Canva has lowered its annual revenue growth forecast to approximately 20 per cent, down from the 30 per cent target established at the beginning of the year, after the AI features it launched proved far costlier to operate than anticipated. The $42 billion Australian design platform's adjustment marks an unusual admission for a major technology company in the AI era: that the economics of running advanced AI tools can force even well-funded players to pump the brakes. The company is now racing to rebuild its infrastructure after what CEO Melanie Perkins described as an over-reliance on expensive third-party models.
The company's second-quarter revenue hit $921.9 million, climbing 25.2 per cent compared to the same period a year earlier, a figure that appears strong on its own but fell short of Canva's internal projections. Some AI capabilities are running as much as six months behind their planned release dates, according to reporting from The Information. Canva's premium AI offering, called AI Pass, carries a price tag of $100 per user per month on top of existing Pro or Business subscriptions. By October 2025, the platform had logged more than 26 million conversations through its ChatGPT application and claims to rank among the top ten domains that ChatGPT refers users to, demonstrating real distribution even as the same channel enables users to create designs without visiting Canva directly.
Co-founder and CEO Melanie Perkins told investors the company had leaned too heavily on frontier models while its own proprietary models weren't ready, and that pricing plus usage controls hadn't kept pace with customer demand. The company claims it achieved a 90 per cent reduction in the expense of serving a single AI task after spending three months rebuilding its AI infrastructure and shifting more processing to its own models. The report notes this is a private company briefing its own backers, meaning the public is seeing only the metrics and narrative Canva chose to share, with no visibility into AI Pass adoption rates, gross margins on AI usage, or how much of the revenue miss stems from the premium tier underperforming versus features simply arriving late.
The revenue shortfall and guidance cut point to a tension now facing every SaaS company rolling out generative AI: the gap between customer excitement and viable unit economics. Canva's rebuilding effort and claimed 90 per cent cost reduction, if sustainable as usage climbs, would represent a fundamental shift in the company's cost structure rather than a temporary fix, transforming what had been a money-losing proposition into something that could scale profitably. The move away from reliance on expensive external models toward proprietary inference infrastructure reflects a broader question for the industry: whether companies can afford to lease intelligence indefinitely or must build their own to survive. The report suggests Canva's experience serves as a concrete data point for every other software company staring at similar bills and wondering whether to slow down or double down on AI features that users want but that threaten to sink margins. If a category leader with billions in valuation is telling its investors that AI costs forced a strategic pause and infrastructure overhaul, smaller competitors with thinner margins and less runway face the same math with fewer options. The platform model matters too—companies that distribute through ChatGPT gain reach but risk training users to bypass them entirely, a trade-off Canva is navigating in real time as it counts both referrals and potential cannibalization from the same source.