Canva, a $42 billion design software company, just cut its 2026 revenue growth forecast from 30% down to 20%. The reason wasn’t slowing demand- demand for its new AI features “significantly exceeded” expectations, according to CEO Melanie Perkins. The problem was that Canva AI costs were too high to serve that demand profitably, and rather than eat the loss, the company chose to slow down.
What Actually Happened
In April 2026, Canva launched Canva AI 2.0 at its Canva Create conference in Los Angeles, the centerpiece of a two-year platform rebuild focused on conversational design- you describe what you want, and Canva assigns the right models and tools to build it. Users loved it. Too much, in a sense: the average cost of serving a single AI task turned out far higher than Canva had planned for, and its own first-party models weren’t ready in time to absorb the load. By Canva’s Q2 2026 investor update, revenue had reached $921.9 million, 25.2% year-over-year growth- solid by most standards, but short of the company’s own guidance, and enough to trigger a rare downgrade.
Why Rising Canva AI Costs Break a Business Model That Used to Just Work
Traditional SaaS economics run on a simple assumption: once you’ve built the software, serving one more user costs almost nothing. That assumption is what let software companies scale revenue far faster than costs for two decades. Generative AI features break that assumption directly- every AI request costs real, ongoing money to run, no matter how many times it’s already been served to someone else. Canva isn’t alone in feeling this. Design tool Figma has reported similar margin pressure from AI features, and Atlassian has reportedly capped how much its own employees can spend on AI tools internally. The zero-marginal-cost logic that built modern software’s profitability is running into a genuine wall.
How Canva Actually Brought Costs Down
Speaking at a Stripe event in Sydney, Canva co-founder Cliff Obrecht said the company has cut the cost of serving an AI task by roughly 90% since April. The approach had three parts: routing more requests to cheaper, purpose-built models instead of expensive general-purpose frontier models by default; building out a 140-person in-house research team to develop first-party models specifically tuned for Canva’s own workloads; and teaching the platform to adapt existing designs from its own library where possible, rather than generating something from scratch every single time. None of that required abandoning AI features- it required treating cost per task as a metric to actively engineer down, not an afterthought.
This Isn’t Just a Story of Canva AI costs
Rising Canva AI costs are a specific, disclosed data point for a pattern playing out much more broadly. Competitive pressure adds to the squeeze too: some users reportedly shifted design work over to ChatGPT after OpenAI’s image generation improved, meaning Canva was absorbing higher AI costs while simultaneously fighting to keep the users those costs were meant to serve. Notably, Canva has stayed profitable for nine consecutive years- a genuine rarity among high-growth software companies- which gave it room to deliberately slow down rather than either eat the losses or ship a broken experience. Most companies facing this exact cost curve don’t have that cushion.
What This Means for Your Business
- If you’re building AI features into a SaaS product, model cost per task before launch, not after demand arrives. Canva’s own team called out that its “pricing, consumption model and usage controls had not caught up with the outsized demand” it saw- build those controls in from day one.
- If you’re buying AI-powered SaaS tools, rising usage-based costs and sudden credit caps aren’t random- they’re vendors passing along exactly this same inference-cost pressure. Our guide to the shift toward usage-based SaaS pricing covers what to watch for as more vendors follow this pattern.
- Don’t assume AI feature usage will match your projections. Canva’s problem wasn’t that people avoided its AI tools- it was that they used them more than anyone modeled for.
Three Cost Levers a Smaller Team Can Copy
Canva paired model routing with a 140-person in-house research team, which most companies do not have. The underlying levers still work at any size:
- Route by difficulty. Send simple, repetitive requests to a smaller, cheaper model and reserve the most expensive model for tasks that truly need it. Review a sample of routed outputs each week to confirm quality has not slipped.
- Reuse before generating. Canva adapts existing designs from its own library where possible. The smaller-scale equivalent is caching common answers, templating frequent outputs, and starting from a stored example instead of generating from scratch.
- Meter and cap. Canva’s own team said its pricing, consumption model, and usage controls had not caught up with demand. Even a basic monthly allowance per user and an alert when spend crosses a threshold prevents an unpleasant surprise.
Questions to Ask Your AI-Powered Software Vendors
- How is AI usage metered, and what happens at the limit: throttling, overage charges, or a forced upgrade?
- Can we set spend caps or alerts on our account?
- How much notice do we get before credit allowances or prices change?
- Are AI features included in the base price or billed separately?
- What data leaves our account when AI features run?
Common Questions About Canva’s AI Costs
Why would Canva cut its forecast when demand was strong?
Because serving that demand cost more than its pricing could support. Rather than absorb the loss, the company slowed growth while it brought the cost per task down.
Did Canva stop offering AI features?
No. According to the reporting above, it kept them and cut the cost of serving an AI task by roughly 90% since April.
Does this matter if I only buy software?
Yes, indirectly. Usage-based charges, credit allowances, and sudden caps are one way vendors pass AI costs on to customers, so it is worth knowing what your plans meter.
Canva AI costs-Final Talk
A $42 billion, nine-years-profitable software company had to publicly slow its own growth because Canva AI costs outran its pricing model- and then cut those costs 90% once it treated the problem as an engineering priority instead of a surprise. Fortune’s original reporting has the fuller picture if you’re watching this space. The lesson scales down: any business adding AI features, or paying for someone else’s, is now operating inside the same unit-economics problem Canva just solved in public.