E-NOUS Foundation an independent measurement standard for AI compute

How much useful work does an AI data center deliver for each unit of energy?

The E-NOUS Foundation publishes an independent protocol that answers that question at the meter, with water and hardware measured alongside, so that an efficiency gain can be verified and financed the way other energy projects already are.

The problem

Power, not chips, now limits how fast AI computing can grow. New grid connections take years; new generation takes longer. In the near term the fastest source of additional compute is the power already connected, used more efficiently. At a site that has run out of power, every megawatt saved is refilled with more work.

Financing that improvement needs a number both sides can settle on, and AI computing has none. Measurement and verification practice already proves savings in buildings and factories by normalizing energy to units of output. AI facilities have no agreed unit of output. PUE stops at the server and says nothing about what the servers produced. Token counts are self-reported. Benchmark scores come from laboratories, not production facilities.

What the protocol does

It supplies the missing unit of output and the rules for verifying its improvement. Version 1 covers one tightly defined service: text-generation inference delivered to outside customers on dedicated capacity. Four rules carry it.

  1. §4
    Count only useful work.Output counts only if a customer received it complete, within the response time the operator promised, from the configuration on record. Padding, retries, cached answers and test traffic do not count.
  2. §3
    Measure at the meter.Every watt inside the facility boundary counts, from the grid connection through power delivery, cooling, processing and storage. Nothing is netted out.
  3. §5
    Idle hardware counts.Hardware is measured as accelerators installed, busy or idle, so unused capacity costs the operator in the measure as it does in capital.
  4. §5
    Water follows the electricity.Water consumed at the site and water consumed generating its electricity are both counted, so a site cannot report zero water by moving its use to the power plant.
E(energy)
useful work per MWh
E(water)
direct and indirect, reported separately
E(hardware)
installed accelerator-hours

Energy, water and hardware are reported as three separate results. The protocol does not combine them into a single score.

How results are verified

The protocol measures a facility against its own past. A baseline is measured before the efficiency measure is installed, and consecutive reporting periods follow with no gaps. The verified improvement is the difference, adjusted for changes the measure did not cause, and it is always stated with its uncertainty. The smallest gain the measurement can detect is calculated in advance, so both parties know before signing whether the gain they are paying for can be measured at all.

The method follows the International Performance Measurement and Verification Protocol, the framework behind most guaranteed-savings contracts in North America, extended to a sector that lacked a unit of output. Independent verifiers make the measurements, and every reported figure must be reproducible from the raw meter and log records.

Independence

The Foundation is a scientific nonprofit incorporated in Delaware in September 2026. Its role is narrow: publish the protocol, maintain it, and accredit the verifiers who apply it. It performs no verification and is party to no contract measured under the protocol.

The protocol is published in full and open to comment. A majority of the board will have no financial interest in any vendor, operator, guarantor, verifier or financing party that uses it, and anyone who holds such an interest must disclose it and step aside from related decisions. Funding sources are published annually. The protocol is technology-neutral: no cooling, power, storage or processing technology is written in as the default. It measures outcomes at the meter, whatever technology produced them.

Status

Draft 1.3 · four outside reviews complete · 21 open problems published · not yet tested on a real facility

The protocol stays a draft until an independent engineer has measured a real facility with it and an independent laboratory has confirmed the measurement is as reliable as it claims. Each stage has a pass test set in advance, so success cannot be declared by assertion.

The two-page summary is available on request. The full draft is in expert review and will be published for comment when that review is complete.

About the name

E is energy, the total energy crossing the boundary of a system. Nous is the Greek word for the intellect that grasps first principles directly, not by sensing and not by calculating, which the Greeks held to be both the highest human faculty and the order behind the world. The protocol asks the old question of a new machine: of all the energy that enters, how much becomes work a mind would recognize.

Contact

info@enous.org

Reviewers, operators, verifiers and financing parties who would like to read the draft: write and say which part of the stack you work in.