Canva cuts forecast as AI costs soar; enterprises shift toward cost control

The Canva revenue cut in August marks a turning point in how enterprises evaluate AI adoption. The company generated more than $900 million a quarter and was growing above 25%, yet still had to lower guidance—not because AI failed to deliver value, but because the input cost eroded margins. Canva's fix—rebuilding with in-house models, Leonardo.AI and task-level routing—reveals a pattern emerging across enterprise AI deployments: the real risk is not whether frontier models work, but whether a single vendor's pricing can compress a company's economics faster than the business can adapt.

What distinguishes this moment from other enterprise IT cycles is that the financial pressure is acute and visible. Uber burned through a full year's budget in roughly four months. Microsoft, despite a partnership with Anthropic, has stated openly that it wants to reduce and ultimately eliminate the cost of paying Anthropic. Lindy switched model providers and cut costs while improving performance on core use cases. These are not ideological choices about self-hosting or data residency—they are CFO-driven responses to invoices that arrived sooner and larger than projected.

The article frames this as a sovereignty issue: enterprises that control their data, evaluations, routing and exit paths can respond to cost shocks. Those locked into single vendors cannot. The key insight is that financial sovereignty does not require owning every layer—Swiss organizations using Microsoft's locally hosted services while relying on open models trained domestically illustrate that point. Rather, sovereignty is about intentional control: knowing where dependencies lie, having tested alternatives, and retaining the ability to shift workloads when costs breach the pain threshold. The paradox the article highlights is that self-hosting is expensive, yet enterprises increasingly view the predictability and optionality of a hybrid posture as worth the premium.