Canva Cuts 1,271 Jobs, 22% of Staff, as AI Costs Break Its SaaS Economics

TL;DR

Canva announced on November 3 that 1,271 people, 22% of the company, are affected by a restructuring that merges its Product, Growth and Platform groups, relocates teams into hubs, and cuts jobs, with CEO Melanie Perkins calling it the hardest moment in the design tool's thirteen-year history. The proximate cause is mundane and expensive: in August, Canva cut its 2026 revenue growth forecast from 30% to 20% because its AI features cost too much to run at the scale of 220 million monthly users. Since April it has cut the cost per AI task by nearly 90%, and users responded by making three times as many designs. The fix now shipping is cheaper vertical models of its own, a broadened OpenAI deal, and a much smaller payroll.

The memo, and what leavers get

Perkins' memo, "An important update on our team," frames the restructuring as refounding the company: one Product Org instead of three, fewer subspecialties, fewer layers, and Sydney as the primary home for product teams. Some of the 1,271 keep jobs that change or move. Others are gone. Everyone leaving gets at least six months of salary, the one-year vesting cliff removed for recent joiners, twelve months of healthcare where offered, and their MacBook.

Then there is the item you will not find in any other layoff memo this decade: the Canva Collective Fund, which hands departing employees US$10,000 grants to start something of their own, takes no equity, and gives them until December 14 to register interest. A severance package with a startup kit inside is a very 2026 object.

From January, most hybrid staff are expected in the office two to three days a week. An all-hands is set for November 10. The company operates across 37 countries, so the timing of individual conversations varies by local process.

The August downgrade

The restructuring did not come out of nowhere. In August, Canva cut its expected 2026 revenue growth from 30% to 20%, a third off, after what Perkins described as demand for new AI features that "significantly exceeded" expectations. The problem was never adoption. It was that each AI task cost more than the business could carry at flat subscription prices, so Canva slowed the rollout while it rebuilt the architecture and cut unit costs. The Australian Financial Review called it an AI bill shock and a rare revenue downgrade. Perkins also said the company had been "relying too heavily on frontier models."

Pitchbook analyst Derek Hernandez put the industry-sized label on it: AI is making SaaS "no longer a zero marginal cost solution," which had been software's quiet advantage since the first subscription invoice. Software used to cost nothing to serve the next user. Now each helpful answer has a meter attached.

Cheaper per task, more tasks

Perkins told Fortune that since Canva AI 2.0 launched in April, the company cut the cost of completing an AI task by nearly 90%. Over the same period, people using Canva AI created three times as many designs as users of the previous version. Those two numbers belong to one story: making a task cheaper invites more tasks. It is the Jevons paradox, the 19th-century observation that more efficient coal engines increased Britain's coal consumption, running on GPUs instead of furnaces.

That is the mechanism every AI feature owner should internalize before the quarterly bill does it for them. Per-seat pricing assumes a customer's cost to serve is flat. Agentic features make it linear in enthusiasm.

Vertical models as a cost strategy

Canva's answer is to stop renting frontier intelligence for routine design work. In late October it shipped Craft, its first image-generation model, then a week later Brush, a vision-language model for design edits. Both came out of Canva Research, its 160-person AI lab. Canva claims Craft made it the second most advanced lab for image generation on the DesignArena image leaderboard, a self-report worth treating as a press release until the rankings move.

Brush is the interesting one for builders. It started from a leading open-weights model, was post-trained on real design and editing tasks, then shaped with large-scale reinforcement learning whose reward function scores visual quality, layout principles and task completion, with professional designers on staff defining what counts as good. Instead of repainting whole designs, Brush runs an agentic loop that writes and executes code against design elements and then inspects the result, so The New Stack described it as writing code to edit designs rather than painting pixels. Editing one headline no longer regenerates the slide.

The launch numbers are the point of the exercise: up to 69% faster on general design edits and 46% lower cost per edit task than the previous design agent, per Canva's own announcement. Brush is live for Canva Pro subscribers now, with wider availability promised.

And a cheaper frontier deal

The same week, Canva broadened its four-year OpenAI partnership for more access to GPT-6 Astra and GPT-5.6, with both companies explicitly working to make the models more economical to run, per Startup Daily. No savings figures were disclosed, which is itself informative: the negotiation is now about inference pricing, not just capability. OpenAI already powers Canva Code, Canva sits among the five most-installed apps in ChatGPT, and in one internal test Astra navigated more than 80 million lines of Canva code and wrote over 1,000 warehouse queries.

The result is a two-tier stack: rent frontier models for hard, rare reasoning, own the cheap model for the high-frequency task. Figma, the public-market parallel, shows the same pressure with disclosed numbers: its free-cash-flow margin fell from 27% in the first quarter to 14% in the second, and it forecast third-quarter revenue growth of 36% against 48% in the June quarter, per its investor release.

Seats are becoming meters

Canva's pricing pages now carry monthly AI usage allowances by plan, with the free tier showing limited access to standard and premium AI features and paid tiers getting larger allowances, a structure it has been tightening through the autumn. Adweek's read: this is the SaaS story of the moment, the per-seat model growing a meter because the seat no longer predicts the cost. If your product bills flat and serves AI, you already have this problem; Canva is just far enough ahead to hit it first, at 220 million users.

What to steal if you run an AI product

• Do the unit economics before the launch, not after the forecast cut. Canva validated demand and then had to slow the rollout because each task cost more than the plan could carry.
• Expect the Jevons effect. A 90% cost cut alongside 3x usage per user is a draw on gross margin. Model adoption growth into your inference bill explicitly.
• Split the stack by frequency. Frontier models for rare, hard reasoning; a fine-tuned open-weights model for the task users do a hundred times a day. Brush's 46% per-edit saving is the arithmetic.
• Let domain experts set the reward. Canva's RL loop is scored by its own professional designers, which is cheaper than buying taste from an API.
• Negotiate inference, not just features. If a vendor wants your distribution as a top-five app, your unit price is a conversation.

Caveats

• The 1,271 figure is everyone affected: layoffs, role changes and relocations combined. Canva has not published the pure headcount cut, so do not read 1,271 as 1,271 firings.
• Canva is private. The 30% to 20% forecast, the 90% cost cut and the 3x usage figure are company statements relayed through Fortune and the AFR, not audited filings.
• Growth slowed for more reasons than inference, including competition from OpenAI's and Google's image models, per Business Insider's coverage of the memo.
• The DesignArena second-place claim is Canva's own line in the Brush announcement.

Key Takeaways

• Canva's November 3 restructuring affects 1,271 people, 22% of the company: merged product orgs, relocations to hubs including Sydney, and layoffs, after its first growth slowdown in a decade.
• In August, Canva cut its 2026 revenue growth forecast from 30% to 20% because AI features cost too much to run; Perkins said demand "significantly exceeded" expectations and the company had leaned too heavily on frontier models.
• Cost per AI task is down nearly 90% since Canva AI 2.0 launched in April, while AI users create 3x as many designs: the Jevons paradox with a subscription pricing model attached.
• The cost fix: in-house models Craft (image generation) and Brush (design editing, 46% cheaper and up to 69% faster per task than the previous agent, built from an open-weights base with designer-scored RL), plus a broadened OpenAI deal covering GPT-6 Astra and GPT-5.6 with work on cheaper serving.
• Figma shows the same pressure with disclosed numbers: free-cash-flow margin down from 27% to 14% quarter over quarter, with growth decelerating from 48% to a 36% forecast.
• Leavers get at least six months' salary, keep their MacBook, and can apply for $10,000 no-equity grants from the Canva Collective Fund to start their own thing.

Sources: Canva team update, Business Insider (full memo), Fortune, AFR, Canva Brush announcement, Canva Craft announcement, Startup Daily, Bloomberg, The Next Web, Canva about page