Canva Lowers Revenue Growth Forecast and Cuts AI Costs
Canva has lowered its revenue growth target for 2026 from 30% to 20% and is slowing the rollout of some AI features to restructure product costs and commercialization models.
The company stated that after the launch of Canva AI 2.0, the average service cost per AI task was too high; there was a high reliance on cutting-edge models, and self-developed models were not yet ready, leading to a disconnect between pricing, usage limits, and exceptionally high demand. After restructuring its infrastructure over the past three months, Canva reported that the cost per AI task has decreased by nearly 90%.
Free and low-cost subscription users are frequent buyers of AI services, with inference costs initially borne by Canva; funds flow from subscription gross profit to external model suppliers, cloud computing power, and self-developed model construction. The slowdown in feature rollout means that short-term new paid conversions are under pressure, but if cost reductions can be sustained, Canva will shift AI from a customer acquisition subsidy to a billable gross profit tool.
ABAB AI Insight
Canva is not just making temporary adjustments. Its acquisition of the generative AI platform Leonardo.AI in 2024 continues the previous strategy of acquiring Affinity, Pexels, Pixabay, and Kaleido: by purchasing creative tools, material libraries, and model teams, it integrates design entry points, copyright assets, and generative capabilities into the same product stack. Leonardo retains its independent platform, but its team, Canva's funding, and creator-authorized content are used to advance visual AI research and development.
This slowdown reveals a gap in the capital path: Canva first quickly launched AI capabilities using third-party cutting-edge models, then attempted to cover inference costs with subscription pricing; however, high-frequency users converted free quotas into ongoing variable costs, while self-developed models and usage controls did not synchronize. The company chose to halt distribution and redo the inference architecture, essentially shifting capital from market deployment to models, infrastructure, and task routing to achieve subsequent unit cost reductions.
This is similar to Adobe's approach of bundling Firefly with creative software subscriptions and limiting high-cost calls through generative quotas, but Canva faces a much larger pool of free users and lower price anchors. Its annual recurring revenue had reached $4 billion by the end of 2025, with 265 million monthly active users, indicating that controlling AI costs is no longer just a research and development issue, but a question of whether the free tier can continue to bear customer acquisition functions.
Essentially, this is about the transfer of pricing power: traditional SaaS charges by seat, with marginal service costs approaching zero; generative AI calculates costs based on usage, with marginal costs rising with usage. Companies that do not bind model costs, usage quotas, and payment tiers will find that the faster user growth occurs, the more gross profit is eroded; only platforms that master self-developed models, task routing, and usage rules can regain pricing power.
ABAB News · Cognitive Laws
The more active free users are, the deeper the cost trap.
AI first burns gross profit, then filters business models.
Growth without pricing power is merely subsidization.