Niantic Spatial

Market position

The company positions itself as an independent AI company applying spatial intelligence across robotics, energy, construction, logistics, large venues and the public sector, drawing on one of the largest datasets of real-world spatial imagery — over 30 billion posed images and more than 300 million user-contributed scans — accumulated through years of location-based gaming and mapping.

The company grounds its models in a proprietary corpus of posed real-world imagery and user-contributed scans built over 25 years of spatial computing work, combining reconstruction, global visual positioning that functions without GPS, and open-set per-point semantics in a single stack; it also authored SPZ, an open-source Gaussian splat file format adopted across the industry.

Technology

A Large Geospatial Model trained on billions of posed real-world images plus terrain, drone and satellite data underpins three product lines. Reconstruction fuses multi-sensor inputs into geo-referenced Gaussian splats and textured meshes at building-to-city scale, using the company's open-source SPZ splat format. The Visual Positioning System is described as powered by 50 million neural networks and 150 trillion parameters, providing positioning and orientation where GPS fails. The Understand layer applies open-set, per-point semantics with cross-modal grounding of visual, spatial and linguistic information to support natural-language queries over 3D maps. Research output includes ACE-G scene coordinate regression (ICCV 2025), Cross-View Splatter feed-forward view synthesis with georeferenced images (CVPR 2026) and PlaceIt3D language-guided object placement (ICCV 2025); the company reports 140+ patents filed or pending.

Go-to-market

Direct enterprise sales for reconstruction, localization and understanding products, supported by partnerships and integrations with technology and industry players including Snap, Meow Wolf, Aechelon Technology (U.S. Coast Guard flight simulator training), Kojima Productions, Vantor, Coco Robotics, Spexi Geospatial, ExxonMobil, Flexion AI and NVIDIA. Scaniverse serves as a free-to-use capture front end feeding the underlying model, and the company publishes research at computer-vision conferences.

Enterprise and government customers in robotics and autonomy, defense and intelligence, oil and gas and utilities, construction, logistics, large venues and smart-city/public-sector applications, plus field operations teams working in GPS-degraded environments. Scaniverse also reaches individual 3D-capture users.

Geography

Headquartered in San Francisco, California, United States, with a self-described global, multidisciplinary team collaborating with research institutions worldwide.

History

The underlying technology and team originated at Niantic, which began as an internal Google startup in 2010 and became independent in 2015. The founding team previously built Keyhole (founded 2001, acquired by Google in 2004), which became the basis for Google Earth and Google Maps; Niantic launched its Visual Positioning System in 2022. On March 12, 2025 Niantic announced the sale of its mobile games division to Scopely for $3.5 billion; the deal closed on May 29, 2025 and Niantic Spatial was spun out as an independent company capitalized with $250 million ($200 million from Niantic's balance sheet and $50 million from Scopely), with all original Niantic investors remaining shareholders. Founding leadership comprised John Hanke as CEO, Brian McClendon as CTO and Thomas Gewecke as president and COO. On March 30, 2026 Hanke appointed Inhi Cho Suh as CEO and moved to executive chairman while remaining chairman of the board.

Risks & controversies

The Wikipedia article on the company carries a notability maintenance tag questioning whether independent, substantive secondary coverage exists. The business is newly separated from its consumer games heritage, and its spatial dataset was substantially accumulated through crowdsourced player activity in location-based games that are now owned by another company.