Skip to content

The public AI Energy Index

The AI Energy Index is Venturi’s free, no-login public view of model energy, water, carbon, and eco-efficiency, served at venturi.systems/ai-energy and built from the same versioned catalog that powers the numbers in your own Venturi instance. This page documents what the viewer shows and where its data comes from.

Availability

The viewer is live at venturi.systems/ai-energy. Its current manifest is 2026-06-13.ap1, generated June 13, 2026.

The methodology behind every number is the same one documented in the energy, water, and eco-efficiency pages.

Where the data comes from

The viewer is a static page that reads a versioned JSON dataset. The live latest manifest names the versioned file under /ai-energy/data/. The current seed catalog identifies itself as unsigned, so these docs do not claim cryptographic verification. A future signed distribution must publish a verifiable signature and key before it is described as signed.

The seven views

View What it shows
Overview / Leaderboard A ranked table of every model: energy with its declared denominator, 1-to-5 rating, water, carbon, task, size class, derivation method, confidence, and freshness date.
Model detail A per-model card with the tri-metric (energy / water / carbon) across short, medium, and long prompt sizes (mean ± standard deviation), latency and throughput, benchmark scores, the provider multipliers (PUE / WUE / CIF), and full provenance and caveats.
Compare Two to four models side by side, with a guardrail that keeps comparisons within the same task and size class.
Time-series Day-to-day tracking of a model’s energy, flagging when its watt-hours move.
Fleet-scale impact Energy, carbon, and water projected to fleet-scale query volumes, presented with the equivalence cards below.
Reasoning breakdown Energy by reasoning effort level, making the reasoning premium concrete.
Methodology The formulas, constants, tiers, provenance, and license attributions (the same material as these docs, with the AI Energy Score named and linked).

Sourced equivalence cards

The fleet-scale view uses only equivalences with a named basis:

Equivalence Divisor
Average U.S. residential electricity customer 10.38 MWh/year, derived from EIA’s 2024 average of 865 kWh/month (EIA)
Olympic swimming pool 2,500 kL
Typical U.S. gasoline passenger vehicle 4.6 metric tons CO2/year (EPA)

Energy totals map to the household card, water totals to the pool card, and carbon totals to the vehicle card. The earlier campus, drinking-water, and transatlantic-flight divisors were removed because their geography, boundary, or unit was not sufficiently defined.

Honest disclosure on every row

The viewer surfaces the same honest-unknown discipline the engine enforces:

  • Per-row derivation_method and confidence are shown inline, so you can see exactly how each number was produced (measured, infra-aware estimate, class analogue, or unestimated) and how much to trust it. Confidence is encoded as a text-and-badge indicator, not by color alone.
  • An uncertainty band accompanies estimated rows.
  • Null is shown as null, never zero. A model with no available data is visibly unrated, not silently counted as emission-free.
  • A persistent coverage-ratio banner on the Overview states how much of the full model population the displayed catalog covers, so a partial catalog is never mistaken for a complete one.
  • A “how is this calculated?” affordance sits next to each metric, linking to the relevant methodology page.

Downloading the data

The live manifest links the versioned JSON dataset. The current seed is explicitly unsigned. No public CSV download is claimed until the viewer exposes one.

Every numeric field in the dataset is nullable (null, never a fabricated zero), and the provenance, derivation_method, and confidence fields are mandatory and non-null on every row.

Accessibility

The viewer targets WCAG 2.2 AA: ratings carry color-independent text (for example, “Model Energy Rating 4 of 5”) rather than stars alone, provenance is exposed to assistive technology, charts have data-table fallbacks, and the tables and filters are fully keyboard-navigable. Viewer state (filters, sort, and the selected model) is encoded in the URL so any view is deep-linkable and shareable.

Where to go next