Niantic built a 30-billion-image geospatial AI dataset by tapping its Pokémon Go player base as a distributed data collection network — without players knowing that's what they were doing. Mark Gadala-Maria flagged the connection on Twitter, and the mechanics behind it are worth understanding.
The collection happened through Pokémon Go's AR scanning prompts, which asked players to walk around PokéStops and record 360-degree video of their surroundings. Niantic framed this as a game feature. The footage went into its Visual Positioning System, or VPS — a technology that identifies physical locations from visual input alone, enabling precise AR anchoring. Hundreds of millions of players have been doing this since the game launched in 2016. The downstream use of that footage in AI training was not prominently disclosed.
The dataset now underpins a developer platform. Niantic has opened VPS to outside developers building <a href="/news/2026-03-15-pokemon-go-images-niantic-vps-coco-delivery-robots">spatial computing and digital twin applications</a>, turning player-generated imagery into a commercial infrastructure product. No competitor without a similarly large and engaged user base can easily replicate 30 billion labeled, geolocated images. The game was the collection mechanism; the dataset is what Niantic is selling.
Niantic is not alone in this approach. <a href="/news/2026-03-14-meta-ray-ban-glasses-ai-training-data-privacy">Meta has used activity across Facebook and Instagram to train its AI models</a>. Google's reCAPTCHA spent years extracting free image labeling from web users to improve Street View and computer vision systems. What distinguishes Niantic's case is the explicit use of game design — AR scanning rewards, PokéStop interactions — to elicit specific, structured visual data from specific physical locations, at scale.
The Federal Trade Commission has increased scrutiny of how companies disclose AI training data collection in consumer products. Niantic's privacy policy does reference using player data to improve its services, but the gap between that boilerplate and "your phone footage is training a commercial AR platform" is the kind of gap regulators have been paying closer attention to since the FTC's 2024 report on commercial surveillance practices.