Why we bill in joules

Cloud billing is built on the wrong units. Joules are the units physics actually charges for.

Cloud billing is built on the wrong units.

vCPU-hours, GB-months, requests, GB-egress — these are the abstractions the hyperscalers reach for because they were the abstractions a 2005 data centre understood. But none of them is what the work actually costs the universe. The work costs joules. Everything else is bookkeeping the cloud invented so it could monetise variance.

The variance is the business

Watch what happens to your bill when you reserve a vCPU-hour. You pay whether or not the code runs. A "10× over-provisioned" workload pays 10×. A workload sitting idle waiting for traffic pays the full reservation. The hyperscaler keeps the difference between what you reserved and what you actually used; that gap is high-margin pure profit.

Egress is the same trick, played in reverse. The marginal cost to AWS of moving a gigabyte out of S3 is sub-penny. AWS charges you eight cents. You can't move your data anywhere else without paying this tax, so you don't. The data has a gravity well around it; the hyperscaler owns the well.

Joules don't have a gap to monetise

Bill in joules and the variance evaporates. Idle workloads draw idle-floor watts, not reservation watts; you pay idle-floor cents. A workload that scales to zero literally pays zero. There's no gap between what you reserved and what you used — there's no reservation. The unit is the thing that physically happened.

This breaks the hyperscaler pricing model on purpose. We can't monetise variance because there isn't any.

What we make money on instead

The honest answer: we mark up the joules. Buying a kilowatt-hour wholesale from Hetzner Helsinki costs us ~€0.12. We sell joules at a markup that lets us pay for the substrate, the people, the support, and a profit margin we don't apologise for. The markup is the same whether you're doing a tiny lookup or a giant reasoning job — it scales linearly with what you actually used.

The wedge isn't "we're cheaper". The wedge is "the bill matches the work". For some workloads we'll come out 60% cheaper than the hyperscalers; for some we'll be comparable; for a few we'll be more expensive because the work genuinely costs more energy than the hyperscalers had to absorb. In every case, the receipt tells you why.

What this unlocks for you

• You can finally optimise. Your inefficient code is currently free — your bill doesn't care how many wasted joules went into it. Joule pricing puts inefficient code on the bill in red ink. The optimisation wins are dramatic; you've just never seen them.
• You can attribute energy per end-customer. Pass X-Customer-Tag on each call; the receipt carries it. Now your B2B SaaS can charge per-feature based on what the feature actually costs you.
• You can file Scope 2 cloud emissions with primary evidence. Joules and grid carbon, signed, every call. Auditor-ready.
• You can architect for energy. Cold paths, batched jobs, region pinning to cleaner grids — these are levers that show up in the bill now, not in vague carbon estimates monthly.

Why no one else does this

The honest answer: legacy. The hyperscalers' billing systems took 15 years to build; rewriting them to measure joules per request is multi-year work that would obsolete their pricing-as-leverage flywheel. They'd rather not.

The dishonest answer is that measuring per-request energy is technically hard. It isn't — Intel RAPL ships in every modern x86 CPU and reports per-package energy at millisecond granularity; NVIDIA's NVML exposes per-GPU draw the same way; Apple's IOReport does the same for Apple Silicon. The technology has been there for a decade. We just decided to wire it to the billing rail.

If you've ever opened a cloud bill and thought "why am I paying for that?" — that question is the inheritance of unit-economics-as-strategy. We picked a different unit. The bill matches the work. Try it.