Canva Cuts Revenue Forecast as AI Costs Run High

There is a particular kind of corporate announcement that arrives dressed as confidence but smells faintly of surprise. Canva's Q2 CY2026 update to shareholders was that kind of announcement.

Australia's most valuable privately held tech company — a design platform that turned Powerpoint-phobia into a $40 billion business — told shareholders this month that its revenue growth forecast for 2026 is being revised down, from 30% to 20%. The reason, as Canva cofounder and CEO Melanie Perkins put it in the shareholder update, was stark: the "average cost of serving an AI task was too high," according to Startup Daily.

That sentence — plain, blunt, and stripped of the usual tech-company rhetorical flourishes — is worth sitting with. Because it describes a problem that extends well past Canva's spreadsheets.

What actually happened

Canva is not struggling in any conventional sense. The company brought in $921.9 million in revenue in the quarter, per B&T — not a number that normally comes with an apology attached. But a miss is a miss, and in the high-growth tech world, the distance between your forecast and your reality tells its own story.

The Australian Financial Review reported that Canva attributed the shortfall directly to the rising cost of building products on top of frontier AI models — specifically noting that this cost surge caused significant product launch delays. The company had committed publicly to an ambitious AI rollout. The rollout is happening, but slower than planned, and more expensively than budgeted.

The Aussie Corporate noted that Canva's push into frontier AI has been described as a "rare revenue downgrade" — the kind of qualifier that signals this is not business as usual for a company that has spent years building an image of smooth, rapid, sustainable growth.

Rampart News flagged the specific mechanism: Canva has failed to roll out its full AI suite to all users, and the revenue miss followed directly from that gap. Features that were supposed to be live — and presumably priced or upsold — are not yet in users' hands. The cost of building them is real. The revenue from delivering them is not yet.

The geometry of the AI cost problem

Here is where Canva's particular situation connects to something broader.

When a tech company integrates a feature it builds itself — a new template library, a better color wheel, a revised collaboration tool — the cost curve is relatively knowable. Engineering hours, infrastructure, testing, rollout. You can model it. You can miss the model, but you're working with familiar variables.

Frontier AI is a different animal. The leading large language models and image generation systems that power the features users actually want — magic background removal, AI-generated copy, intelligent design suggestions — are expensive at inference time in ways that scale with usage in ways that are hard to predict before you've shipped. Every user request against a frontier model has a compute cost. Multiply that across tens of millions of Canva users, running those tasks at the frequency the product encourages, and you are no longer in a territory where historical cost modeling is particularly useful.

Perkins framed the downgrade as a deliberate trade-off — the company is choosing to absorb higher costs now to preserve competitive positioning later. That framing is not wrong, but it's also not complete. There's a difference between "we chose to spend more than we originally planned" and "the costs turned out to be higher than we understood they would be." Perkins' comment that the average cost of serving an AI task was too high suggests the latter was also true.

The AFR, which earlier reported that Canva made a significant public pivot toward AI positioning, put it with appropriate directness: the company has admitted it miscalculated how expensive the shift would be.

That is a candid thing for any company to say. It is also, if you are watching the broader tech landscape, a story being told in a dozen different earnings calls and investor letters right now — just usually with more euphemism.

What Canva's position actually is

It's worth being precise about what kind of problem this is, because not all AI cost problems are created equal.

Canva is not burning cash it doesn't have. A company posting nearly $922 million in a single quarter has operating leverage that most businesses would sell organs for. The downgrade is real, and a one-third reduction in projected growth rate is not trivial — but the baseline from which that growth is being measured is enormous.

The more interesting question is structural: can Canva recover the margin it's sacrificing on AI compute, and on what timeline?

There are two basic paths. One is that Canva successfully passes the cost of frontier AI features through to users — either by pricing AI-enhanced tiers more aggressively, or by monetizing the expanded capabilities in ways that justify the infrastructure spend. The other is that model costs fall fast enough — through competition between frontier AI providers, through Canva's own optimization work, or through improvements in the underlying model efficiency — that the math improves on its own.

Both paths are plausible. Neither is guaranteed. And the product launch delays are the real tell here: if Canva can't get its full AI suite to users, it can't test which features users will actually pay a premium for, which means it's spending heavily on something whose revenue ceiling it hasn't yet measured.

The Aussie Corporate called this a "rare revenue miss" — and that word "rare" is doing real work. Canva has historically been disciplined about its numbers. The miss registers as notable precisely because the company's track record made it unexpected.

The question Canva hasn't fully answered yet

There's a version of this story in which Canva's transparency — here's what happened, here's why, here's what we're choosing to do about it — is exactly what good corporate governance looks like. Revenue targets are revised. A hard problem is named plainly. Investors are updated with specific language rather than reassuring vagueness.

There's also a version in which a company that bet its brand on an AI pivot is now explaining, quietly, that the pivot costs more than the plan assumed, and that it doesn't yet know what users will pay for the output.

Both versions can be true simultaneously. They usually are.

What matters now is not the forecast reduction itself, but whether Canva can demonstrate — in the quarters ahead — that the delayed AI suite is worth what it cost to build. If the features ship and users adopt them and the revenue growth recovers, the 2026 miss will look like a speed bump. If the cost curve stays elevated and the product rollout stays slow, a one-third reduction in projected growth rate will start to look like a different kind of data point.

The company has answered the question of what AI costs. It hasn't yet answered the question of what AI is worth.

---

Carmen Rodriguez is Buzzrag's labor and workplace correspondent. She covers worker power, organizing, and the economic systems that shape how people work and earn.