A Medium piece by Jose Ruacho making the rounds makes a structural argument worth flagging: the price you pay for Claude Pro, ChatGPT Plus, Copilot, and virtually every frontier-model product is not the price of the inference. It is the price of the subsidy. The actual cost of the compute is covered by capital that is not trying to recoup on a twenty-dollar monthly subscription — it is trying to buy market share, usage patterns, and the future ability to reprice once dependency is established.
I've been calling this "comfortable drift" in the context of AI tool sprawl and token budget exposure. The Ruacho framing is the same pattern at a different layer. Teams are not just accumulating tool dependencies — they are building workflows around a cost structure that does not reflect the underlying economics of providing the service.
The repricing event is coming. What it will feel like is a sudden 3x price increase. What it actually is: normalization. The infrastructure cost being passed through rather than absorbed by capital.
The operational question for small teams is not whether to use AI tools. It is whether your workflows assume current pricing is a floor or a ceiling. If you are building headcount plans, client commitments, or internal automation around a cost structure that assumes $20-100/month per seat stays in that range permanently, you are building on a subsidy that belongs to someone else's balance sheet.
I don't have a clean answer. But I'd rather have the conversation before the invoice, not after.