Usage-based pricing is the sharpest model in software — and the sharpest sword cuts both ways

The last chapter’s software charged by the seat: buy as many as you have employees. Step into data infrastructure — Snowflake, Datadog, Confluent — and the meter changes entirely. Snowflake doesn’t ask how many people you have; it charges for queries run, data stored, compute burned. Datadog charges by hosts monitored, logs ingested. This is usage-based pricing, and it reflects a deeper shift: data itself is becoming the workflow, and whoever owns the data infrastructure owns an increasingly critical layer of the enterprise.

The sharp side is beautiful. Expansion is written into the physics of the product: as a customer’s business grows, its usage grows automatically — no salesperson, no new contract — so net revenue retention often runs well above seat-based peers, and pricing aligns tightly with consumption. It’s the purest compounding machine in software.

But the same edge cuts back. The mechanism that makes expansion automatic makes contraction automatic too. When budgets tighten, a customer doesn’t cancel — they just run fewer queries, and revenue drops instantly. Worse, data software faces a phenomenon of its own: “cloud cost optimization,” where whole teams are spun up in downturns purely to cut data spend. So usage growth is high but not automatically guaranteed — it’s procyclical and can swing violently. A boom-era NRR here is the most dangerous number to extrapolate.

None of this makes usage-based businesses worse — at their best they’re the purest compounding machines in software. It makes them harder to underwrite. A seat-based revenue line and a usage-based one that both read “30% growth” carry completely different risk: the seat number is contractually rigid for the term, while the usage number can evaporate the moment a customer optimizes or the cycle turns. Same growth rate, very different durability — and the market too often pays the same multiple for both.

Then there’s the competitive structure, a three-way melee unique to this layer. Independents are pincered by open source (the free version is “good enough” for many) and by the cloud giants — and here’s the trap: your landlord is also your rival. Snowflake runs on the same clouds (AWS, Azure, GCP) whose owners sell competing data products. Which raises the question the whole chapter builds to: is the data moat actually real? “Data is the new oil” is a slogan; data gravity — the fact that accumulated data is painful to migrate — is a real, if bounded, moat. Distinguish the two, apply a cycle discount to usage growth, and remember: the pricing and cyclicality pains data software feels today may be what the whole SaaS industry feels tomorrow.

This is Chapter 10 of “Compounding in the Cloud.”