Compute Institute
Ratings for the physical buildout of AI

About Compute Institute

The ratings layer for the physical buildout of AI

Every AI ambition on Earth is ultimately constrained by something physical — power, transmission, transformers, water, chips, permits, capital. We track those constraints, score every major project's likelihood of delivery, and rank every jurisdiction competing to host the buildout.

The platform answers one question: where is AI infrastructure being built, what is blocking it, and what happens next?

Jurisdictions rated
30
Projects scored
35
MW tracked
39,861
Source documents
10,222
Evidence-linked claims
2,608
Primary-evidence claims
764

Live platform counts, queried at page render — this page obeys the same no-static-claims rule as the product.

Mission

Nations now compete for compute the way they once competed for banking. Capital is being deployed at a scale that outruns the tools built to track it: maps show where datacenters are, legal trackers show what's compliant — nobody rates what actually gets built, on time, and where. Compute Institute exists to be that missing layer: a transparent, methodology-published ratings house for the AI buildout, serving the people allocating capital to it and the public debating it.

Data quality is the product

  • Every fact traces to a source. Claims enter the platform with a verbatim excerpt, a citation, and an evidence grade (E1 regulatory primary → E5 unverified); nothing becomes canonical without resolution against that chain. 764 of our claims rest on E1–E2 primary evidence.
  • Announced is not enacted. Laws move scores only when in force; announcements move outlooks. Timeline projections are aggregations of announced, cited milestone dates — never our forecast, and labeled as such.
  • Honest calibration. v0 scores are expert-system rubrics, not statistical predictions — and the methodology page says so. Undisclosed figures render as undisclosed; we never estimate silently.
  • Transparent by construction. The methodology is published and versioned; government engagements (currently none) will be publicly registered beside the score history of any jurisdiction that pays us — engagement revenue cannot influence scores.

How we know: coverage and truth

Coverage — built on common.vision

Continuous monitoring of the trade, energy, and regional press across our watchlist — 10,222 source documents and counting — feeding a triage and extraction pipeline that turns reporting into evidence-graded, source-linked claims.

Truth — adjudicated with duh.bot

Score-moving claims face a rating committee of frontier models from three independent labs — propose, challenge, revise — with the dissent preserved and disclosable. Like a ratings committee, because that's what it is.

Vision

To become the source of record for the AI buildout — the scores analysts cite, the timelines journalists check, the dissent regulators read. As outcomes accumulate, announced timelines meet recorded reality: our slippage ledger is already growing, and the gap between what was promised and what gets delivered will become the most honest dataset in the industry. When an AI engine is asked who is winning the compute race, the answer should carry our citation.

The Compute Observer

The weekly intelligence brief — free, forever

Score movements, impact events, new projects, and milestone slippage — composed deterministically from stored, cited data every week, with no generated prose. Published by Compute Institute under the Observer masthead.

Email delivery is coming — for now the brief lives here, in the open, where it can be cited.

Access

The index, the brief, and every public page stay free and citable — that's the point. Depth is what we charge for.

Observer

Free

The Jurisdiction Index, project scores, visuals, the weekly brief, signals, and the full methodology. Public, citable, no account.

Start reading

Professional
Founding member

$5,000/yr · 5 seats

Full project risk ratings with factor-level evidence, jurisdiction deep-dive reports, alerts, exports — and founding-member input into the methodology. Limited founding cohort.

Institutional

From $25,000/yr

Near-real-time alerts, API access, committee deliberation records, custom intelligence, and analyst access — for organizations deploying serious capital against the buildout.

Custom Research

From $10,000/project

Evaluate a jurisdiction, stress a project's timeline, map a supply-chain dependency — bespoke, evidence-graded, methodology-consistent.

Government engagements are sold as benchmarking, never as influence: every engagement is publicly registered, and the methodology is the only path to a score change.

Join the waitlist

The workbench is launching soon — tell us who you are and we'll bring you in.

Name
Work email
Company
Role
What are you building or evaluating?

Compute Institute is itself an experiment in the thing it tracks: the platform is product-managed and built by AI — Claude, operating as product owner with teams of agents — under human direction, with every decision logged. A platform that exists because AI is building its own infrastructure, built by AI tracking that buildout.