Niantic Spatial

Scroll sideways for more →

Niantic Spatial, Inc. is an American geospatial artificial intelligence company that builds maps and 3D reconstructions of the physical world for machines, including humanoid robots, delivery robots, augmented reality glasses, and AI agents. The company was created on May 29, 2025, when Niantic, Inc. completed the sale of its games business, including Pokemon GO, to the Savvy Games Group subsidiary Scopely for $3.5 billion and spun out its geospatial technology under founder John Hanke with roughly $250 million in capital^[1][3][4]. Its stated mission is to build a Large Geospatial Model (LGM), a spatial intelligence counterpart to large language models, trained on more than 30 billion posed images contributed largely by players of Niantic's location-based games^[5][7][24].

Background

Niantic Spatial's technology and leadership trace back through two decades of digital mapping. John Hanke co-founded Keyhole in 2001, a satellite imagery startup that Google acquired in 2004 and turned into the basis of Google Earth and Google Maps; Brian McClendon, Niantic Spatial's chief technology officer, was the other Keyhole pioneer^[5][6]. Hanke founded Niantic in 2010 as an internal unit at Google, released the location-based game Ingress in 2012, and spun the company out of Google in 2015^[6]. Pokemon GO, launched in 2016 with The Pokemon Company, became one of the most successful mobile games ever; Forbes put its cumulative revenue at roughly $8 billion by 2025, with about 100 million people playing during 2024^[6].

The games doubled as a data collection engine. Players who opted in scanned publicly accessible places with their phone cameras, feeding Niantic's Visual Positioning System (VPS), a computer vision service that lets a device work out its position and orientation to centimeter accuracy from a single image^[7]. In November 2024, Niantic researchers Eric Brachmann and Victor Adrian Prisacariu published "Building a Large Geospatial Model to Achieve Spatial Intelligence," announcing that the company was using this corpus to train a single global model rather than millions of independent local ones. At the time Niantic reported 10 million scanned locations worldwide, about 1 million of them activated for VPS, roughly 1 million fresh scans arriving each week, and more than 50 million local neural networks trained, with over 150 trillion combined parameters. The blog argued that the dataset was distinctive because it was captured from a pedestrian perspective, including places cars cannot reach, and that a global LGM "extrapolates locally by interpolating globally," filling gaps in any one location with knowledge learned from thousands of similar ones^[7].

The announcement also drew criticism. 404 Media broke the story under the headline "Pokemon Go Players Have Unwittingly Trained AI to Navigate the World," and Forbes and Game Developer likewise reported that many players had been unaware their scans were training a commercial AI model^[8][9][25]. Niantic responded that scanning has always been optional, that it applies only to publicly accessible locations, and that "merely walking around playing our games does not train an AI model"^[7][8].

Spin-off from Niantic

On March 12, 2025, Niantic announced it would sell its entire games division to Scopely, the mobile games publisher owned by Saudi Arabia's Public Investment Fund through Savvy Games Group, for $3.5 billion. With an additional $350 million cash distribution from Niantic's balance sheet, equity holders received about $3.85 billion in total. Pokemon GO, Pikmin Bloom, Monster Hunter Now, and the companion apps Campfire and Wayfarer went to Scopely along with the teams behind them^[1][2].

The geospatial technology business stayed with Hanke as a new company, Niantic Spatial, Inc., capitalized with about $250 million: $200 million from Niantic's balance sheet and a $50 million investment from Scopely^[1][3]. The transaction closed on May 29, 2025, the date Niantic Spatial treats as day one of its independent existence^[4][5]. Forbes reported that around 400 gaming employees transferred to Scopely, that Niantic Spatial retained roughly 200 people, and that more than 65 positions were cut in the restructuring^[6]; a California WARN filing put the count at 68 roles at the San Francisco headquarters, cut in two waves on May 20 and June 15, 2025^[29].

Hanke led the company as CEO through its first ten months. On March 30, 2026, Inhi Cho Suh, a former IBM executive and past president of Docusign, became chief executive, with Hanke moving to executive chairman; the company framed the change as a shift toward commercial scaling of its physical AI business^[19]. Thomas Gewecke serves as president and chief operating officer^[24].

Technology

Niantic Spatial's core claim is that the next phase of AI needs grounded knowledge of specific real places, not just language or plausible video. A January 2026 company essay laid out a three-layer view in which large language models supply reasoning, world models supply learned physical intuition, and Large Geospatial Models supply ground truth about actual locations, answering "Where am I? How am I oriented? What am I looking at?"^[18] This positions the LGM as complementary to, and competitive with, world-model efforts such as World Labs, which generate plausible 3D environments rather than reconstructions of particular real ones^[18].

The platform is sold as four products^[18][20]:

Scroll sideways for more →

VPS localization is accurate to centimeters, where satellite positioning is off by meters at best, and it works in urban canyons and indoor spaces where GPS fails outright^[20][21]. For robotics, Scaniverse can export a capture as a USDZ file combining a Gaussian splat, which carries real-world lighting and texture, with an aligned collision mesh, ready to load into NVIDIA Isaac Sim and Isaac Lab as a training environment^[20]. By 2025 the company said its LGM training corpus exceeded 30 billion posed images; its site now also cites more than 300 million user-contributed scans^[5][24]. Research teams are based in London and Oxford^[5].

The company kept two games as ongoing demonstrations of its technology rather than as a primary business: Ingress Prime, the 2012 title whose players seeded the original location database, and the AR pet game Peridot^[1][2].

Robotics and physical AI

Robotics became the clearest expression of the company's pivot from consumer AR to enterprise physical AI. Niantic Spatial pitches its reconstructions as the missing "real2sim" half of robot training: instead of training in generic synthetic scenes, a robot can train in a faithful replica of the exact site where it will work^[20].

Real2sim2real humanoid navigation with Flexion and NVIDIA

On July 20, 2026, Niantic Spatial and Flexion, a Zurich-based humanoid autonomy startup founded in 2024 that raised a $50 million Series A led by DST Global Partners, with participation from NVIDIA's NVentures, in November 2025^[27], published a joint demonstration with NVIDIA of a complete real2sim2real pipeline: an RGB-only navigation policy trained entirely in simulation inside a reconstruction of a real office, then transferred zero-shot to a physical humanoid robot navigating that office, with no on-site fine-tuning^[22][23].

The pipeline works as follows^[22]:

1. A person walks the deployment site once with an off-the-shelf 360-degree RGB camera.
2. Niantic Spatial estimates camera poses and trains a photorealistic 3D Gaussian splat of the site, preserving lighting, materials, and clutter as captured.
3. A collision mesh is derived from the same capture using MVSAnywhere, a zero-shot multi-view stereo model that produces cleaner geometry on low-texture surfaces such as white walls, where conventional multi-view methods struggle.
4. Splat and mesh are exported together as a single gravity-aligned, metric-scale USDZ file matching NVIDIA's NuRec volume specification, loading directly into Isaac Sim and Isaac Lab.
5. Flexion runs massively parallel reinforcement learning inside the digital twin on a single GPU, using domain randomization plus large-scale offline-trained image encoders, for millions of rollouts.
6. The resulting RGB-only policy runs on the real robot's onboard camera, with no depth sensor required for navigation.

In benchmarks of 1,024 rollouts per evaluation, with the same spawn and target poses reused across policies, the companies reported^[22][23]:

Scroll sideways for more →

The blog states that RGB trained in the Gaussian splat reconstruction was the only RGB configuration that matched or beat the conventional depth baseline; RGB policies trained in a synthetic office or on a generated mesh underperformed depth, with the gap widening in the harder scene^[22]. The companies highlighted three situations where RGB beat depth sensing: semantically defined hazards with little geometric signature, such as a blue floor mat that barely registers in a depth stream; thin structures like tripods, railings, and cables that depth methods miss or blur; and transparent surfaces such as glass doors and windows^[22]. None of those exact obstacle configurations were present in the original scene capture; the policy handled them by generalizing from related training experience. The post is equally direct about the limit of the approach: "RGB is not magic," and the policy generalizes less well to objects far outside its training experience^[22]. The companies argue the combined pipeline compresses deployment timelines from months of on-site adaptation to days, and Flexion separately demonstrated Reflect v1.0, its long-horizon autonomy stack combining navigation, interaction, planning, and recovery^[22][23]. The published material does not identify the humanoid hardware model used^[22][23].

Delivery robots

In March 2026, Niantic Spatial became an infrastructure partner to Coco Robotics, the Los Angeles sidewalk delivery company, deploying VPS so Coco's robots can hold accurate position in urban canyons where GPS degrades or drops entirely, and navigate precisely in high-interaction spots such as pick-up zones^[16].

Partnerships and customers

Scroll sideways for more →

The Snap deal was announced during Evan Spiegel's keynote at AWE 2025 and gives roughly 400,000 AR developers the ability to anchor content at mapped locations, while some 900 million Snapchatters can discover those experiences and contribute scans that expand the map^[10][11]. Hanke described the ambition as building "spatial intelligence that helps people and machines better understand, navigate, and engage with the physical world"^[10].

Not every inherited product survived the enterprise pivot. In March 2026 Niantic Spatial released 8th Wall, the WebAR platform Niantic acquired in 2022 in what it then called its largest acquisition, as free open-source software under an MIT license while winding down its hosted services, including user logins and the cloud editor^[17].

Competitive position

Niantic Spatial occupies a distinct spot in the spatial AI field. Against world-model companies such as World Labs, whose Marble generative world model became commercially available in November 2025^[28], it argues that generated environments are "plausible" while its reconstructions are ground truth for real places, which matters when a robot or drone must act safely at a specific site^[18]. Against traditional mapping providers, it claims a moat in its pedestrian-perspective corpus: more than 30 billion posed images collected over a decade of location games, covering walkways, interiors, and other spaces vehicle-mounted sensors never see^[5][7]. Its named markets are robotics, defense and intelligence, oil and gas, logistics, construction, and large venues^[18][20][21]. The same data provenance remains its most persistent source of criticism, since the corpus underpinning products now sold to enterprises and defense customers was gathered through consumer games, a tension press coverage has tracked since the LGM announcement^[8][9][15]. The issue sharpened in June 2026, when reporting on the Vantor partnership asked whether navigation models headed into military drones had benefited from player scans; both companies said raw Pokemon GO scan data was not shared with Vantor, though coverage noted they did not address whether previously trained models had drawn on it^[15].

See also

• Spatial intelligence
• Gaussian splatting
• Digital twin
• Sim-to-real transfer
• Flexion
• World Labs
• NVIDIA Isaac Sim
• Physical AI

References

1. "Pokemon GO maker Niantic is selling its games division to Scopely for $3.5B." TechCrunch, March 12, 2025. https://techcrunch.com/2025/03/12/pokemon-go-maker-niantic-is-selling-its-games-division-to-scopely-for-3-5b ↩
2. "Scopely to Acquire 'Pokemon Go' Maker Niantic's Games Business for $3.5 Billion." Variety, March 12, 2025. https://variety.com/2025/gaming/news/scopely-pokemon-go-niantic-sale-3-billion-1236334814/ ↩
3. "Niantic Spatial spin-off nets $250m funding after Scopely acquisition." PocketGamer.biz, March 2025. https://www.pocketgamer.biz/niantic-spatial-spin-off-gets-250m-funding-after-scopely-acquisition/ ↩
4. "Pokemon Go under new ownership as Scopely closes $3.5bn Niantic deal." PocketGamer.biz, June 2025. https://www.pocketgamer.biz/pokemon-go-under-new-ownership-as-scopely-closes-35bn-niantic-deal/ ↩
5. "Day One for Niantic Spatial." Niantic Spatial, May 29, 2025. https://www.nianticspatial.com/blog/niantic-spatial-day-one ↩
6. Nieva, Richard. "Pokemon Go Made Niantic Billions. Now It's Ditching Gaming For AI." Forbes, May 21, 2025. https://www.forbes.com/sites/richardnieva/2025/05/21/niantic-scopely-pokemon-go/ ↩
7. Brachmann, Eric and Victor Adrian Prisacariu. "Building a Large Geospatial Model to Achieve Spatial Intelligence." Niantic, November 12, 2024. https://www.nianticspatial.com/blog/largegeospatialmodel ↩
8. Collins, Barry. "'Pokemon Go' Players Are Training AI Models To See The World." Forbes, November 21, 2024. https://www.forbes.com/sites/barrycollins/2024/11/21/pokmon-go-players-are-training-ai-models-to-see-the-world/ ↩
9. "Niantic's new AI model may have been built by unaware Pokemon Go players." Game Developer, November 2024. https://www.gamedeveloper.com/business/niantic-pokemon-go-ai-model ↩
10. "Niantic Spatial and Snap's Multi-Year Strategic Partnership to Build AI-Powered Map." Niantic Spatial, June 10, 2025. https://www.nianticspatial.com/en/blog/vps-snap-investment ↩
11. "Snap Inc. & Niantic's New Partnership Could be a Big Moment for City-scale AR Positioning and Experiences." Road to VR, June 2025. https://roadtovr.com/snapchat-niantic-spatial-partnership-vps/ ↩
12. "Niantic Spatial and Aechelon Technology Partner to Enhance U.S. Coast Guard Training." Niantic Spatial, September 8, 2025. https://www.nianticspatial.com/blog/uscg-simulator ↩
13. "Aechelon and Partners Niantic Spatial, ICEYE, BlackSky and Distance Debut Project Orbion: the First AI-Enabled Digital Twin of the Earth." PR Newswire, September 2025. https://www.prnewswire.com/news-releases/aechelon-and-partners-niantic-spatial-iceye-blacksky-and-distance-debut-project-orbion-the-first-ai-enabled-digital-twin-of-the-earth-302549818.html ↩
14. "KOJIMA PRODUCTIONS AND NIANTIC SPATIAL ANNOUNCE NEW PARTNERSHIP." Kojima Productions, September 2025. https://www.kojimaproductions.jp/en/niantic-spatial-announcement ↩
15. "Pokemon Go Scans Quietly Trained the Navigation Tech Now Headed Into Military Drones." DroneXL, June 9, 2026. https://dronexl.co/2026/06/09/pokemon-go-scans-niantic-vantor-military-drone-navigation/ ↩
16. "Niantic Spatial Partners with Coco Robotics to Accelerate the Future of Autonomous Delivery." Niantic Spatial, March 10, 2026. https://www.nianticspatial.com/blog/coco-robotics ↩
17. "Niantic's WebAR Creation Platform '8th Wall' Goes Open Source as Hosted Services Go Offline." Road to VR, March 10, 2026. https://roadtovr.com/niantic-webar-platform-8th-wall-open-source/ ↩
18. Hayden, Hugh. "Niantic Spatial & The Large Geospatial Model." Niantic Spatial, January 13, 2026. https://www.nianticspatial.com/blog/world-models-2026 ↩
19. "Welcoming Inhi Cho Suh as CEO of Niantic Spatial." Niantic Spatial, March 2026. https://www.nianticspatial.com/blog/welcome-inhi-cho-suh ↩
20. "Niantic Spatial for Robotics." Niantic Spatial, accessed July 23, 2026. https://www.nianticspatial.com/robotics ↩
21. "Niantic Spatial for Oil and Gas." Niantic Spatial, accessed July 23, 2026. https://www.nianticspatial.com/en/oil-and-gas ↩
22. "Niantic Spatial, Flexion, and NVIDIA: Closing the Sim2Real Gap for Humanoids." Niantic Spatial, July 20, 2026. https://www.nianticspatial.com/blog/flexion-humanoid-real2sim-sim2real ↩
23. "Niantic Spatial, Flexion, and NVIDIA: Closing the Sim2Real Gap for Humanoids." Flexion, July 21, 2026. https://flexion.ai/news/niantic-spatial-flexion-and-nvidia-closing-the-sim2real-gap-for-humanoids ↩
24. "About Niantic Spatial." Niantic Spatial, accessed July 23, 2026. https://www.nianticspatial.com/about ↩
25. "Pokemon Go Players Have Unwittingly Trained AI to Navigate the World." 404 Media, November 19, 2024. https://www.404media.co/pokemon-go-players-have-unwittingly-trained-ai-to-navigate-the-world/ ↩
26. "Meow Wolf and Niantic Spatial Announce Plans to Explore an Expansion of the Meow Wolf Universe." Niantic Spatial, June 2025. https://www.nianticspatial.com/blog/meow-wolf-partnership ↩
27. "Flexion to use Series A to build sim-to-real, AI systems powering humanoids." The Robot Report, November 2025. https://www.therobotreport.com/flexion-raises-50m-build-ai-systems-power-humanoids/ ↩
28. Bellan, Rebecca. "Fei-Fei Li's World Labs speeds up the world model race with Marble, its first commercial product." TechCrunch, November 12, 2025. https://techcrunch.com/2025/11/12/fei-fei-lis-world-labs-speeds-up-the-world-model-race-with-marble-its-first-commercial-product/ ↩
29. Carter, Justin. "Niantic is laying off 68 employees after selling its game business for $3.5 billion." Game Developer, 2025. https://www.gamedeveloper.com/business/niantic-is-laying-off-68-employees-after-selling-its-game-business-for-3-5-billion ↩
30. "Niantic Spatial and Vantor Partner to Deliver Unified Air-to-Ground Positioning in GPS-Denied Areas." Business Wire, December 16, 2025. https://www.businesswire.com/news/home/20251216330019/en/Niantic-Spatial-and-Vantor-Partner-to-Deliver-Unified-Air-to-Ground-Positioning-in-GPS-Denied-Areas ↩