Canva's $7.1bn markdown is an inference bill, not a mood swing

Blackbird and Airtree cut their carrying value for Canva from $42bn to $34.9bn, a 17% write-down worth $7.1bn, or A$10bn, as reported by Emma Rapaport and Paul Smith in the Australian Financial Review on 14 August ^[1][2][3]. The number that carries more weight is the one Canva's own independent valuer produced, because it sets the price at which employees can sell shares, and it has moved from $38.9bn to $31bn over the past year ^[4][5].

That is a $7.9bn cut, roughly 20% and therefore deeper in percentage terms than the one the investors took ^[6][7]. Neither figure is a transaction; nobody bought or sold Canva at either price ^[8].

The markdown followed the revenue news by a week. On 3 August the AFR reported that Canva had cut its expected growth rate by a third, to 20%, under a headline saying the company had an AI bill shock ^[9][10]. A third off implies a prior expectation of about 30% ^[11]. Second-quarter revenue was $921.9m, up 25.2%, and missed internal guidance ^[12].

Melanie Perkins puts the problem on the cost side rather than the demand side, saying appetite for the new AI features "significantly exceeded" expectations ^[13]. In an email to Fortune she said the launch "validated the demand, but also showed us we needed to reduce the cost of completing an AI task to support a broad rollout", and that "rather than broadly rolling out a product before the underlying economics were ready, we decided to slow the rollout while we rebuilt the architecture, reduced unit costs and strengthened the business model" ^[14][15]. Cost per task has fallen nearly 90% since Canva AI 2.0 launched in April, she says, while users are creating three times as many designs ^[16]. Those two numbers do not cancel: at a tenth of the unit cost and triple the volume, the bill sits near 30% of where it started, not 10% ^[17].

Derek Hernandez, a senior research analyst at Pitchbook covering SaaS and AI, gave Fortune the structural reading: "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" ^[18]. His analogy is that building a Ford F-150 is training, and the petrol and the mechanic are inference, "because that's the point of using the product" ^[19]. Serving one more user of a design tool used to cost almost nothing; every AI-assisted image now arrives with a compute bill attached ^[20].

Hernandez also said Canva and Figma "hit the same wall about five days apart, but they cited it in different places" ^[21]. Figma grew 48% and raised its outlook, and its stock fell 16% on margins ^[22]. Its free cash flow margin fell to 14% in the second quarter from 27% in the first, and it has guided third-quarter growth to 36%, down from 48% ^[23][24].

Large buyers are meeting the same bill and converging on the same answers. Amazon ran a Claude job that went 860% over budget before failing, Microsoft has put spending limits on internal AI use, and EY built a router that sends tasks to cheaper models rather than defaulting to the frontier, which is close to what Perkins describes rebuilding ^[25][26][27]. Rory O'Driscoll of Scale Venture Partners told the 20VC podcast that "there's going to be a lot of people paying the bill in 26 and 27 for a certain amount of hesitancy in 23 and 24" ^[28].

Canva could not sit it out, because AI is the expansion plan rather than a feature. The company has been pushing beyond design into enterprise workflows with tools including Canva Code ^[29], and Perkins told Fortune in 2023 that the AI market was too fragmented, with the strategy since being to bring those pieces inside one platform ^[30].