AI researchers want to map the 3D world. That means going vertical—and possibly nuclear
Spatial intelligence is an emerging approach to deploying AI in the physical world. By combining mapping data with artificial intelligence, it aims to deliver “smart data” tied to specific locations—even indoors and across multiple stories. With this technology, maps become more intelligent, evolving into 3D representations that offer contextual information not just about places, but also about the people and activities within them. While Google Maps and other platforms have long mapped streets and roads, the vertical, or Z-axis (i.e. spaces above and below ground) has remained largely uncharted. Until now.
The ability to project dynamic intelligence into our world has long been a staple of science fiction and has indeed inspired some real world innovations in Silicon Valley such as early cellphones at Motorola, which were inspired by the Star Trek communicator; or the metaverse, which was inspired by Neil Stephenson’s novel Snow Crash); any one of several technologies inspired by the film Minority Report, and holograms inspired by Star Wars, to name a few. Today, with large language models handling the delivery and interpretation of information, maps are struggling to catch up. Embedding AI more deeply into our physical environments not only gives contextual relevance to the structures and parts of our physical world, it creates new channels for data collection and analysis—about people, behaviors, and interactions. This means, in effect, that information about places is no longer limited to a flat map; it can now be dynamic, hyper-localized, and personal.
Who’s building spatial intelligence—and why
Many companies are working on spatial intelligence from different angles. Some, like Descartes Labs, synthesize satellite, weather, and market data into actionable intelligence. Blacksky supports military and commercial partners by tracking rapid changes in geography. Mapbox customizes route planning by adding layers of intelligence—like dining options, scenic routes, or even music. Carto helps businesses analyze spatial data to identify patterns and turn those insights into strategy or revenue. While companies like Carto work on backend analytics, others like Mapbox provide direct consumer utility. And the AR industry continues to geolocate content.
However, today’s spatial analysis is still largely limited to flat maps. To broaden its utility, mapping must become more immersive—accounting for 3D objects, vertical space, and highly localized environments.
Another ambitious player in the field, World Labs, was cofounded by Stanford professor Fei-Fei Li, often called the “godmother of AI.” The company is building Large World Models (LWMs) that can perceive, generate, and interact with 3D worlds—complete with physics, semantics, and control. With $230 million in funding from a sizable roster of Silicon Valley luminary VC’s and companies, World Labs aims to generate limitless virtual spaces, essentially layering privatized dimensional data onto the digital world. Dr. Li’s dual role—leading this company while also advising the state of California on AI policy—raises potential conflicts of interest, especially as other companies lack such access or influence.
The energy problem behind the tech
Building and operating spatial intelligence systems demands massive power, and already AI’s energy consumption is straining the grid. As AI’s energy appetite grows, some tech companies are turning to nuclear power. Microsoft, for example, is funding the restart of Three Mile Island in Pennsylvania; and Amazon has acquired land near the plant. Alphabet/Google signed a deal to purchase energy from multiple small modular reactors through Kairos Power, and Nvidia, has been backing nuclear-powered AI deployments with PG&E at California’s Diablo Canyon by managing the facility’s First Commercial Deployment of an “On-Site Generative AI solution” for their Nuclear Energy Sector. Meanwhile Meta, whose prior attempts towards nuclear energy were thwarted by bees, has not given up its effort, and has released an RFP seeking nuclear power partners with either small or large reactors to help them reach their energy usage goals.
It is worrying that the companies that have created the “move fast and break things” ethos are now getting into the nuclear energy game—and selling their generative AI software to run these nuclear power plants. Many small distributed reactors are being proposed by startups with limited nuclear experience, some near population centers. There are also questions about the dangers and resilience of AI-run software and testing practices, especially when applied to systems as critical as nuclear energy.
The return of nuclear power in service of AI and spatial intelligence raises serious concerns, given past nuclear efforts in the U.S. have been fraught with issues such as meltdowns, leaks, and equipment failures. Now, with climate change intensifying and weather patterns destabilizing, previously “safe” waste storage sites may also be at risk.
Spatial intelligence holds undeniable appeal to many, both for utility and profit. The ability to access real-time, contextual information about any space—even vertical ones—is indeed seductive. But underneath that convenience lies a complex reality: a world where every space could be subject to tracking, surveillance, and monetization. And that, combined with a nuclear push, plays into powering the tech sector and government’s vision for new kinds of smart cities, hinted to have less regulation and broader “”experimentation,” with big energy needs. As AI’s energy demands grow, the quiet revival of nuclear power by tech folks suddenly “disrupting power” to build, restore, and run private nuclear power plants, may carry many more unforeseen consequences that will impact all of our lives.