Back in April, Canva launched Canva AI 2.0, calling it a "reimagining of how the world creates". The update introduced conversational design, letting users start projects with an idea, sketch or even a voice note. But shortly after the launch, Canva found itself unexpectedly having to press pause on the rollout.

"Canva AI 2.0 is a complete re-architecturing" of Canva AI. We think it's the best AI creation experience you can get," Canva co-founder Cameron Adams told us at a press event at the brand's new London headquarters this week.

But as Canva began rolling the features out in April, it faced a unique problem – and one that Adams thinks other brands can learn from. The demand for Canva AI quickly exceeded its expectations. "Too many people wanted to use the product. There were soon 75M people using it, presenting huge cost and server load increases". So Canva paused the roll-out. With some re-engineering, it managed to reduce the cost-per-task by almost 90%, ending up with models that were 5x faster and 30x cheaper.

Adams says this exactly how business need to think in order to succeed in a rapidly shifting AI landscape. It's not just about raw power, but how to keep costs down.

"There's a lot of focus right now on the big models, such as Claude and ChaGPT," Adams says. "But the best tech isn't always the winner. Betting only on a model is a very fragile approach. We’re going to be constantly in a cycle of models overtaking each other." This is why Canva now has over 120 researchers working on AI models internally and using open source.

Because customers, Adams argues, ultimately don't care specifically which model(s) you're using. They just care how it affects them. "We want to make something accessible, reliable and connected to the way people want to work. We’re playing the long game, translating new tech into everyday human capability," he says. "We don’t settle for short term vision."