Frequently Asked

01 What is Datum Index?

Datum Index is an independent editorial observatory of reputation for AI inference and infrastructure providers. We publish a composite reputation score, a long-form dossier and a data-confidence disclosure for each of the 20 providers we cover, alongside long-form analysis and a commented benchmark of the field.

02 How is the reputation score calculated?

The reputation score is a 0–100 editorial composite, not a first-party metric. It weighs reliability and track record most heavily, then trust and compliance, then transparency and economics, and finally raw capability and performance. It is a considered editorial judgement rather than a mechanical formula. The full weighting model is on our Methodology page.

03 Is Datum Index paid or sponsored by the providers it ranks?

No. We accept no payment, placement fee or sponsorship from any provider we cover, and no provider can buy a higher rank or a kinder verdict. Every entry is held to the identical standard. Our independence statement is on the About page.

04 Can I quote the prices and latency figures as fact?

Not as current fact. Prices are indicative USD-per-1M-token reference points for a representative model, and latency and throughput are order-of-magnitude figures, not SLAs. Both move often and vary by model, region and load. Always confirm against the provider directly before quoting.

05 What do the confidence marks (high, medium, seed) mean?

They tell you how far to trust a figure. "High" means confirmed against the provider's public pricing or documentation on the review date. "Medium" means based on public sources but not re-confirmed on the exact date. "Seed" means an illustrative value pending first-party verification — it must not be quoted as fact. The full legend is on the Methodology page.

06 How often is the data reviewed?

The observatory is reviewed on a rolling basis. This issue was last reviewed 10 July 2026, against dataset schema v1.0.0. Each dossier shows its own review date so every figure can be traced.

07 How do you decide which providers to include?

We cover serious, publicly available AI inference and infrastructure venues — frontier labs, hyperscalers, aggregators, routers, GPU clouds and silicon specialists. Inclusion is not an endorsement and omission is not a judgement; it usually means a dossier is not yet complete to our standard.

08 A figure about our company is wrong. How do we correct it?

Evidence changes the record. A provider who can substantiate a figure against a public source can move our confidence mark or correct the number. We respond to evidence, not pressure, and we upgrade confidence marks in public as verification allows.

Read the full methodology