The 2023-2026 model price war — and why your cost model is already obsolete — step 4 of 8
The incumbent trap — when your cost model is a fossil
In Chapter 26 we used the phrase "process debt" to describe what locks incumbent service businesses out of an AI-native rebuild. The same concept applies, with a different surface area, to LLM products that priced themselves against a model that no longer exists.
Call this process debt of pricing models. It's what happens when the price curve has moved but everything about your company — your pricing page, your sales motion, your headcount, your support playbook, your CFO's spreadsheet — still assumes yesterday's costs.
The shape of the trap
A company shipping an LLM feature in 2023 made dozens of decisions calibrated to GPT-4 pricing. None of them looked load-bearing at the time:
- The pricing page lists $99/month. The deck explained the price in terms of "compute cost per request × expected requests."
- Sales was hired to sell the $99 plan. Commissions, quota, OTE all calibrated for that price point.
- The product team scoped features against "how many calls per user can we afford?" The answer was 100/day.
- Support was sized for a customer base willing to pay $99 — i.e., power users who use the product a lot.
- The CFO model assumed 70% gross margin on those calls and modeled three years out from that.
By 2026, the underlying API costs have fallen 10-100×. In a frictionless market, this should be pure margin upside. In practice it is a strategic catastrophe, because:
- A new entrant prices the same feature at $9/month and still has 80% margin. The $99 product is undefendable on the pricing page — buyers compare and churn.
- Dropping the price to $9 to compete blows up the sales motion. Reps quit because they can't make quota on a $9 ACV. The deck doesn't work. The CFO model collapses.
- Adding features fast enough to justify the $99 requires shipping 2-3× the surface area of the new entrant, which the team is not structured for.
- Keeping the price flat and losing customers slowly is the path of least resistance — and the path to irrelevance.
This is the same trap that hit Jasper. It is the trap currently hitting every "AI for ___" company that raised at 2023 valuations, shipped against 2023 model prices, and built an org chart against 2023 sales motions. The cost curve moved underneath them and the company couldn't move with it.
Why incumbents can't escape
The new entrant doesn't have any of this debt. They priced from day one against Haiku-4.5 economics. Their sales motion, their support team, their product surface, their CFO model — all sized for $9 ACV from the start. They get the same margin on a $9 product that the incumbent got on $99, because their cost basis is two orders of magnitude lower.
The incumbent is stuck. Their options are:
- Stay at $99, lose customers slowly. Death by attrition.
- Drop to $9, blow up the sales motion. Self-inflicted layoffs and a quarter or two of pain.
- Add 10× more features at $99 to justify the price. Requires a bigger team than the unit economics now support.
There is no "just lower the price" button. The whole company was built around a number that no longer exists. This is what process debt looks like inside an LLM company — and it compounds in the wrong direction every quarter the model prices keep falling.
The strategic lesson
If you are building today, assume the model price you're paying today is the highest price you will ever pay for that level of capability. Price your product against where the cost curve will be in twelve months, not where it is today.
If you are working at an incumbent that priced against 2023 or 2024 models, the strategic question is not "how do we cut API costs?" It is "how do we restructure the entire company — pricing, sales, features, headcount — to be viable against a competitor with a 10× lower cost basis?" That question is uncomfortable. It's the only one that matters.
The next step is a quick check on which features become viable as prices fall. Then the build steps put a real cost model in your hands so you can see this dynamic on your own products.
The 2023-2026 model price war — and why your cost model is already obsolete — step 4 of 8
The incumbent trap — when your cost model is a fossil
In Chapter 26 we used the phrase "process debt" to describe what locks incumbent service businesses out of an AI-native rebuild. The same concept applies, with a different surface area, to LLM products that priced themselves against a model that no longer exists.
Call this process debt of pricing models. It's what happens when the price curve has moved but everything about your company — your pricing page, your sales motion, your headcount, your support playbook, your CFO's spreadsheet — still assumes yesterday's costs.
The shape of the trap
A company shipping an LLM feature in 2023 made dozens of decisions calibrated to GPT-4 pricing. None of them looked load-bearing at the time:
- The pricing page lists $99/month. The deck explained the price in terms of "compute cost per request × expected requests."
- Sales was hired to sell the $99 plan. Commissions, quota, OTE all calibrated for that price point.
- The product team scoped features against "how many calls per user can we afford?" The answer was 100/day.
- Support was sized for a customer base willing to pay $99 — i.e., power users who use the product a lot.
- The CFO model assumed 70% gross margin on those calls and modeled three years out from that.
By 2026, the underlying API costs have fallen 10-100×. In a frictionless market, this should be pure margin upside. In practice it is a strategic catastrophe, because:
- A new entrant prices the same feature at $9/month and still has 80% margin. The $99 product is undefendable on the pricing page — buyers compare and churn.
- Dropping the price to $9 to compete blows up the sales motion. Reps quit because they can't make quota on a $9 ACV. The deck doesn't work. The CFO model collapses.
- Adding features fast enough to justify the $99 requires shipping 2-3× the surface area of the new entrant, which the team is not structured for.
- Keeping the price flat and losing customers slowly is the path of least resistance — and the path to irrelevance.
This is the same trap that hit Jasper. It is the trap currently hitting every "AI for ___" company that raised at 2023 valuations, shipped against 2023 model prices, and built an org chart against 2023 sales motions. The cost curve moved underneath them and the company couldn't move with it.
Why incumbents can't escape
The new entrant doesn't have any of this debt. They priced from day one against Haiku-4.5 economics. Their sales motion, their support team, their product surface, their CFO model — all sized for $9 ACV from the start. They get the same margin on a $9 product that the incumbent got on $99, because their cost basis is two orders of magnitude lower.
The incumbent is stuck. Their options are:
- Stay at $99, lose customers slowly. Death by attrition.
- Drop to $9, blow up the sales motion. Self-inflicted layoffs and a quarter or two of pain.
- Add 10× more features at $99 to justify the price. Requires a bigger team than the unit economics now support.
There is no "just lower the price" button. The whole company was built around a number that no longer exists. This is what process debt looks like inside an LLM company — and it compounds in the wrong direction every quarter the model prices keep falling.
The strategic lesson
If you are building today, assume the model price you're paying today is the highest price you will ever pay for that level of capability. Price your product against where the cost curve will be in twelve months, not where it is today.
If you are working at an incumbent that priced against 2023 or 2024 models, the strategic question is not "how do we cut API costs?" It is "how do we restructure the entire company — pricing, sales, features, headcount — to be viable against a competitor with a 10× lower cost basis?" That question is uncomfortable. It's the only one that matters.
The next step is a quick check on which features become viable as prices fall. Then the build steps put a real cost model in your hands so you can see this dynamic on your own products.