AI is a meter, not a subscription. Canva found that out the expensive way

Canva has been marked down. Blackbird and Airtree have cut their valuation of Australia’s biggest tech success story by 17%, from a 2025 peak of about US$42 billion to roughly US$34.9 billion, reported by Startup Daily. Secondary markets are pricing it lower again, the touted 2027 US listing may slip, and revenue growth guidance has come down from 30% to 20%.

Plenty of people will write about the number. The reason behind it is far more useful to a small business, and almost nobody is talking about it.

Canva slowed its own AI rollout on purpose. Not because the features did not work, and not because customers did not want them. Because the cost of serving each AI task did not stack up. Melanie Perkins has said the company chose to slow the rollout while it rebuilt the architecture and reduced unit costs, and that this hit distribution and near-term growth.

Read that again with your own business in mind. One of the best-run software companies in the country looked at what it cost to run AI every time a user pressed the button, and decided it could not afford to ship it as fast as it wanted.

The bit nobody tells you when they sell you AI

Traditional software is a fixed cost. You pay $49 a month, and whether your team uses it twice or two thousand times, the bill is the same at the end of the month. Most business owners have twenty years of habit built on that.

AI does not work like that. Every time it runs, something gets consumed. Every summarised email, every drafted quote, every document read, every image generated has a real cost attached to it, and that cost scales with two things: how often it runs, and how much work each run does.

It is a meter, not a subscription. The bill grows with use, and the whole point of a good automation is that it gets used a lot.

This is why Canva’s problem is worth your attention even though your business is not a $30 billion design platform. The mechanic is identical. It just showed up on their books first, because they have a hundred million people pressing the button.

The question to ask before you buy

When someone pitches you AI, the useful question is not “can it do this?” Almost always, yes. The question is:

Which task, how often, what does each run cost, and what is it worth?

Say you automate quote follow-ups. You send 200 quotes a month, so that is 200 runs, each one short. At a few cents a run, you are looking at pocket change against work you were previously losing entirely. That is an easy yes, and it is roughly the arithmetic behind the first five minutes deciding the sale.

Now say you point AI at every inbound email, reading long attachments and drafting replies, across three thousand emails a month. Same technology, wildly different meter reading. It might still be worth it. But it is a decision with a number attached, not a feature you switch on.

The honest comparison is always the same: cost per run, times runs per month, against hours saved times what an hour of your time is actually worth. If nobody selling you AI will give you the first number, that is the answer to your question.

The good news, which is real

Here is the part the valuation headlines skip. Canva did not give up. Blackbird’s Rick Baker says the company has since cut its cost of serving AI by 90%, unlocking AI at scale with “unit economics that actually stack up”.

Ninety per cent, in about a year. That is the direction this whole industry is moving, and it is why panic is the wrong response. The thing that was too expensive to run at scale in 2025 is affordable in 2026, and the thing that is marginal today will be obvious by next winter.

The practical lesson for a small business is not “avoid AI because it costs money”. It is: scope it to the tasks where the maths already works, and revisit the rest in six months. Automate the quote chase now. The idea that needs AI reading every document you own can wait until the meter comes down, and it will.

The quieter risk underneath

There is a second thread in the Canva story worth noting. Investor Jason Lemkin’s blunt observation was that AI agents have started routing users around Canva entirely, doing the job without ever opening the tool.

That is the same pattern we wrote about when Airtasker made itself bookable inside ChatGPT. If your value sits in a step the customer could skip, an agent will eventually skip it. Worth thinking about, though it is a slower burn than the cost question and a different article.

The func.digital take

We sell AI automation, so it would be easy to skip the bit where AI has a running cost. But pretending it is free is how businesses end up with a subscription they resent and an automation nobody trusts.

The businesses that do well with this are not the ones that bought the most AI. They are the ones that picked two or three tasks where the numbers obviously worked, measured them properly, and left the rest alone until the cost came down. Canva just did that at enormous scale and took a valuation hit for the delay. You can do the same thing on a Tuesday afternoon, for free, with a notepad.

If you want help doing the arithmetic before you commit to anything, that is part of a free digital systems audit. We will look at the tasks eating your week, work out which ones are worth automating at today’s prices, and tell you plainly which ones are not there yet. Get in touch, and let’s put a number on it.