What is spatial computing in an enterprise context?

Spatial computing is the class of systems that understand the three-dimensional physical space they operate in and let digital information be anchored to it persistently. The term originates in 2003 with Simon Greenwold's thesis at the MIT Media Lab, which defined it as human interaction with a machine that retains and manipulates references to real objects and spaces. Technically it rests on three capabilities: simultaneous localization and mapping of the environment (SLAM), three-dimensional scene reconstruction from visual sensors or LiDAR, and persistent anchoring, meaning the ability to fix a piece of information to a precise physical point and retrieve it identically in a later session, even from a different device. It is not a synonym for augmented or virtual reality: those are display modes, spatial computing is the understanding layer that makes them possible, and it exists even without a headset, for example on a smartphone overlaying a maintenance instruction on the photographed machine.

The boundary with Physical AI

The distinction from Physical AI is not terminological, it is functional. Physical AI acts autonomously in the physical world through an actuator: a robotic arm, a vehicle. Spatial computing, in its most common enterprise application, does not act on its own: it mediates between a plant's data model and a human operator, guiding them with instructions anchored to the correct physical object. A technician following a test procedure overlaid on the real machine is not a case of AI acting on the world, it is a case of AI helping a person interpret it better. The latency constraint is the one from edge AI: the operator's head moves, and a perceptible delay between the movement and the overlay's update breaks the illusion of anchoring and makes the system unusable.

An enterprise example

In a manufacturing plant, a junior technician follows an extraordinary maintenance procedure they have never performed alone: the system recognizes the specific machine by its geometry, retrieves its service history and current parameters from the maintenance system, and overlays step-by-step instructions anchored to the exact bolt or valve to work on, not to a generic illustration. The same pattern holds for a guided line inspection or for field training of a new hire. The condition that decides whether the project works is not the chosen device, headset, tablet or phone: it is whether a data model of the plant already exists, with unique identifiers for every asset and a connected history, to anchor that information to. Without that model, the system has a space to map but nothing meaningful to show on top of it, and it stays a trade-show demo.

Why it matters for decision-makers

Spatial computing is not a hardware project, it is a data project with a three-dimensional interface on top. Before evaluating a device, a company should verify whether its asset inventory, technical documentation and maintenance history are structured so they can be retrieved in real time and tied to a precise physical location. If that foundation does not exist, the first investment belongs there, not in the headset: buying it first means funding a pilot with nothing to anchor to, one that runs out the moment the demo ends.