AI Isn't One Race

The companies in AI aren't all competing for the same thing. Nvidia isn't fighting Anthropic for users; Mistral isn't fighting Microsoft for enterprise contracts. So AI Scoreboard doesn't reduce them to a single ranked number. Instead it maps who is winning each competitive space — and shows the evidence behind every call.

Each verdict is directional and explainable, not mathematically definitive. We'd rather be clearly reasoned and honest about uncertainty than precise about something unknowable.

The Five Races

SpaceThe question it answers
Frontier ModelsWho is building the most capable AI?
Consumer ProductsWho is winning end users at scale?
Enterprise & B2BWho do organizations trust in production?
Developer PlatformWho do builders default to?
Compute & InfrastructureWho controls the compute the industry runs on?

Most companies seriously compete in only two or three of these. Naming where a company is absent is as informative as naming where it leads.

Evidence, Not A Formula

Each race is judged against a handful of observable KPIs — benchmark indexes, web traffic, enterprise contracts, API adoption, capex, market share — drawn from public sources. But we never sum those metrics into a score. The KPIs are evidence that supports a stated editorial verdict you can check, not inputs to a hidden calculation.

The output for each race is a plain-language read of the current state, plus where each company stands and why.

Where Companies Stand

Within each race, every company gets a position rather than a number:

PositionMeaning
LeadingAt the front of this race.
ChallengingClose behind the leaders.
In the mixA real competitor, not at the front.
EnteringNew to this race, building fast.
FadingLosing ground.
Not competingNot meaningfully in this race.

A Roster That Moves

There is no fixed list of companies. A company appears because the data places it in a race, and the tracked universe is simply everyone who shows up in at least one space. That's how players the headlines overlook — Chinese frontier labs, chip-supply and foundry companies — earn a place the moment the evidence warrants it, and how a fading company quietly drops out.

Cadence & Data Status

The board updates on a monthly published cadence, while the underlying research is refreshed more often behind the scenes. Every dataset carries a status:

StatusMeaning
Editorial DraftHuman-judged from public signals; not yet fully sourced or reviewed.
PublishedReviewed, sourced, and timestamped.
ArchivedA past month, kept for comparison and trend history.

The current board is an Editorial Draft. It is real and reasoned, but still being verified — treat it as informed commentary, not a final ranking.

Sourcing & Honesty

Every verdict pairs with source notes and a confidence rating — High, Medium, or Low. When fresh data conflicts with our assumptions, we treat it as something to verify: we flag anomalies rather than quietly discarding them. Estimated numbers are labeled as estimates, and we'd rather show our reasoning than a false decimal.