The cloud software playbook just hit a wall, and it's made of GPU bills.
The Summary
- Canva cut its growth forecast by a third because AI features cost more to run than they generate in revenue, breaking the model that made SaaS profitable
- An analyst says AI is breaking the "secret sauce" of software: selling the same code to millions of users at near-zero marginal cost
- Figma faces the same economics, signaling this isn't a Canva problem but a category problem for design tools and likely every SaaS company racing to ship AI
The Signal
Canva was supposed to be the proof that you could grow fast and print money. The Australian design platform hit profitability while scaling to 220 million users. Then it went all-in on AI features, and the economics broke.
Here's what changed: traditional SaaS runs on code you write once and sell forever. The cloud bills scale linearly, but revenue scales exponentially. That gap is where software fortunes come from. AI features flip the script. Every user query hits an LLM. Every image generation burns GPU cycles. The costs scale with usage, not just infrastructure. More users means more compute, which means margin compression instead of margin expansion.
"AI is breaking SaaS's 'secret sauce': selling the same code to millions of users at near-zero marginal cost."
The growth forecast haircut tells you everything. Canva didn't suddenly lose product-market fit. It didn't lose users. It added features users want, and those features cost too much to deliver at the price point users will pay. The value is there. The unit economics aren't.
Figma's in the same boat, which means this isn't about execution. It's about the fundamental cost structure of AI-powered software versus the pricing models inherited from the pre-AI era. Design tools are canaries in the coal mine because they shipped generative features early and hard. Text-to-design. AI image editing. Smart layouts. All compute-heavy. All running on models that cost real money per inference.
Key dynamics at play:
- SaaS margin magic depends on near-zero marginal cost per user
- AI features have linear or worse cost scaling tied to usage
- Users expect AI features bundled into existing subscription prices
- No one's figured out how to charge enough to cover the compute without killing adoption
The Implication
Every SaaS company watching this is doing the same math. Do we ship AI features to stay competitive and tank our margins? Or do we hold back and watch competitors (or startups) take our users? The answer for most will be ship first, figure out economics later. That means a wave of SaaS companies are about to discover their growth stories have GPU-shaped holes in them.
The companies that win will either crack inference costs through better models and infrastructure, or they'll pioneer pricing that actually reflects compute intensity. Tiered AI credits. Per-generation fees. Usage-based pricing that doesn't feel like punishment. Someone will figure it out. Until then, expect more headlines like Canva's.