Canva Voluntarily Slows Growth, Figma Absorbs Inference Costs: AI App Gross Margin Structures Face First Financial Test
Claire Weston
Canva cut its full-year revenue growth forecast to 20% after AI inference costs spiraled; Figma's stock fell roughly 15% in a single day post-earnings — both reports expose the same structural problem: the more popular an AI feature gets, the deeper it eats into gross margins.
The AI features are popular — so what's the problem?
The two leading design-software companies, Canva and Figma, just delivered the same uncomfortable message: AI feature demand far exceeded expectations, but so did the cost of serving every single inference call.
This means → "Popular" does not automatically mean "profitable." The more users engage, the more money the company burns — unit economics (whether revenue per user covers cost per user) actually deteriorate.
In plain terms = before AI, serving free users cost almost nothing. After AI, every user action burns inference compute — the free lunch turned into a real bill.
Why did Canva hit the brakes voluntarily?
Canva COO Cliff Obrecht told colleagues directly: before AI, the cost of serving free users was "very low." After AI launched, "those costs rose dramatically, unit economics fundamentally changed, making it far more important to bring AI costs down."
The company made a rare call: it deliberately paused a planned AI feature rollout designed to drive paid-subscription growth, pulling its full-year revenue growth forecast down to 20%.
This means → Canva chose "stop the bleeding first, grow later" — accepting slower short-term growth rather than letting inference costs hollow out margins.
Who picks up Figma's tab?
Figma CFO Praveer Melwani stated on the investor call: "We do not charge customers for products still in beta. We absorb the inference costs ourselves, with no offsetting consumer revenue."
Figma guided Q3 revenue growth down from 48% in the June quarter to 36%; multiple AI tools remain in beta with no clear path to monetization.
In plain terms = Figma is subsidizing users to try its AI for free — the product gets applause, but the company foots the entire bill. A ~15% single-day stock drop was the market's verdict.
Can in-house models ride to the rescue?
Both companies view in-house AI models as the structural fix — Canva says its internal models generate images and video faster and cheaper than top AI-lab models.
But those models were not ready in time for this rollout cycle. Canva still relied on costlier external models at the critical moment; Figma has only begun blending internal models with frontier ones.
This reflects a deeper timing mismatch: building proprietary models takes years; capital markets demand earnings growth by the quarter.
What does this mean for the broader AI-application sector?
The two earnings reports set a crucial reference point: product-level appeal for AI apps is now validated, but the path to converting that appeal into a sustainable profit model remains unclear.
This means → the real cost of AI transformation is not just R&D spend — it is embedded, more fundamentally, in the inference fee behind every single user call. Until monetization models mature, that cost keeps eroding gross margins.
The next proof point is already on the clock: whether in-house models can be deployed before the next product cycle will determine if these companies can break out of the cost trap.
Content is for reference only, not financial advice.