Cutting Your Price 5x Won't Kill Your Margin. Your Architecture Will

Key takeaways

• You cannot discount your way to a durable low price. You have to re-engineer unit cost first.
• Oodle launched AI agent observability at $10 per million spans, claiming roughly one-fifth the cost of incumbents, by putting storage on object storage (S3) and compute on serverless.
• Its pricing is deliberately flat and legible: about $0.30 per GB ingested, $0.001 per GB per month retained, queries free.
• The transferable lesson: price is downstream of COGS. Fix the cost structure, and the low price defends itself.
• Model your gross margin at the new price before you publish it, not after.

The myth: a low price is a race to the bottom

Founders are told that undercutting incumbents is suicide, that you are just training customers to expect cheap and torching your margin on the way down. That is true if the only thing you change is the number on the pricing page. It is false if you change what a unit actually costs you to serve.

The interesting pricing moves in AI infrastructure right now are not discounts. They are cost re-architectures wearing a low price as a badge.

What did Oodle actually do?

Oodle priced agent observability at $10 per million spans and claims roughly 6x cheaper than a common alternative, about one-fifth the cost of incumbents like Datadog. The headline is the price. The story is the architecture: traces land in object storage (S3) instead of an always-on cluster, and queries run on serverless compute instead of reserved nodes, reportedly keeping P99 query time under a second.

That matters because legacy observability spend is dominated by idle capacity: clusters kept warm 'just in case.' Move the bytes to object storage and the compute to on-demand, and your marginal cost per trace collapses. Only then is $10 per million a price you can defend instead of a loss you subsidize.

Why is flat, legible pricing a weapon?

Oodle's pricing is also a positioning choice: about $0.30 per GB ingested, $0.001 per GB per month retained, and unlimited queries at $0. No per-seat, no per-query meter running in the customer's head.

Legible pricing is underrated. Usage-based models often fail not because they are expensive but because they are unpredictable, and buyers hate an unpredictable bill more than a high one. Charging on a single input the customer controls (bytes ingested) and giving away the thing they do constantly (querying) removes the anxiety that kills usage-based adoption.

The founder takeaway: price is downstream of COGS

The order of operations most founders get wrong: they pick a price to win deals, then hope the margin shows up. The durable version is the reverse. Re-architect the cost of serving one unit, measure your real cost of goods, then set a price that is both aggressive to the buyer and comfortable for you.

You do not need an S3-native rewrite to apply this. You need to know your unit economics cold: the cost to serve one user or one million units, at your current architecture and at a proposed one. That is exactly the modeling Calcaas is built for: set your token and infra costs, test a price, and see the gross margin before you commit to it in public.

Takeaway: a 5x cheaper price is not brave, it is arithmetic, if and only if you moved your cost structure first. Model the margin, then make the move.