The zero-marginal-cost era just got a lot more complicated.
Canva built one of the rarest profiles in tech: a high-growth startup that actually made money. Then it tried to become an AI-powered enterprise platform, and the economics hit a wall.
The company cut its expected revenue growth rate by a third — down to 20% — after the cost of delivering AI features came in far higher than anticipated. CEO and co-founder Melanie Perkins told Fortune that user demand for the new AI tools 'significantly exceeded' expectations, which sounds like a win until you read the next line.
'This validated the demand, but also showed us we needed to reduce the cost of completing an AI task to support a broad rollout,' Perkins said. Rather than push a product whose underlying economics weren't ready, Canva chose to slow the rollout, rebuild its architecture and reduce unit costs.
That is a disciplined call. It is also a costly one.
The inference problem is industry-wide
Derek Hernandez, Pitchbook's senior research analyst covering the intersection of SaaS and AI, put it plainly: 'AI is making SaaS no longer a zero marginal cost solution, which has really been what I would call a lot of software's secret sauce up until now.'
Inference costs — the recurring expense of processing every AI request a user makes — don't behave like traditional software costs. They scale with usage. The more customers love the product, the more it costs to serve them. That inverts the model that made SaaS so attractive to investors for the better part of two decades.
Canva's public-market parallel, Figma, is living the same story from a different angle. Figma's free-cash-flow margin fell to 14% in the second quarter from 27% in the first. The company is forecasting third-quarter revenue growth of 36%, a deceleration from 48% in the June quarter.
Hernandez told Fortune that Canva and Figma are the 'biggest signals' that AI is breaking SaaS's traditional model. 'Canva and Figma both hit the same wall about five days apart, but they cited it in different places,' he said.
Perkins noted that Canva has reduced the cost per AI task by nearly 90% since launching Canva AI 2.0 in April — an agentic upgrade to the platform. The catch: users of Canva AI are now creating three times as many designs as in the previous version, which means total inference spend keeps climbing even as per-unit costs fall.
IPO math enters the equation
The timing matters. Fortune reported last year that an employee share sale valued Canva at $42 billion, with experts pointing to a potential 2026 IPO. Hernandez now told Fortune that Canva might be targeting next year instead.
'I'm sure they're trying to protect their profitability, especially if they want to go to public investors,' Hernandez said.
That is the right instinct. Public markets will price Canva on margins and predictability, not on the promise that AI costs will eventually come down. By tapping the brakes now, management is signaling that it understands what investors will actually pay for.
The bottom line for free enterprise
Canva's situation is a clean illustration of a principle the market enforces without sentiment: technology does not repeal cost structures, it relocates them. The SaaS model promised near-infinite scalability at near-zero marginal cost. Generative AI moved the cost from storage and bandwidth to compute and inference — and the bill is real.
Companies that acknowledge the economics early and rebuild their architecture accordingly will be the ones standing when the IPO window opens. The ones that paper over the costs with growth narratives will find public investors far less forgiving. Capital rewards clear rules — and right now, the rule is that AI has to pay for itself.
