AI is cheap… the same way your Uber was.
What subsidised AI really costs, and why creative thinking might matter more when the bill arrives.
I was listening to Nilay Patel interview Dara Khosrowshahi recently, when something quietly alarming slipped in. Uber has already burned through its AI token allocation for the YEAR only 3-4 months in. The Uber CEO didn’t dress it up much. The way to keep funding that usage?
“The trade-off is going to be headcount.”
Not exactly the kind of future-facing optimism we’re usually sold by technology companies.
Because if anyone understands subsidised technology, it’s Uber.
You may be old enough to remember that Uber didn’t just disrupt taxis through better design and convenience. They did it with investor money. Rides were artificially cheap, heavily subsidised to drive behaviour change at scale. And it worked like a charm.
Then the subsidies faded.
Prices crept up. The gap between taxis and rideshare narrowed. And suddenly the “new normal” looked a lot like the old one, just with a better interface and no free bottle of water.
AI feels like it’s following the same playbook.
Right now, whether you’re paying $20 or $200 a month for your AI subscription, you’re almost certainly not covering the ‘true’ cost of what you’re using. The gap is being filled, again, by investor capital, infrastructure bets, and a race for market dominance.
Which raises an uncomfortable question: what happens when that subsidy disappears?
If Uber’s experience is anything to go by, the shift won’t be dramatic, it’ll be gradual. Costs will rise. Access may tighten. And businesses that have built themselves around “cheap” AI will start looking for ways to rebalance.
That’s where the comment about headcount stops sounding like a throwaway line and starting to sound like a preview, and it may hit closer to home for anyone using AI to promote their brand or service.
Because AI isn’t being adopted evenly.
At one end, you’ve got casual users, people generating a birthday invite, a quick social post, something disposable. At the other, you’ve got power users: developers, product teams, businesses building entire systems on top of AI APIs.
And sitting right in the middle is the group that’s most exposed, small businesses and everyday operators who’ve quietly woven AI into their workflows.
Tools like Canva are a perfect example. What started as a design accessibility platform is rapidly becoming AI-first, rewriting text, generating images, suggesting layouts before you’ve even finished the thought. On the surface, it’s incredibly useful.
But it also raises a structural question: how sustainable is that level of generation? Because unlike dragging a text box onto a page, every AI prompt carries a real cost, compute, infrastructure, energy. And right now, much of that cost is still being absorbed behind the scenes. (Again)
Canva’s ‘confident’ play to become AI first. https://www.canva.com/newsroom/news/canva-create-2026-ai/
When pricing shifts, it won’t just affect the power users burning through API credits. It will start to show up in the tools everyday businesses rely on. And that’s where things get interesting.
Because once the cost of “just generating something quickly” increases, behaviour changes.
Casual users may opt out, go back to free templates, or simply doing less. But businesses that have built AI into their day-to-day operations don’t have that luxury. For them, it becomes a compounding expense. Small at first, then quietly significant. The risk isn’t a dramatic collapse. It’s gonna be slow creep - just like Uber. Higher tiers. Usage caps. Features that were once frictionless becoming metered.
And then there’s a quieter consequence, one that has less to do with cost, and more to do with how we think. As someone who works in brand and strategy, the real concern isn’t that tools like Canva can generate anything you want. It’s that they’ll generate everything you ask for, without questioning whether you should be asking for it in the first place.
We’re seeing more content being produced, faster than ever. But not necessarily better content. Not more effective content. Just… more. Three posts a week because that’s the rule. Another variation because it’s easy. Another campaign because the tool suggested it. But strategy doesn’t work like that.
Good creative isn’t about filling space.
It’s about making decisions. What to say. What not to say. When to show up, and when not to. And there’s a real risk that, in a world where generation is frictionless, that layer of thinking gets skipped entirely. Because why stop and consider the right move, when you can instantly produce ten “good enough” ones?
As the cost of AI continues to rise, the way we use it will inevitably change. AI won’t disappear. But the volume might.
The throwaway outputs. The “just generate another one” mentality that’s been quietly fuelling an avalanche of content, most of it unnecessary, some of it useful, all of it carrying a cost we don’t really see.
Because right now, it still feels free. Or at least, cheap enough not to question.
Our little AI buddies.
Are we doing ourselves a favour,
or a disservice?
The people who benefit most won’t be the ones generating the most, they’ll be the ones who know what’s worth generating in the first place. The ones who can direct the tool, not just use it. The ones who can get to a strong outcome in fewer steps, with fewer prompts, and less ‘waste’.
In that version of the future, creative skill doesn’t disappear. It becomes more efficient. More deliberate. More valuable.
And my wish is that these costs reins in some of the excess. Because the real shift isn’t just economic, it’s cultural.
From using AI because we can,
to using it because we should.
_______
To be fair, I’m not immune to the lure of AI.
I can feel how easy it is to slip into the rhythm of, just one more prompt, one more attempt to get it “just right.” It feels productive. It feels like progress. Sometimes it’s just motion or stimulation. A very convincing form of distraction.
By end of 2024, the AI industry must generate $600 Billion in end-user revenue just to break even on hardware and operational costs. (Sequoia Capital) This revenue currently sits at $150B and the gap between the required and actual revenue is widening every quarter. So we know they are coming for free and low cost plan users.
I’m becoming more aware of that trade-off, not creatively, but economically. Because if every prompt carries a cost (even if I’m not the one paying for it yet), then it forces a different question:
Is this actually moving the work forward, or just filling the space? That’s a discipline we haven’t really needed yet. But we will soon.
_______