AI costs are showing up before revenue in public cloud companies
In several cases, cost of revenue is rising faster than revenue following AI rollout — in some cases by a wide margin.
A few examples from recent filings:
One company: revenue +9%, cost of revenue +46% → gross margin down ~370 bps
Another: revenue +15%, cost +28% → margin compression despite growth
Another: revenue +11%, cost +19% → early divergence showing up in COGS
The pattern is consistent: AI workloads are landing in cost of revenue immediately (inference, GPUs, storage, bandwidth), while revenue either lags or is bundled into existing products.
So the P&L effect shows up as margin pressure first, before any clear revenue lift.
Mechanically, this makes sense:
inference is metered and continuous
pricing is often fixed, bundled, or not yet optimized
usage can scale faster than monetization
Which means companies can increase “AI usage” and still degrade unit economics in the near term.
The broader implication:
Right now, most of the AI conversation is about models, capabilities, and adoption.
But for public companies, this is a unit economics problem first.
If cost of revenue continues to grow faster than revenue, margins compress — and eventually that forces pricing changes, feature gating, or reduced usage.
Curious if others are seeing similar patterns internally, especially around inference cost vs pricing.