The Ugly Truth About Pokemon Go...

TL;DR

Niantic used Pokémon Go's AR features to secretly build a 30-billion-image dataset training AI navigation systems for delivery robots.

Key Points

• 1.Pokémon Go was fundamentally a data collection operation. Niantic's CTO Brian McClendon revealed the game's true purpose was crowdsourcing 30 billion real-world urban images to train a visual positioning AI system accurate to several centimeters.
• 2.Traditional GPS fails in urban environments. McClendon explained GPS can drift 50 meters in city canyons, putting users on the wrong block — Niantic's Spatial division aims to replace it with camera-based visual positioning.
• 3.Niantic disclosed data collection in its terms of service, but players didn't grasp the implications. In-game prompts warned of AR mapping data collection, yet users never anticipated it would train commercial AI sold to robot delivery companies.
• 4.Niantic sold Pokémon Go to Scopely in March 2025 but retained all collected data. This confirmed the game was always a means to an end — the 30 billion images were the real asset, worth far more than the game's microtransaction revenue.
• 5.143 million players unknowingly built one of the largest real-world visual datasets in AI history. Unlike Google's mapping vehicles, Pokémon Go captured landmarks, storefronts, and sidewalks from every angle, lighting, and weather condition at a scale no fleet could replicate.
• 6.Gamification was the data extraction mechanism. Dopamine loops and rare Pokémon rewards motivated players to perform AR scans for free — many even paid Niantic via microtransactions while simultaneously providing the labor that built Niantic's AI business.
• 7.Niantic's Spatial AI is now being licensed to delivery robot companies. The visual navigation system allows robots to pinpoint location within centimeters using only camera input, enabling automated deliveries without GPS — a market projected to vastly outpace Pokémon Go's earnings.