A web of space.

The last decade of artificial intelligence was built on a lucky accident. By the time anyone thought to train a machine on everything humanity had written, humanity had already written it down and left it lying in public. The web was a training set for thirty years before anyone called it one. The models that changed everything did not gather their data. They inherited it.

The next frontier gets no inheritance. Machines are leaving the browser. They are being asked to move through hallways, reach around furniture, and understand that a kitchen is not a warehouse. For that they need what the language models had: a vast, honest record of the world they are entering. No such record exists. There is no web of rooms.

The field's first answer is simulation — build synthetic rooms, endlessly varied. Simulation is necessary and insufficient. Synthetic scenes are various in the wrong direction: they multiply what a designer can imagine and miss what real homes have by default — clutter, wear, daylight through a dirty window, the extension cord taped to the floor. The hard part of the physical world is exactly the part nobody thinks to generate. The long tail cannot be invented. It has to be found.

Finding it is a cost problem. The world is too large to scan on purpose; no lab's budget photographs a hundred million homes. But one industry already does. Every day, as a matter of ordinary commerce, real estate photographs, measures, and labels the interiors of real, lived-in homes — because that is how homes are sold and rented. Every listing is a scan someone already paid for. Marketing budgets are, by accident, the largest spatial-capture program on earth. The data the frontier is starving for gets produced daily as a by-product, used for six weeks as advertising, and thrown away.

Spatial Web is built backwards from that fact. We operate companies that real estate pays for today — capture, distribution, intelligence, voice. Every job they complete adds one more photoreal, measured, labeled interior, joined to a verified market record, to a single corpus. The corpus feeds our research: models that read rooms the way language models read text. The businesses fund the capture. The capture builds the corpus. The corpus feeds the work. Nothing here waits on a grant or a breakthrough; the engine runs on rent and closings.

Five companies run that engine today — Listing3D, EchoCV, 3DApartment, PropertyPulse, and NexusCall. Ordinary businesses, with an extraordinary by-product. The first models learned from a web of text that nobody planned. The next ones will need a web of space, and this time somebody has to build it on purpose.