Where AI companies are actually built

There is a version of the AI map everyone carries around: San Francisco, then a long gap, then everywhere else. The data mostly agrees — but the gap is larger than people think, and the cities behind it are not competing at the same game. Each one has a specialism, and the specialism is more useful than the ranking.

What this map can and cannot see

Two things to know before reading any of the numbers below. First, 2,700 live tools carry no headquarters in our records — 42% of the live catalog. They are excluded from every share here rather than assigned somewhere, so treat these as counts of what we have located, not a census of the world.

Second, our coverage is not uniform. Singapore was swept deliberately and close to exhaustively in 2026, which is why it sits second by volume. That placement reflects how hard we looked, not how much is there relative to London or Tel Aviv. Nowhere else in this piece is a hub’s rank an artefact in that way, but that one is, and it would be dishonest to present it otherwise.

The concentration is the story

The Bay Area holds 1,761 products, more than the next nine hubs put together. Whatever you think about whether that is healthy, it is the operating condition: if you are building AI anywhere else, your competitors, your reference customers and your acquirers are disproportionately in one place, and your calendar has to account for it.

The fourth-largest “hub” is not a city at all. 241 products are headquartered nowhere in particular — distributed teams with no office to put on a map. That is a bigger population than London, Tel Aviv or Boston.

Every hub has a specialism

Ranking cities by volume mostly measures how big they already were. The more useful read is what each place concentrates on, because that is what determines whether it is the right place for a particular company.

Tel Aviv is a security cluster, and has been since long before the current wave — the talent pipeline out of military signals units predates AI by decades and simply redirected. Toronto and Beijing both centre on AI Models, for the same underlying reason in two different systems: a concentration of research labs with people who train rather than integrate. NYC builds for commerce and finance because that is who is in the building.

Bangalore is the outlier worth noting. It leads on Writing, has the lowest average opportunity score of any hub at 52, and the highest saturation. That is a cluster competing in the most contested market in the catalog, which is a structural problem rather than a talent one.

Size and score are different questions

Average opportunity score per hub ranges from 52 to 71, and it does not follow volume. Los Angeles (71 across 31 products), Toronto (69) and New York (69) all score above the Bay Area (66). Small hubs can score high simply because a handful of companies picked good problems — but the pattern across the bigger ones is real: the places with a specialism score better than the places without.

Four hubs, four kinds of company

The highest-opportunity product we have located in each of four hubs, and what it does.

What to do with this

If you are choosing where to base a company, the specialism matters more than the ranking. Being the fortieth security company in Tel Aviv buys you a hiring pool and a customer network that being the only one in a city without a cluster does not. Being the four hundredth coding tool in the Bay Area buys you the same pool and a much harder fight.

If you are investing, the hub tells you something about the default strategy of a company before you meet them. A Bay Area company is likelier to be building infrastructure and pricing for scale; an NYC company is likelier to have a named customer before it has a product.

The full ranking, including average scores and dominant funding stage per hub, is in our geography report.

Every hub we track

Photo: K / Pexels. Colour-graded for Falcoscan.

← All articles