orders.

The base case rests on a specific constraint: Australia needs 312,000 additional technology workers by 2030, and AI/ML skills have an estimated obsolescence cycle of 2.5 years.[JSA] This means that even in a linear growth scenario, the market's ability to absorb and deploy AI software is paced by talent availability, not by capital or technology. The government's AUD 600 billion productivity target from AI — including AUD 115 billion from generative AI — is a ceiling defined by execution capacity, not ambition.[Industry.gov.au]

The upside scenario requires two things to happen simultaneously: a meaningful expansion in AI-skilled labour (through graduate pipelines, offshore models, or skills-retraining programs) and a sovereign AI funding commitment from government large enough to catalyse the procurement wave that policy momentum has been building toward. The December 2026 privacy deadline could paradoxically accelerate this — compliance requirements force enterprises to formalise their AI usage, which creates auditable procurement records and, eventually, a disclosed software market.[OAIC]

The downside scenario is not a crash — it is displacement. If AI software spend largely replaces existing SaaS and automation budgets rather than adding new spend, growth rates flatline without anyone declaring a failure. Roger Montgomery analysis cited in the research highlights that current global AI valuations embed the assumption that end-customer spending of USD 1.3 to 3.1 trillion materialises — if it does not, Australian AI software multiples compress regardless of local adoption rates.[Montgomery] The signal to watch is not adoption percentage — it is whether the big four banks and Telstra report AI as additive spend or as budget reallocation in their next two annual reports.