Pokémon Go Built a Surveillance Grid. You Built

The summer of 2016 was genuinely unhinged. People were wandering into traffic, trespassing in cemeteries, and crowding public parks at 2am — all to catch a Snorlax. If you were between the ages of 12 and 35 that year, you know exactly what I'm describing because you were probably out there too. I'm not judging. The game was good.

What you probably weren't thinking about, while your phone battery cratered and your feet blistered, was that you were also doing survey work. Unpaid, enthusiastic, extremely thorough survey work. For a company that was never really a game company.

A recent video from The Infographics Show lays out the case that Pokémon Go's primary product was never Pikachu — it was you, and specifically what your camera captured while you were looking for Pikachu. The argument is worth taking seriously, even where it oversimplifies, because the parts that are documentably true are genuinely strange.

Here's the structure: Niantic's founder, John Hanke, didn't start in gaming. He started in mapping. In 2001, he founded Keyhole, an early digital mapping company that secured funding from In-Q-Tel — the CIA-affiliated venture capital firm that invests in technologies with intelligence applications. (In-Q-Tel operates at arm's length from the CIA, which is a distinction that matters when you're already writing a piece about surveillance.) Google acquired Keyhole in 2004, and the technology became the foundation for Google Earth and, eventually, Street View. Hanke went with it. Niantic spun out of Google in 2015 with investment from Google, Nintendo, and the Pokémon Company, and launched Pokémon Go in 2016.

So the person who built Pokémon Go had spent the previous fifteen years building tools to map the physical world in three dimensions. That's not a coincidence you can hand-wave.

Street View, for all its ambition, has a hard ceiling: it runs on cars, which means it runs on roads. Everything else — parks, building interiors, university quads, hiking trails, the weird little path between two apartment complexes that only locals know about — is what the industry calls "the last mile." Notoriously hard to map. Enormously valuable to map.

Niantic solved this with roughly 232 million players (the widely-cited peak monthly active user figure from July 2016, per multiple industry trackers). Those players walked their cameras through every inch of territory Street View couldn't reach, and they did it joyfully, because they were trying to catch a Drowzee. The Infographics Show estimates that hiring professional surveyors to cover equivalent ground would have cost over $15 billion. Niantic didn't pay that. They sold in-game currency instead.

This is the part that's darkly funny if you let it be: there was a mechanic in the game called AR mapping tasks, where players would physically walk around a PokéStop recording 3D video scans with their phone camera. Completing these tasks earned PokéCoins, which could be spent on items like Poffins — basically a supercharged pet treat for your Pokémon. When you ran the actual math on time spent versus PokéCoins earned versus the real-money value of those coins, the video argues players were effectively earning around a dollar an hour. And then spending it back in Niantic's store. The video puts it pointedly: "Labor was exchanged for currency that just flowed right back to the company, often at inflated prices."

There's a name for that model. It's called a company town.

Now, here's where I want to pump the brakes slightly on the framing — because The Infographics Show presents this as a secret heist, and the reality is messier than that. Niantic's data collection wasn't hidden in any legally meaningful sense. Terms of service exist. Location data permissions exist. Nobody reads them (this is a known and depressing fact of digital life), but the mechanism was disclosed. The question of whether disclosure-via-ToS constitutes genuine informed consent is real and worth having — but it's a different argument than "Niantic secretly built a surveillance grid." What Niantic did was build something enormous in plain sight, which is arguably more unsettling.

In 2024, Niantic announced it was building a "large geospatial model" — an AI trained on more than 30 billion images sourced from player-contributed data. Fifty million neural networks, trained on a decade of your vacation photos, your commutes, your local park that you scanned while trying to get a Poffin. The video notes that even players who rage-quit years ago couldn't pull their data back — once it's folded into the model, you can't unmap the park. The video cites an estimate that retraining the model to remove a specific user's data would cost between $6 and $10 million; I haven't found an independent technical source to confirm that number, so treat it as the video's estimate rather than established fact, but the underlying point — that AI models can't simply "forget" a data point on request — reflects a real and widely documented challenge in machine learning.

What Niantic built from all of this is called a Visual Positioning System: not just a map, but a 3D reconstruction of real-world environments that can locate a point in space with far greater precision than standard GPS. Niantic's own technical materials describe the VPS as capable of centimeter-level accuracy, which is a significant step beyond the several-meter radius of consumer GPS. If Street View tells you which street you're on, the VPS, per the video, "knows which brick you're looking at." That's a different category of location data.

The downstream applications are where this gets genuinely open-ended. Targeted advertising that pings your phone as you walk past a storefront is the commercial version. Law enforcement access — the video raises the scenario of locating a suspect before they leave a neighborhood — is the more fraught version. Neither of these is hypothetical in the sense of being science fiction; both represent documented uses of existing location technology, just at lower precision. The question is what changes when precision goes up by an order of magnitude.

Niantic's patent — formally titled "Generating 3D Models of Real-World Objects" — describes a system for building spatial models from crowdsourced imagery, and it puts them in a structural position that the video compares to OpenAI's in the LLM space: you can't easily build a competing model without comparable data, and Niantic has a nine-year head start on collecting it. The Niantic Spatial Platform (the successor to their Lightship developer platform) is already being positioned as the foundation layer for other companies building AR experiences. Disney and Sony building AR apps on top of Niantic's map would mean paying rent on a digital twin of the world that Pokémon players built.

The game currently has somewhere in the range of 30–35 million monthly active users depending on measurement methodology and reporting period — a fraction of its 2016 peak. Niantic has been transparent that its focus has shifted toward licensing and enterprise applications. The video's darkest observation is that once Niantic has finished extracting what it needs from the data, there's no structural reason to keep the game running — and if the servers go dark, the Pokémon go with them, while the surveillance infrastructure stays.

That's the thing about the summer of 2016 that nobody told you at the time: the Snorlax was never really yours. The park, apparently, is.

Jamie Cho covers policy and legislation for Buzzrag.