Game as Data Engine

Pokémon GO is widely known as a mobile AR game. Less understood is that over its decade of operation, Niantic quietly built what may be the world's largest repository of real-world spatial image data — an estimated 30 billion geotagged images collected by hundreds of millions of players walking through cities, scanning landmarks, and exploring neighborhoods.

In 2024, Niantic spun out its spatial AI division as a standalone entity, Niantic Spatial, with the explicit goal of commercializing this data asset. The first meaningful proof point is a partnership with Coco Robotics, a last-mile delivery robot company in Los Angeles — signaling a "game → data → AI model → industrial application" value chain that few companies have managed to execute.

Ten Years in the Making

Niantic was founded in 2010 as an internal Google startup, spun out independently in 2015. Its original thesis was augmented reality games as a platform — but the deeper asset being accumulated was something few investors initially appreciated: dense, real-world spatial imagery at scale.

The spin-off structure reflects a deliberate strategic pivot: the gaming business and the spatial AI business have different customers, revenue models, and growth trajectories. Separating them allows Niantic Spatial to pursue enterprise contracts without the consumer gaming narrative obscuring its pitch.

The Spatial AI Stack

Niantic Spatial's product is a layered technical stack — from raw data to developer-facing APIs — that enables any application requiring precise real-world location awareness.

1. VPS (Visual Positioning System)

VPS replaces GPS with camera-based localization. Rather than triangulating satellite signals — which suffer from urban canyon effects[1] that cause 20–50m errors in dense cities — VPS matches real-time camera frames against a pre-built 3D map of the environment to determine position. Accuracy reaches centimeter-level, even indoors and underground where GPS fails entirely.

The critical differentiator: the 3D map underlying VPS was built from Pokémon GO player scans. No competitor has a comparable dataset. Building an equivalent from scratch would require years of dedicated field capture and billions of dollars in operational cost.

2. World Model

Beyond positioning, Niantic Spatial is developing a World Model — an AI that understands spatial context, not just coordinates. It can distinguish whether a given location is a sidewalk, a building entrance, or a road crossing; infer foot traffic patterns; and understand the semantic meaning of physical environments.

This moves the product from "where am I" (GPS-equivalent) to "what is this place" — a capability critical for autonomous systems making real-world decisions.

3. Lightship AR SDK

The developer-facing platform allowing third-party apps to access Niantic's spatial data and VPS capabilities. Positioned as an infrastructure layer for AR developers, similar to how AWS provides compute infrastructure without dictating what runs on top of it.

Where Spatial AI Gets Deployed

Niantic Spatial sits at the intersection of several large and fast-growing markets, each of which requires precise real-world spatial understanding as a foundational capability.

The autonomous delivery robotics market alone is projected to reach $80B+ by 2030[2], with last-mile delivery representing approximately 50% of total logistics cost. The inability to navigate reliably in dense urban environments is the primary technical barrier — which is exactly what Niantic Spatial's VPS addresses.

Selling Infrastructure, Not Software

Niantic Spatial's commercial model mirrors infrastructure platform businesses — charging for access to capabilities rather than end applications.

Primary Revenue Streams

• API / SDK Licensing: Enterprise contracts granting access to VPS and spatial data APIs. Coco Robotics represents the first announced B2B customer in the robotics vertical.
• Data Licensing: Selling access to the underlying 30B image dataset for AI training and HD map building to autonomous vehicle and robotics companies.
• Custom Integration: Professional services for enterprise deployments requiring tailored spatial models or city-specific data enrichment.

The Infrastructure Moat

Unlike software businesses where competitors can replicate features, Niantic Spatial's moat is the dataset — and datasets of this type accumulate over years, not quarters. A competitor starting today would need to deploy hardware infrastructure globally or find an equivalent crowdsourcing mechanism, both of which face significant scale and cost challenges.

What Could Go Right

1. 01 Robotics as a beachhead, spatial computing as the upside. Delivery robots are a tractable first customer — high pain point, clear ROI, willingness to pay. But the larger prize is spatial computing platforms (Apple Vision Pro, Meta Orion) that need world-anchored localization at scale. Niantic Spatial is positioning itself as the invisible infrastructure layer beneath both.
2. 02 The data flywheel compounds. Every new customer deployment generates additional real-world scan data that improves the model. Unlike pure software, quality improves with usage. This creates a winner-takes-most dynamic if Niantic Spatial can sign enough early enterprise customers to create feedback loops ahead of competitors.
3. 03 GPS-independence becomes a regulatory requirement. As autonomous systems proliferate in urban environments, regulators may require redundant localization that does not depend on GPS. Niantic Spatial would be uniquely positioned to serve as the backup — and eventually the primary — positioning layer for certified autonomous systems.
4. 04 International coverage advantage. Pokémon GO had near-global player penetration. The resulting dataset covers cities across Asia, Europe, and South America — markets where incumbent HD map providers (primarily US and EU focused) have sparse coverage. This opens markets that competitors structurally cannot serve yet.

What Could Go Wrong

1. 01 Go-to-market execution risk. Having a great dataset is not the same as building a successful B2B enterprise sales motion. Niantic is primarily a consumer company. Selling into robotics procurement cycles, integrating with ROS[3]-based robot software stacks, and managing enterprise SLAs requires capabilities Niantic Spatial is building from scratch.
2. 02 Dataset freshness and maintenance. Physical environments change constantly — new buildings, construction, demolished landmarks. Keeping 30B images current requires a continuous data pipeline. As Pokémon GO engagement declines[4], the primary data collection mechanism weakens. Niantic Spatial needs to build alternative update pipelines before this becomes a quality problem.
3. 03 Privacy and regulatory exposure. A database of 30 billion images of public spaces, including incidental captures of people and private property, represents significant regulatory risk in the EU (GDPR), and increasingly in the US. Any enforcement action or mandatory data deletion could materially impair the core asset.
4. 04 Well-funded competitors with alternative approaches. Apple, Google, and Microsoft all have large-scale spatial mapping programs. LiDAR-based approaches (Mobileye, HERE HD Live Map) offer different but overlapping capabilities. The camera-only VPS approach is cost-competitive but may face performance ceilings in edge cases that richer sensor stacks handle better.

Bottom Line

Niantic Spatial is attempting one of the more interesting infrastructure pivots in recent tech history: converting a consumer gaming data asset into B2B spatial AI infrastructure. The core insight — that game mechanics can be designed to crowdsource the physical world at scale — is validated. The 30 billion image dataset is real and defensible in the near term.

The open questions are commercial: Can a consumer-DNA company build enterprise go-to-market? Can the dataset stay fresh as the game that generates it matures? And can Niantic Spatial land enough customer deployments fast enough to create the data feedback loop before better-capitalized competitors build equivalent coverage via alternative means?

The Coco Robotics partnership is a meaningful first proof point. The next 12–18 months — measured in the number of additional enterprise contracts signed and the quality of VPS performance data published — will determine whether this is a genuinely differentiated infrastructure business or a dataset looking for a business model.

Related Spaces to Watch