For the past three years, the conversation about artificial intelligence in Australian boardrooms has been mostly about opportunity. In 2026, it has become a conversation about money as well. Two of the country’s best-known technology companies have shown, in their own disclosures, that AI can transform a business and strain it at the same time.

Canva’s candour

Canva is the clearest example. In her letter to shareholders for the June quarter, co-founder and chief executive Melanie Perkins said the company had been “relying too heavily on frontier models” and that the “average cost of serving an AI task was too high.” Canva cut its 2026 revenue growth forecast from 30% to 20%, slowed the rollout of its AI features and rebuilt its systems around its own models. It says the cost of running its AI has since fallen by about 90%. Its long-standing investors marked down their holdings in the following weeks.

This was not a story of AI failing. Customers wanted Canva’s AI tools; that was the problem. Each use cost money, and pricing had not kept pace. It is a lesson that applies far beyond software.

Atlassian’s trade-off

Atlassian has made a different trade. In March, chief executive Mike Cannon-Brookes announced the company would cut about 10% of its workforce, or around 1,600 people, “to self-fund further investment in AI and enterprise sales, while strengthening our financial profile.” Atlassian has since reported a return to a small GAAP operating profit for the 2026 financial year.

Whatever one thinks of the decision, it makes the economics explicit: money for AI has to come from somewhere, and at Atlassian a significant share of it came from payroll.

Banks are spending, too

The pattern extends to the old economy. Commonwealth Bank kept its annual investment spend at $2.4 billion and said it put more than $1 billion during the 2026 financial year into protecting customers from fraud, scams, cyber threats and financial crime, including AI-powered defences. Banks see AI as both a tool and a threat, because criminals are using it as well.

What boards should be asking

In our view, the lesson from 2026 is that AI should be treated like any other major capital program: with clear cost lines, measurable returns and honest reporting. Directors do not need to understand how large language models work. They do need to ask some basic questions:

  • What does each use cost? If customers or staff use an AI feature more, does the cost rise faster than the revenue or savings it produces?
  • Who controls the supply? Reliance on a small number of external model providers is a concentration risk, both for price and for continuity.
  • What are we giving up? If AI spending is funded by cuts elsewhere, the board should understand what capability is being lost.
  • How are we reporting it? Shareholders are entitled to know when AI costs materially affect margins or forecasts, as Canva’s investors learned.

There is also a governance point. Directors’ duties of care and diligence apply to AI decisions in the same way as any other. A board that approves a large AI program without understanding its economics, or that lets management make public claims about AI capability that cannot be supported, is exposed.

Not a reason to stop

None of this is an argument against investing in AI. Canva’s response, to rebuild rather than retreat, may well prove to be the right one, and Atlassian’s results suggest its bet is paying for itself so far. The point is that the early phase of AI enthusiasm, when experiments were cheap and benefits were assumed, is over.

Australian companies that treat AI as a cost to be managed, not just a story to be told, will be better placed than those that discover their AI bill the way Canva did: after customers had already embraced it. That discipline is now part of good governance.