What Happened

On 14 August 2026, the Australian Financial Review reported that Canva's longest-standing backers, Blackbird and Airtree, marked down the company's valuation from $42 billion to $34.9 billion — a 17% drop equivalent to $7.1 billion in USD (A$10 billion). Canva's own independent internal valuer cut even deeper, moving from $38.9 billion to $31 billion over the past year, a $7.9 billion reduction.

Neither figure represents a transaction. Nobody bought or sold Canva shares at these prices. But the internal valuation matters because it sets the price at which employees can sell shares — and two separate valuers moving in the same direction is a signal, not an opinion.

The markdown followed Canva's 3 August disclosure that it cut its expected growth rate by a third, to 20%. Q2 revenue reached $921.9 million, up 25.2%, but missed internal guidance. CEO Melanie Perkins told Fortune that demand for AI features 'significantly exceeded' expectations, but the cost of serving them via frontier models was unsustainable. Rather than broadly rolling out AI before the economics worked, Canva slowed the rollout to rebuild its architecture.

Perkins claims cost per task has since fallen by nearly 90% since Canva AI 2.0 launched in April. But users are now creating three times as many designs, so the savings don't fully translate to the bottom line.

Why It Matters

This is the clearest documented case of AI inference costs breaking the SaaS business model. Pitchbook analyst Derek Hernandez framed it directly: 'AI is making SaaS no longer a zero marginal cost solution, which has really been what I would call a lot of software's secret sauce up until now.'

Serving one more user of a design tool used to cost almost nothing. Every AI-assisted image now arrives with a compute bill attached. The analogy Hernandez offers: building a Ford F-150 is training, and the petrol and mechanic are inference — 'because that's the point of using the product.'

Figma hit the same wall roughly five days later. It grew 48% and raised its outlook, but its stock fell 16% on margin compression. Free cash flow margin dropped from 27% in Q1 to 14% in Q2. Figma guided Q3 growth down to 36% from 48%. One company is public and got repriced in a day. The other is private and got repriced by its shareholders. Same wall.

The pattern extends beyond design tools. Amazon reportedly ran a Claude job that went 860% over budget before failing. Microsoft has imposed spending limits on internal AI use. EY built an AI router to send tasks to cheaper models rather than defaulting to frontier. These are converging responses to the same problem.

Who Is Affected

SaaS companies adding AI features are the most directly exposed, especially those whose margins depend on near-zero marginal serving costs. The Canva and Figma examples show that even companies with strong growth and product-market fit face margin compression when AI features scale.

Enterprise IT buyers face the same pressure internally. Microsoft's spending limits and Amazon's budget overruns suggest that even the largest companies are struggling to control AI compute costs at scale.

AI startup founders building on frontier model APIs need to treat inference costs as a core P&L line item. Investors are now repricing companies that can't demonstrate a path to positive AI unit economics — Canva's markdown is proof.

Strategic Implications

For AI startup founders: Model your inference costs as a percentage of revenue from day one. Canva's 90% cost-per-task reduction shows optimization is possible, but only after a painful period of margin compression that triggered a valuation cut. Investors are now repricing companies that can't show a path to positive AI unit economics. If you're pitching, bring a cost-per-task model to the meeting.

For developers building with AI APIs: Build model routing into your architecture now. EY's AI router and Canva's architecture rebuild both point to the same solution: route tasks to the cheapest model that can handle them, rather than defaulting to frontier models. This is no longer an optimization — it's a margin requirement. Open-weight models and smaller specialized models are your routing targets.

For non-technical business owners evaluating AI tools: Scrutinize the AI cost model of any SaaS tool you adopt. Tools that bundle AI features without transparent pricing on usage limits may face margin-driven price hikes or feature restrictions. Ask vendors how they handle inference costs at scale, and whether they use model routing. If they can't answer, that's a red flag.

What to Watch Next

Three signals will determine whether Canva's cost problem is fixed or deferred: (1) the next revenue print — whether 20% growth holds; (2) whether the 90% cost-per-task reduction holds up as usage continues to scale; and (3) whether other SaaS companies get marked down for the same reason. Hernandez calls Canva and Figma 'the biggest signals so far,' which implies more markdowns are coming.

Canva's IPO timeline is also worth tracking. The company was valued at $42B in an employee share sale and expected to list in 2026. Hernandez now suggests next year, reading the slowdown as investor-facing margin protection.

Frequently Asked Questions

Q: Why did Canva's valuation get cut?

A: Canva's backers cut its valuation by $7.1 billion (17%) because AI inference costs from frontier models made its AI features unprofitable at scale. The company missed internal revenue guidance and reduced its growth forecast to 20%, prompting both investor and internal valuers to mark down the company.

Q: What does Canva's valuation cut mean for other SaaS companies?

A: It signals that AI features are breaking the zero-marginal-cost model that SaaS was built on. Companies like Figma have already seen similar margin compression. Any SaaS company adding AI features without a clear path to positive unit economics faces potential valuation pressure from investors.

Q: How is Canva fixing its AI cost problem?

A: CEO Melanie Perkins says Canva slowed its AI rollout to rebuild its architecture and reduce unit costs, which have fallen ~90% since April. However, usage has tripled over the same period, so the net savings are smaller than the percentage suggests. The fix is still being validated.