Ask yourself one question before you build another workflow around a cheap AI tool: who's actually paying for how cheap it is? Not you, not fully. Nvidia, Microsoft, and Amazon are absorbing enormous losses right now to make AI compute and API pricing look like a bargain, because winning the platform war matters more to them this year than winning the quarter.
That's not a stable price. It's a promotional one. And if your growth plan depends on it staying this cheap, you're building on a foundation that was never designed to hold.
A lot of growth plans getting built right now quietly assume AI tools will keep costing what they cost today. That assumption is doing more work than most CEOs realize. It's propping up build-it-yourself workflows instead of proven systems. It's shaping headcount decisions. In some cases it's baked directly into pricing. If the assumption is wrong, a plan that looked cheap and clever six months ago turns into a maintenance burden nobody budgeted for.
Here's why the assumption is shaky. Subsidized pricing is a land-grab strategy, not a cost structure. The infrastructure players eating the difference right now are doing it to win adoption and build switching costs, with the expectation that pricing normalizes once the market is theirs to lose. That's not a prediction about when it happens. It's just how this kind of subsidy has worked every other time a platform has offered one.
You don't need to predict the exact moment AI pricing corrects. You need a growth plan that still works if it does. So ask a more useful question than "which AI tool should we use": which parts of your current plan would still make sense if your AI costs tripled?
If the honest answer is "most of it," you're in good shape. If the honest answer is "not much of it," that's not really an AI problem. It's the same planning problem that shows up as inconsistent pipeline, over-reliance on one channel, or a growth plan that was never actually de-risked in the first place — AI pricing just happens to be the thing that would expose it first.
Clarity does. A real ICP, real positioning, and a system built around pipeline outcomes don't get cheaper or more expensive when AI pricing moves, because they were never priced in AI tokens to begin with. A workflow that only makes sense because tokens are subsidized right now does get more expensive, and fast, the moment that subsidy ends.
This is also, not coincidentally, why a CRM tends to be one of the most AI-resistant pieces of software a company owns. It doesn't compete on the assumption that AI stays cheap. It competes on data gravity and switching cost — the same kind of durability worth building into your own growth plan, instead of into whichever tool happens to be discounted this quarter.
Before you commit further to a plan built on today's AI pricing, it's worth pressure-testing whether it would survive a correction. That's exactly the kind of question we work through in a Growth Workshop.
If you're not sure your growth plan would hold up if AI got more expensive, that's worth thirty minutes now, before it becomes a real problem instead of a hypothetical one.
This post pulled from Episode 2 of the Growth Unfiltered Show, "Salesforce's Best Trading Day in Years: Does That Settle the SaaSpocalypse?" Antoine Vial, Brian Graf, Welcome Read, and first-time guest Braison Murgai dig into the subsidized-token argument in full, along with the rest of the SaaSpocalypse debate. Watch the full episode below.