What Happens to Your Margin When Your Model Provider Raises Prices

A product built primarily as an interface and workflow layer over a foundation model inherits that model's cost structure directly. Unlike a traditional software cost base, where infrastructure is a small, controllable share of total cost, an AI wrapper's largest cost line can be a rate set unilaterally by a vendor, changed on their schedule, with your customer contracts and pricing usually locked in well before you know a change is coming.

This is a genuinely different risk profile than the one most SaaS financial models were built around, and it's worth naming explicitly rather than folding it quietly into a generic "cost of goods sold may vary" assumption. Treat model provider pricing the way you'd treat a key supplier concentration risk in a manufacturing business, because structurally, that's closer to what it actually is.

Calculate what share of your cost of goods sold is model API spend, and separately, what your gross margin would look like at a range of hypothetical rate increases, ten percent, twenty five percent, fifty percent. If a plausible price increase would meaningfully compress your margin, that's the number to bring to a board or investor conversation before it happens, not after, since a business that's already stress tested this looks very different from one that's blindsided by it.

Do this exercise per product line if you run more than one, since the exposure is rarely even across a portfolio. A feature that calls the model once per session looks very different under a price increase than one that calls it dozens of times, and blending them into one company-wide number hides which part of the business actually carries the risk.

Work through the exposure and your defenses in this order:

1. Calculate what share of your cost of goods sold is model API spend.
2. Model gross margin at a range of hypothetical rate increases, such as ten, twenty five and fifty percent.
3. Model the effect on your existing contracts specifically, especially those signed under the old cost assumption with months left to run.
4. Build in multi-provider flexibility so you have a real fallback and some bargaining position if pricing moves against you.
5. Capture efficiency gains such as prompt caching, smaller models for simple sub-tasks and tier routing as margin.
6. Add a usage based pricing component so your customers' bills move partly in line with your own cost.

Building your product so it can run on more than one model provider, even if you default to one, gives you a real bargaining position and a genuine fallback if a preferred provider's pricing moves against you. This isn't free: maintaining compatibility across providers costs engineering time and can mean giving up some of a specific model's particular strengths. Weigh that cost against how exposed your margin actually is before deciding whether multi-provider support belongs in this year's roadmap or next year's.

Prompt caching, smaller models for simpler sub-tasks, and better routing between model tiers are all efficiency gains that can offset a rate increase without touching your customer pricing at all. Building the habit of capturing efficiency gains as margin, rather than letting them quietly fund more usage at the same margin, gives you a buffer that's already in place by the time a price increase happens.

A pricing model with some usage-based component, rather than a flat fee entirely disconnected from your own variable cost, naturally absorbs some of a cost increase without a renegotiation, since the customer's own bill moves somewhat in line with usage. This isn't a substitute for the other levers, but a pricing structure that's already partly variable is easier to adjust than one that's entirely fixed and was set before the cost structure was fully understood.

A price increase almost never arrives at a convenient moment in your billing cycle, and the contracts most exposed to margin compression are usually the ones already signed under the old cost assumption, with months left to run. Model what happens to margin on your existing book of contracts specifically, not just on new business you'll price going forward, since the old contracts are the ones you can't simply reprice.

If a meaningful share of revenue sits in contracts with a long remaining term and no cost-based adjustment clause, that's the segment carrying the most risk, and it's worth knowing that concentration before a price increase happens rather than discovering it while renegotiating under pressure. A short list of which contracts have no adjustment mechanism and how much runway is left on each one turns a vague sense of exposure into something you can actually manage.

For any contract renewing soon, this is also the moment to add whatever adjustment language you decided made sense, since it's far easier to add a capped, transparent clause at renewal than to introduce one mid-term on a contract that's already running under different terms.