# Scaling Semantic 3D Data for Spatial AI

Labeled, large-scale 3D and geospatial datasets become the scarce input for spatial and embodied foundation models, and each credible release pushes 3D perception toward the open-vocabulary treatment that reshaped 2D vision.

- Conviction: 38 / 100 (weakening)
- Horizon: Emerging (watchlist)
- Tracking since: 2026-07-06T00:00:00.000Z
- Last updated: 2026-07-07T14:00:02.329Z
- Canonical: https://polylog.news/ai/trends/spatial-3d-foundation-data
- Publisher: Polylog
- Affected regions: Global

## Recent score history

- 2026-07-06: 40
- 2026-07-07: 38

## Recent evidence

- [confirms] A Claimed Billion-Scale Facade Dataset Surfaces, Awaiting Verification (2026-07-06): A research post advertises a claimed billion-scale, centimeter-accurate cross-continental building-facade point-cloud dataset with fine-grained semantic labels for training 3D perception models, though it remains awaiting verification.