The economics of artificial intelligence have become the quiet subplot of every technology earnings call, and this week it was Atlassian’s turn to reassure investors that the bill for all this cleverness will not swallow the company whole. Co-founder and co-chief executive Mike Cannon-Brookes told the market that the Sydney-founded software maker has its AI costs firmly in hand, framing the technology as the single best thing to happen to the business rather than a looming drag on its finances.
The message matters because the question of who actually pays for generative AI, and how much, is no longer academic. Every large language model query carries a real compute cost, and vendors that have baked AI features into products customers already pay a flat rate for are now discovering that usage can climb faster than revenue. For a company like Atlassian, whose Jira, Confluence and Trello products sit inside millions of workflows, even a small per-query cost multiplied across a vast user base can add up to a serious line item.
The news
According to Forbes Australia, Atlassian used its latest investor update to flag that rising usage of Rovo, its AI agent and search layer, will weigh on margins in the 2027 financial year. That is a notable admission from a company that has long prided itself on efficient, product-led growth, and it puts a number and a timeline on a cost that many software firms prefer to leave vague.
Yet Cannon-Brookes was at pains to play down the significance of that margin impact, reportedly describing it as barely registering on the company’s overall trajectory. The gist of his argument is that the incremental cost of running Rovo is manageable, that the price of inference is falling as models get more efficient, and that the productivity gains AI delivers to customers will more than justify whatever Atlassian spends to power it. In his telling, the company is taming the cost curve rather than being dragged along by it.
Rovo sits at the centre of this pitch. Launched as Atlassian’s answer to the wave of AI copilots and agents sweeping through enterprise software, it is designed to search across a company’s scattered tools, answer questions and increasingly take actions on a worker’s behalf. The commercial logic is that once teams rely on Rovo to navigate their own institutional knowledge, they become both stickier customers and candidates for higher-priced tiers, offsetting the compute the feature consumes.
Two ways to read it
There are two competing interpretations of Atlassian’s stance, and both are worth taking seriously. The optimistic reading is that Atlassian is simply being transparent about a cost that every AI-exposed software company carries, and that its willingness to name FY27 as the pressure point reflects confidence rather than concern. On this view, a company that can quantify its AI cost drag is a company that understands it, and the falling price of model inference means the margin hit may prove smaller than feared by the time it actually arrives.
The more sceptical reading is that reassurance from a chief executive is exactly what you would expect regardless of the underlying reality, and that flagging a margin impact two years out is a way of managing expectations before the numbers turn up in the accounts. Investors have grown wary of AI promises that arrive with heavy capital and operating costs attached, and the market has punished software names that let AI spending outrun the revenue it generates. A single upbeat quote does not settle that debate.
The contrast with Canva sharpens the point. The Sydney design platform, another jewel of the Australian technology scene, has been candid about the way AI features are inflating its own cost base, to the point of reshaping how it treats free users. Two large homegrown software companies are wrestling with the same structural problem in public, and the fact that one sounds relaxed while the other sounds stretched tells you the answer is far from settled across the industry.
What it means for Australia
For Australia, this is more than an accounting curiosity. Atlassian and Canva are the two companies most often held up as proof that Australia can build globally significant software businesses, and their financial health shapes everything from investor appetite for local startups to the career ambitions of a generation of engineers. If AI costs can be absorbed profitably by a company of Atlassian’s scale, that is an encouraging signal for the smaller Australian firms now racing to bolt generative features onto their own products.
It also feeds directly into the national conversation about sovereign AI capability and the infrastructure that underpins it. The cost of running AI is, at bottom, the cost of compute, and compute is precisely what a wave of local data-centre and AI-factory investment is trying to make cheaper and more abundant on Australian soil. Every dollar Atlassian shaves off its inference bill is a dollar that would otherwise flow to overseas cloud providers, which is why the economics of AI at the product level and the push for domestic infrastructure are really two halves of the same story.
There is a workforce angle too. Rovo and tools like it are being sold on the promise that they make knowledge workers meaningfully more productive, and Australian employers are watching closely to see whether that promise survives contact with real budgets. If AI features remain affordable to deploy at scale, adoption across the local white-collar economy is likely to accelerate. If costs prove stubborn, companies may ration access in ways that blunt the technology’s reach.
What’s next
The real test will arrive in the numbers rather than the narrative. Investors will be watching Atlassian’s coming results for evidence that Rovo usage is translating into upgraded contracts and durable revenue, and for the first concrete signs of the FY27 margin pressure Cannon-Brookes has flagged. The trajectory of model inference costs, which have been falling steadily but unpredictably, will do much to determine whether his confidence looks prescient or premature.
For now, Australia’s two software standouts offer a useful natural experiment in how to make AI pay. One is projecting calm and the other is visibly recalibrating, and the gap between them is a reminder that the hardest question in artificial intelligence right now is not what it can do, but what it costs to keep it running.
Sources: Forbes Australia
