Canva's investors reduced its valuation by $7.1 billion, and its own appraiser made an even larger cut.

      Emma Rapaport and Paul Smith detailed the markdown for the Australian Financial Review on 14 August. The valuation is presented in two formats, both of which are accurate: it stands at $7.1 billion in US dollars and A$10 billion, the figure highlighted in most headlines. The adjustment reduces Blackbird and Airtree’s value from $42 billion to $34.9 billion, a decline of 17%. Both firms are among Canva's longstanding investors.

      The more significant adjustment originated from within Canva. The company's independent valuation has decreased more substantially than that of its investors, with the AFR noting that this internal downgrade exceeds theirs. This figure is crucial as it represents not just an opinion but the price at which employees can sell their shares. Over the past year, it has fallen from $38.9 billion to $31 billion.

      Thus, the write-down amounts to $7.1 billion by one measure and $7.9 billion by another, with both valuation methods showing a similar trend. Importantly, neither reflects a transaction; no one has bought or sold Canva at those prices, and this distinction is significant for what follows.

      The markdown was triggered by earlier revenue news. On 3 August, the AFR reported that Canva had reduced its expected growth rate by a third, bringing it down to 20%. The headline of that story directly identified the reason behind it: Canva faced an AI bill shock. The company struggled to implement AI tools due to the rising costs associated with using advanced models. In the second quarter, revenue hit $921.9 million, up 25.2%, but failed to meet its internal expectations.

      Melanie Perkins indicated that demand was the issue, not a lack of it. The desire for new AI features "significantly exceeded" expectations. In her account to Fortune via email, she provided a clear narrative about the situation, being unusually open regarding the sequence of events. She stated that the launch "validated the demand" but also revealed the need to lower the cost of completing an AI task to facilitate a wider rollout.

      Then came the decision: "Instead of broadly launching a product before the underlying economics were in place, we opted to slow the rollout while we improved the architecture, decreased unit costs, and bolstered the business model.” Perkins mentioned that the cost per task has dropped by nearly 90% since the launch of Canva AI 2.0 in April. Users are now creating three times the number of designs, indicating that the savings do not align with the percentage reduction.

      Derek Hernandez, a senior research analyst at Pitchbook covering SaaS and AI, provided insights into the structural implications. He noted that software economics are at stake, stating, "AI is making SaaS no longer a zero marginal cost solution, which has been one of the key advantages of software until now." He further explained his analogy: building a Ford F-150 is akin to training, while the fuel and the mechanic represent inference, "as that's the purpose of using the product."

      Previously, serving an additional user of a design tool incurred minimal costs. Now, every AI-assisted image comes with an associated computing cost. Hernandez also connected this to public markets, noting that Canva and Figma encountered similar challenges within just five days of each other, albeit with different implications.

      Figma, experiencing 48% growth, raised its outlook but saw its stock decline by 16% due to margin concerns. Its free cash flow margin dropped to 14% in the second quarter from 27% in the first, with third-quarter growth expectations set at 36%, down from 48%. One company navigated a public repricing in a day, whereas the other, being private, was repriced by its shareholders.

      This issue of rising costs is affecting larger buyers as well, with Amazon experiencing a project that exceeded its budget by 860% before it failed. Responses are starting to converge: Microsoft has implemented spending limits on internal AI use, while EY developed an AI router to allocate tasks to less expensive models instead of defaulting to the cutting-edge options, aligning closely with Perkins’ vision of reconstruction.

      Rory O’Driscoll from Scale Venture Partners highlighted timing on the 20VC podcast, stating, "Many will be facing costs in 2026 and 2027 due to hesitations in 2023 and 2024." AI is not merely an added feature for Canva; it forms the basis of its expansion strategy. The company has been diversifying beyond design into enterprise workflows, introducing tools like Canva Code. Perkins previously told Fortune in 2023 that the AI market was overly fragmented, leading to a strategy aimed at consolidating those elements within a single platform.

      This presents a dilemma for the entire industry, as companies cannot afford to bypass the AI boom, but fully embracing it can threaten the economics they strive to safeguard. Lovable, a competitor in Europe, operates on a similar premise, but lacks a comparable margin