Sources: Canva slashed revenue growth forecast as heavy use of new AI features drove up costs and slowed their rollout, while more Canva users turned to ChatGPT (The Information)

Canva introduced a suite of AI-enhanced design capabilities with the expectation that they would boost engagement and revenue. In practice, the features consumed far more compute resources than anticipated, inflating cloud and licensing bills. The resulting expense pressure forced the company to cut its projected revenue growth, a clear signal that the AI rollout was not financially sustainable under current operating conditions. Engineers must therefore treat AI feature launches as high-impact changes that require upfront cost modeling.

The cost surge also slowed the pace at which Canva could ship additional AI improvements, creating a feedback loop where delayed enhancements reduced the perceived value of the service. Users, noticing slower innovation or higher latency, began shifting their workflows to ChatGPT, which offered comparable AI assistance without the same performance penalties. This user migration underscores the importance of maintaining a reliable, low-latency experience when competing against entrenched AI platforms.

From an observability standpoint, the incident highlights gaps in Canva's ability to correlate usage spikes with real-time expense data. Engineers should instrument AI endpoints to emit metrics on request volume, latency, and per-call cost, feeding them into dashboards that trigger alerts when spend thresholds are approached. Without such visibility, cost overruns can remain hidden until they impact financial guidance, as happened here.

Adopting tighter observability does carry overhead: teams must integrate cost-aware telemetry, possibly augmenting existing monitoring stacks with billing APIs and custom aggregation logic. This adds development and maintenance effort, and the instrumentation must be accurate across multiple cloud providers to avoid blind spots. Moreover, the approach may not scale for every low-traffic feature, so engineers need to define clear criteria for when cost-focused monitoring is justified.

The practical limit of this strategy appears when the marginal cost of additional monitoring outweighs the savings it can generate. In environments where AI usage is sporadic or the financial impact is minimal, extensive cost telemetry could become a drain on engineering resources. Therefore, organizations should prioritize observability investments for high-volume, revenue-critical AI services, while accepting coarser metrics for less critical components.