The geography of AI

Where AI gets built is usually discussed through funding announcements and conference locations. This report counts products instead: 3,741 live tools with a headquarters on record, across 20 hubs, with each hub’s specialism, score profile and funding profile set side by side.

Coverage, and two things this cannot measure

2,700 live tools — 42% of the live catalog — carry no headquarters. Every share in this report is calculated against the 3,741 that do, never against the full catalog. A hub’s count is therefore a count of what we have located there, not a census.

Second, our coverage is uneven by design. In 2026 we swept Singapore exhaustively rather than by sample, which is why it appears second by volume. That rank reflects our effort, not its size relative to London or Tel Aviv. Nothing else in this report carries that distortion, but that one is severe enough that we have not adjusted for it and would rather say so plainly.

Third, headquarters is a poor proxy for where work happens. The fourth-largest entry in this table is “remote”, which is not a place at all.

Finding one: one hub, then everyone else

The Bay Area’s 1,761 products exceed the combined total of the next nine hubs (1,599). No other technology category we are aware of concentrates this hard, and the practical consequence for companies elsewhere is that their competitive set, their talent market and their likely acquirers all sit in one time zone.

Finding two: the country picture is narrower still

The United States accounts for 2,457 located products. Singapore follows at 374 for the coverage reason above; after that the drop is steep — the United Kingdom at 182, India at 148, Israel at 132, France at 131.

Israel is the number worth pausing on. 132 products from a country of under ten million people, concentrated almost entirely in one market. Per capita it is the densest AI ecosystem in this data by a wide margin.

Finding three: hubs specialise, and the specialisms are legible

These are not arbitrary. Tel Aviv’s security concentration follows a talent pipeline that predates the current AI wave by decades. Toronto and Beijing both centre on AI Models, in each case around a cluster of research institutions. New York builds for commerce and finance because its customers are down the road. Bangalore leads on Writing, which is the one case where the specialism is a liability rather than an asset.

Finding four: scores do not follow size

Los Angeles scores 71 on opportunity across 31 products and 30.1 on saturation — the widest gap of any hub. Toronto and New York both reach 69. The Bay Area, with fifty times more products than Los Angeles, scores 66 against 37.6.

Small hubs can score well simply because a handful of companies chose good problems, so the top of this table should be read with its sample sizes attached. The robust version of the finding is the bottom: Bangalore at 52 against 53.9 is the only hub in the catalog whose average saturation exceeds its average opportunity, and that is a structural position rather than a sampling accident — it follows directly from the market mix.

Finding five: hubs differ by stage as much as by market

Seattle’s dominant stage is public, because its AI products largely sit inside established companies. Bangalore’s and remote’s is bootstrapped. The Bay Area’s and Tel Aviv’s is Series A. Those are three different economies wearing the same label, and a single ranking by product count hides all of it.

Every hub

The narrative version of this data, with what it means for choosing where to build, is in our guide to the same numbers.

Photo: Dzenina Lukac / Pexels. Colour-graded for Falcoscan.

← All reports