You might not believe it.

Canva, a global software leader with 260 million monthly active users and $3.5 billion in annual revenue, is grappling not with a lack of users but with users loving its platform a little too much.

Here’s what happened.

In April this year, Canva launched AI 2.0: users can generate editable designs, documents, websites, and marketing materials with just a single prompt.

With a stronger product, Canva pushed AI to the core of its offerings.

But four months later, things took an unexpected turn. Canva not only slowed its AI rollout but also revised its 2026 revenue growth forecast from around 30% to about 20%.

The reason? Canva’s costs were soaring, and the more active users were, the less profitable they became.

Canva is not alone. In Q2 this year, design software leader Figma reported a 48% YoY revenue increase—but operating costs surged by 117%. The earnings release triggered a 15% post-market plunge in Figma’s stock.

When every user click incurs real costs, even the lightest software business becomes “heavy.”

C-end software, once hailed as one of the best businesses of the past 20 years, is transforming into something far more familiar: a gym.

Encouraging spending while fearing overuse.

Before AI, Canva operated a classic SaaS model.

It attracted a massive user base with a free online editor, then converted users into Pro or enterprise clients through premium templates, copyrighted assets, and other products.

These offerings shared a key trait: the more users and frequent the usage, the lower the unit cost.

This model fueled Canva’s growth. By the end of 2025, it had 260 million MAUs, over 28 million paying customers, $3.5 billion in annualized revenue, and 95% of Fortune 500 companies as users.

But AI disrupted this perfect system. Canva discovered an awkward truth:

The more active users were, the less profitable they became.

Two main reasons explain this.

First, free users—once a valuable traffic asset—became a real cost.

To retain its vast free user base, Canva didn’t lock AI entirely behind a paywall. Free users still get limited AI usage: 200 standard AI calls or 20 advanced AI calls per month, shared across the account.

The issue? AI’s cost structure differs entirely from traditional software.

Every template view costs Canva nearly nothing. But every AI-generated image or model call consumes GPU and inference resources.

Thus, the free traffic that once drove Canva’s growth now became a financial burden.

The second problem was worse: rising costs, but users refused to pay more.

In 2024, Canva attempted to migrate early Teams users to a new per-seat pricing model.

Some U.S. users previously paid $120 annually for five seats. The new standard: $10/user/month, minimum three seats. A five-person team’s annual fee jumped from $120 to ~$600—a 4x increase.

This struck Canva’s most sensitive nerve.

Canva had lured users from Adobe by being simple and affordable—a budget alternative to professional tools like Adobe.

A sudden price hike triggered backlash. Canva caved, allowing some early Teams customers to keep their original pricing.

Canva’s struggle isn’t unique. Rival Figma faced the same dilemma.

In March, Figma introduced AI Credits, converting AI features into usage quotas to limit calls. The reason? Figma, too, couldn’t sustain the costs.

But quota limits didn’t curb Figma’s rising expenses.

In Q2 2026, Figma’s cost of revenue skyrocketed from $27.89 million a year earlier to $60.47 million—a 117% YoY surge, far outpacing revenue growth. GAAP operating losses hit $117.3 million, compared to a $2 million profit the previous year.

A key driver? Increased spending on AI-related tech infrastructure and hosting. The earnings release triggered a 15% post-market stock drop.

Canva and Figma’s predicament reflects a broader issue for traditional C-end software companies:

The land-grab revenue models of the past no longer work in the AI era.

Since AI drives up costs, the obvious solution is to raise prices.

But the first AI apps to try this soon learned it wasn’t that simple.

The first approach? Hike subscription fees.

Adobe led the charge. After introducing Firefly in 2023, it raised its main plan from $52.99/month to $59.99. In 2025, it launched Creative Cloud Pro at $69.99.

But prices rose while AI usage soared even faster.

Adobe’s latest earnings show subscription revenue up 13.7% YoY—but subscription costs jumped 16%.

Here’s the awkward (awkward) truth for pure subscription models in the AI era: revenue is fixed, but compute costs fluctuate.

Thus, a second pricing model emerged: usage-based pricing.

Replit adopted “Effort-Based Pricing” in 2025, charging by the workload of tasks completed by its Agent. Simple tasks might cost $0.25; complex ones over $1.

This way, heavy users directly cover their extra compute costs. But it shifts risk from the company to the user.

Agents differ from regular software. After a button click, they might plan, search, call tools, and iterate multiple times in the background. Users see one task but remain unaware of the tokens consumed or steps taken.

On July 11, 2025, Replit overcharged ~6% of paying users due to a billing error and had to issue refunds plus $10 credits.

In Replit’s community, developers complained: a project that once cost $2–3 now cost $30 under the new model. One user quit after being charged $32 for a failed task.

This highlights the biggest flaw in usage-based pricing: prices become volatile, raising users’ psychological costs.

Worse, when everything breaks down into tokens, calls, and compute costs, users start doing the math:

Why not buy APIs directly from OpenAI or Anthropic?

This erodes the software premium at the application layer.

Salesforce faces similar scrutiny. Despite its Annual Recurring Revenue surpassing $1.2 billion (up 205% YoY), markets wonder: Are customers paying Agentforce for Salesforce’s decades of customer data, business processes, and enterprise software expertise—or just for a wrapped layer of model calls?

Enter the third approach: outcome-based pricing.

This aligns best with AI’s value logic. Users don’t care about tokens or Agent iterations—only whether the job gets done.

Zendesk was an early adopter.

Since 2024, it charged based on AI-“auto-resolved” customer service tickets, starting at ~$1.50 per resolution. The platform only bills when AI truly solves a problem.

Compared to token-based pricing, this better matches the value customers seek.

But it faces a tougher question: What counts as “resolved”?

Loose standards let platforms claim more resolutions, but customers suspect AI merely closes tickets. Strict standards leave many AI-generated values unbilled.

Under Zendesk’s early rules, if AI replied and the user didn’t interact for 72 hours or escalate to a human, the chat counted as auto-resolved.

One Zendesk community member audited 192 auto-resolutions and found at least 40% still required human review.

Thus, while outcome-based pricing seems ideal, execution is hardest. Tokens are easy to count; outcomes are hard to define.

After circling back, more AI apps now converge on the same solution: hybrid pricing.

A base subscription fee covers essentials, with additional charges for extra AI usage or outcomes.

For example, Intercom’s customer service AI requires a $39/month base seat plus $0.99 per resolved outcome. In March and April, Figma and Canva shifted from fixed subscriptions to “base + overage” models.

On the surface, AI software price lists are growing complex.

Behind the scenes, the industry is redistributing compute risk.

Under pure subscriptions, the company bears the risk—more user activity means higher costs.

Under pure usage-based pricing, users bear the risk—task costs become unpredictable.

Outcome-based pricing centers disputes on definitions—what truly counts as “done”?

Hybrid pricing is the most practical compromise for now. Companies secure a revenue floor via subscriptions, then pass extra costs from heavy users via usage or outcome fees.

But this remains a pricing Band-Aid, masking deeper shifts in the software business under AI.

AI is slowly transforming this business into something else—a gym.

Companies want users hooked but not overusing their services.

Today’s AI apps navigate this tension, constantly seeking new equilibrium.