The Masthead
About Datum Index
An independent editorial observatory for the reputation of AI inference and infrastructure. Trustpilot with the temperament of a magazine — reviews and scores, yes, but rendered as considered journalism rather than a dashboard.
Datum Index began from a simple frustration: the AI-infrastructure market had acquired every kind of benchmark except the one that mattered. There were leaderboards for tokens per second and tables of price per million, but nothing that answered the question a serious buyer actually asks — which of these providers can I trust with production traffic, and how sure can I be of the numbers they publish about themselves? We set out to hold that question steady, in prose, with our reasoning shown.
We are a publication, not a platform. We write dossiers, not database rows. We keep a house voice — measured, sceptical, occasionally opinionated — because a reputation is a judgement, and judgements are made by people willing to sign their name to them.
We accept no payment, no placement fee and no sponsorship from the providers we cover. The only reputation we protect is our own. The Datum Index Desk
Our Independence
The statement
Independence is not a slogan here; it is the product. We therefore state our commitments plainly, and invite readers to hold us to them.
What independence means for us
- No paid placement. No provider can buy a higher rank, a kinder verdict, or a place in the index. Rank follows reputation, full stop.
- No sponsorship of coverage. We take no fee, commission or sponsorship tied to any provider we assess.
- No favourites. Every entry — including any provider we happen to score well — is held to the identical standard and the identical scepticism.
- Confidence in the open. Where we are unsure, we say so with a visible confidence mark rather than laundering an estimate into a fact.
- Corrections welcomed. A provider who can substantiate a figure can move our confidence mark. Evidence, not pressure, changes the record.
How We Work
The record, and its limits
Our assessments read the public record: pricing and documentation pages, status histories and incident disclosures, compliance attestations, and the accumulated verdict of practitioners. We do not run first-party latency probes, and we do not pretend that a provider's self-reported uptime is an audited fact. What we add is editorial synthesis and an honest confidence mark — the two things a raw leaderboard cannot give you. Our full weighting model and the confidence legend live on the Methodology page.
The observatory currently covers 20 providers and is reviewed on a rolling basis; this issue was last reviewed 10 July 2026. The underlying dataset is versioned (schema v1.0.0) so that every figure can be traced to a review date.