For the first time in the recorded Gartner series, software is now the largest category of technology spending in Australia. It has overtaken IT services. The 2026 forecast puts software at close to A$60 billion out of a A$172.3 billion total, growing 13.6 percent in a year that grew total IT spend 8.9 percent. That is not a rounding shift. It is a category reversal.
The obvious question is what a CIO does with that fact on 7 July, three weeks into FY27, with a board budget approval calendar that mostly closes by late August. The less obvious question, and the one Australian executives are asking behind closed doors this month, is whether last year's budget shape is still defensible at all.
Four forces are moving at once. Software has moved past IT services as the largest line item. Gartner expects 75 percent of software spend to carry generative AI functionality by 2028. Inference token pricing has fallen roughly 80 percent between early 2025 and early 2026, with another 90 percent cost reduction forecast by 2030. And the procurement side is going the other way, with multi-year mega-deals like the DTA's VSA6 Microsoft agreement, which commenced 1 July, locking in pricing and capabilities across the public sector for five years.
None of these are surprises taken separately. Taken together, they mean the budget an Australian enterprise built in January no longer resembles the one it needs in November.
What actually changed in the Australian IT spending mix
Gartner's September 2025 forecast, updated with mid-year data during 2026, gives a clean picture. Software will exceed A$60 billion in Australia this year, growing 13.6 percent. IT services, which held the top slot for most of the last decade, will grow more slowly and drop to second position. Data centre systems will grow 22.5 percent to A$10.1 billion, with server spending up 30 percent to A$7.7 billion inside that total, driven almost entirely by AI-optimised infrastructure. Information security spend will pass A$7.5 billion. Public cloud services in Australia will reach A$33.6 billion, a 17.9 percent increase.
Inside public cloud, the sub-category story is where the FY27 planning tension sits. SaaS remains the largest cloud segment at A$16.4 billion, but its growth has slowed to 13.8 percent, in line with a maturing market. IaaS, at A$7.1 billion, is now the fastest growing category at 24.1 percent, because the AI workloads coming out of pilot need somewhere to run. PaaS, around A$10 billion, is growing 20.9 percent.
Two structural signals sit inside those numbers. The first is that Australian software procurement, historically dominated by seat-based licences and services wrapped around them, has flipped into a phase where the software itself is doing the work that services used to be sold to do. The second is that IaaS spend is accelerating not because storage or general compute needs have doubled, but because inference and model training now require dedicated GPU capacity that most Australian buyers did not procure in previous cycles.
Why the software category overtook IT services
Two things happened at once. Systems integrators and consulting firms lost some of their previously bundled discovery, integration and change work to software vendors shipping ready-configured functionality. That story is not unique to Australia. What is more distinctly Australian is the second thing. The public sector, financial services, mining and healthcare buyers who kept services revenue elevated for years have re-signed as software customers on multi-year subscription terms that pull revenue out of the services line and into the software line.
The VSA6 deal, negotiated by the Digital Transformation Agency and Microsoft and effective 1 July 2026, is the clearest live example. Every dollar an APS agency spent last cycle on integration partners to stitch Microsoft 365, Copilot, Dynamics 365 and Azure together into working ways of working is now, at least partly, a dollar sitting inside the software line as licensed capacity. The consulting spend has not evaporated. It has been converted into subscribed access to a stack that ships the assembly job as part of the product.
Private sector CFOs read this pattern too. When the largest single technology procurement in Australia this year is a licensing agreement rather than a systems integration engagement, the ratio between what your business spends on software and what it spends on services should probably not resemble the pre-2024 shape. Boards are asking that question of finance chiefs now, and finance chiefs are asking it of CIOs.
The inference pricing collapse changes what a good AI budget looks like
Inference cost per million tokens has fallen roughly 80 percent between early 2025 and early 2026 for comparable-quality models. GPT-4o class input pricing has dropped from around US$5.00 to US$2.50 per million tokens, with newer small models available at US$0.55. Gartner has forecast a further 90 percent reduction by 2030. Australian invoices track those movements almost immediately, because Azure AI Foundry, AWS Bedrock and Vertex AI price in USD and bill in AUD, so the local buyer sees the price cut when it lands.
The implication is not that Australian enterprises should shrink their AI budgets. The average enterprise AI budget has grown from around US$1.2 million a year in 2024 to US$7 million in 2026, roughly six times, at the same time the unit cost has fallen sharply. That combination tells you the total workload has grown by roughly thirty times in two years. If the workload keeps growing at anything close to that rate, price cuts do not fund themselves; the volume swamps them.
The FY27 planning implication is that unit price forecasting is now an insufficient basis for the AI line. What matters is the shape of the demand curve inside your own environment. How many tokens per employee per day at the current baseline. What the demand looks like after each new agentic workflow is deployed. Where the peaks fall. Whether the workload should sit on public inference or move to reserved capacity. Most Australian enterprises do not have this visibility because they never had to build it. They will now need to.
The Australian AI infrastructure question is no longer theoretical
Australia now has domestic infrastructure options for AI inference that did not exist twelve months ago. On 3 June 2026 Megaport announced A$458.9 million in new AI infrastructure contracts, roughly A$199 million in annual recurring revenue, and a fully underwritten A$827.3 million capital raise to fund a globally distributed AI inference cloud. The plan puts around A$350 million into a GPU pool anchored on NVIDIA silicon, deployed across a footprint of more than 1,100 connected data centres in 31 countries, sold to enterprise customers on both contracted and consumption terms.
DXN Limited, a smaller Perth-headquartered ASX operator, announced an A$8.8 million initial contract to deliver a 1.36 megawatt HPC modular data centre to a US-based neo-cloud, with direct-to-chip liquid cooling and GPU racks running up to 150 kilowatts each. That deal is small on its own. The signal underneath it is that Australian-built HPC modules are now being manufactured in Welshpool for export inside the AI infrastructure supply chain.
For an Australian CIO the practical implication is that the traditional two-option framing (build or buy) is now a three-option question. Build your own capacity is still one path. Buy hyperscaler consumption inside AWS, Azure or Google is still the second. The third, largely absent from FY26 plans, is procuring dedicated inference capacity from an Australian or Australian-connected provider on contracted terms that avoid FX exposure, sit inside Australian data sovereignty perimeters, and can be sized against your actual workload rather than a hyperscaler minimum.
The build versus buy versus consume decision no longer has an obvious default answer. That is exactly why it needs to appear in the FY27 plan as a line, not as a footnote.
Why services spend is not vanishing, just changing shape
Software overtaking services in the top-line data does not mean the services opportunity disappears. It means the services that survive are the ones software cannot ship. Integration around AI agents talking to legacy systems, data engineering to make company data actually usable by models, safety and evaluation work on agentic workflows, red team assessments against prompt injection, non-human identity governance, and the change management on top of any of it, all remain services problems the software line cannot handle alone.
The demand mix inside Australian services firms confirms the shift. The workload has moved from long implementation projects toward shorter, deeper engagements around specific AI deployment risks. AvePoint's 2026 study reported that 88.4 percent of surveyed organisations had experienced at least one AI agent related breach in the twelve months to mid-2026. That is not a market for another Microsoft 365 rollout. It is a market for skilled human judgement on containment, observability and governance.
CFOs planning FY27 will be tempted to cut services spend because the top-line numbers give them cover to do so. Doing that indiscriminately will cost more than it saves. The right move is to compress services spend on legacy delivery work while preserving, or increasing, the specialist services layer around AI deployment and cyber resilience.
Where cyber spend belongs in this budget
Information security spend is forecast to pass A$7.5 billion in Australia in 2026. That number is only defensible if it maps to the specific risk shape agentic AI is producing. ASIC's 26-092MR open letter from 8 May 2026 made cyber resilience an explicit licensing obligation for AFS licensees, with a twelve-point action list and four governance expectations directed at boards. That letter is now being tabled at board risk committees across the financial sector.
Two categories of security investment do real work in FY27. The first is non-human identity, because every AI agent needs a distinct identity with least privilege access, ephemeral credentials, and a joiner-mover-leaver lifecycle. Delinea's 2026 identity report found 89 percent of Australian respondents facing at least one identity visibility gap, with 52 percent naming AI-related environments as their most persistent gap. That is a controllable problem with known technology, but it is not the problem the last-cycle IAM roadmap was scoped for.
The second is observability that logs every agent tool call, model input, model output, data access and downstream action. Without that, incident investigation is speculation. Both categories require capital and vendor selection now to be operational by the end of FY27.
What the FY27 budget should actually reflect
The budget shape most Australian enterprises submitted for FY26 held roughly this ratio: services and integration the largest single line, software next, cloud third, hardware and data centre fourth, security fifth, and a small AI experimentation envelope carried in a corporate innovation cost centre. That shape assumes the pre-agentic world.
The FY27 shape needs to reverse the top two lines, absorb the AI experimentation envelope into an operational software and inference line that scales with actual workload, and lift the specialist services layer and information security line to reflect what agentic AI actually requires to run safely. Public cloud consumption should be modelled with a specific IaaS growth assumption at 20 percent or more, not blended into a whole-of-cloud number. Data centre and server capex should either appear as a real number or be explicitly ruled out with reasoning, because the buy path is now a genuine option for the first time in a decade.
Governance capacity is the last piece. BCG's much-quoted 10 percent algorithms, 20 percent technology and data, 70 percent people and processes framework holds up under Australian conditions. The 70 percent line covers process design, training, safety evaluation, change communication, and the accountability structures that let a board actually sign off on an autonomous agent going live. Most Australian enterprises have historically funded the 10 percent and the 20 percent and hoped the 70 percent would happen through the general operating budget. That has not worked.
The board conversation this month
Every board approving an FY27 budget between now and September will want three things stated in one page. First, the shape shift, in plain numbers. What percentage of technology spend is now going to software, cloud and AI-specific infrastructure, and how has that ratio moved. Second, the containment position on AI risk, mapped to the ACSC's agentic AI guidance and, for financial services entities, to ASIC 26-092MR. Third, the vendor concentration and lock-in position, with an explicit statement about how much of the plan sits inside a single vendor's stack and what a two-year exit would cost if a strategic reason to move ever emerged.
Boards will approve budgets that answer those three questions on evidence. They will push back on budgets that carry forward last year's category ratios without acknowledging the shifts underneath. A finance chief who defers those questions to FY28 is betting the environment stays still for another twelve months. Nothing in the current data suggests it will.
Frequently asked questions
What is Australia's IT spending forecast for 2026?
Gartner forecasts Australian IT spending to reach A$172.3 billion in 2026, an 8.9 percent increase on 2025. Software is now the largest category at close to A$60 billion, up 13.6 percent, having overtaken IT services for the first time. Public cloud services will reach A$33.6 billion, data centre systems A$10.1 billion, and information security more than A$7.5 billion.
Why has software overtaken IT services in Australia?
Because software vendors now ship ready-configured functionality that used to require systems integration work, and because large public sector and enterprise buyers have signed multi-year subscription agreements that convert what was formerly services revenue into software revenue. The DTA's VSA6 Microsoft agreement, which commenced 1 July 2026, is the clearest recent example of that conversion at scale.
How much are Australian enterprises spending on AI?
Global enterprise AI budgets have moved from around US$1.2 million per year in 2024 to US$7 million in 2026 on average. Australian enterprises track this trend, with 80 percent of government CIOs naming AI as a top budget priority and 85 percent expecting IT budgets to rise in 2026. Actual spend is distributed across software subscriptions, public cloud IaaS for inference workloads, dedicated GPU capacity, and services for deployment and governance.
How much has AI inference cost dropped in 2026?
Per-million-token inference pricing has fallen approximately 80 percent between early 2025 and early 2026 for comparable-quality models. GPT-4o class input pricing moved from around US$5.00 to US$2.50 per million tokens, with smaller efficient models available at around US$0.55. Gartner forecasts a further 90 percent reduction in inference costs by 2030.
Is it cheaper for an Australian enterprise to build or buy AI infrastructure in 2026?
The answer depends on workload shape and data sovereignty requirements. Buying hyperscaler consumption remains the default for variable and low-volume workloads. Building dedicated capacity, or contracting Australian-connected providers such as Megaport for reserved inference capacity, becomes competitive for high-volume, latency-sensitive, or sovereignty-sensitive workloads. The three-option framing (build, buy, or consume from an Australian provider) is new for FY27 and should appear in the plan explicitly.
What is Megaport building for AI?
Megaport announced on 3 June 2026 an A$827.3 million capital raise to build a globally distributed AI inference cloud, anchored by around A$350 million in NVIDIA GPU capacity, deployed across more than 1,100 connected data centres in 31 countries. New AI infrastructure contracts worth A$458.9 million in total contract value and A$199 million in annual recurring revenue were announced simultaneously.
Why is data centre spending growing so fast in Australia?
Gartner forecasts data centre systems spending to grow 22.5 percent to A$10.1 billion in Australia in 2026, with server spending inside that up 30 percent to A$7.7 billion. The driver is AI-optimised infrastructure, meaning liquid-cooled, high-density GPU racks provisioned specifically for model training and inference workloads, which existing data centre estates were not built for.
What percentage of software spend will include GenAI by 2028?
Gartner forecasts that by 2028, 75 percent of Australian software spend will be on software with generative AI functionality embedded. That is a category-wide expectation, not a niche feature, and shapes how buyers should evaluate current renewal and procurement decisions.
Should enterprises lock in AI pricing with hyperscalers?
Multi-year hyperscaler AI pricing agreements deliver cost certainty and can protect budget planning against volatility, but they also embed vendor concentration risk and reduce optionality if pricing continues to fall or if better-fit providers emerge. The right position sits between locking a base workload on committed pricing and keeping incremental workload flexible enough to reallocate. The VSA6 Microsoft deal for the APS is a large-scale reference point for how the trade-off looks when it is negotiated at scale.
What is the risk of vendor concentration for enterprise AI?
Deep integration of identity, data, collaboration and AI services with one vendor increases switching costs, weakens future negotiating position, and creates a single control-plane surface for operational and cyber risk. Boards approving FY27 budgets should require an explicit statement of concentration risk, including what a two-year exit from the primary vendor would cost, before signing.
How do you write an FY27 AI budget?
Anchor the plan on a workload projection rather than a unit-price forecast, because unit prices are falling but workload volumes are growing faster. Split the plan across software subscriptions, public cloud IaaS, dedicated inference capacity, specialist services for deployment and governance, and non-human identity plus observability inside the security line. Preserve the BCG 10/20/70 ratio between algorithms, technology and data, and people and processes. Map every line to a specific commercial outcome and an owner who is accountable for delivering that outcome inside the year.
What this means for Wai's clients
Australian technology leaders reading this in July have a short window to fold these shifts into an FY27 budget the board will approve on evidence. The organisations moving fastest are treating the software category reversal, the inference cost curve, and the domestic AI infrastructure question as connected decisions rather than three separate ones. That is the frame Wai works from with the technology, SaaS, enterprise and public sector clients we support through ARC, our authority and AI visibility infrastructure. It is the same frame that will let a CIO stand in front of a board next month with a plan that reflects the market they are actually operating in.