In the film *The Matrix*, the character Oracle is often translated as "Prophet," yet her true power lies not in predicting the future, but in deciphering the fundamental operating logic of the digital world, the Matrix. She understands its rules, knows each individual's position within it, and perceives what will unfold next. This is less a prediction and more a profound comprehension of the world itself.

At the Cisco AI Summit this February, Fei-Fei Li shared a story about biological evolution. She noted that human language, a source of our pride, has existed for only roughly 500,000 years in evolutionary history, whereas spatial intelligence, represented by vision and touch, began its neural evolutionary race as far back as the Cambrian period, 500 million years ago. Language, therefore, is merely an interface to intelligence, not its entirety. If AI cannot grasp the three-dimensional physical world or possess physical intuition, it will remain confined to a flat, two-dimensional pixel plane, able to see and speak but never truly "reach" the real world.

By 2026, spatial intelligence is accelerating from academic theory to industrial practice. At CES 2026, Jensen Huang defined physical AI as the "next trillion-dollar industry," asserting that AI must move from the digital realm into the physical world to generate genuine value. However, enabling AI to truly understand the physical world requires far more than training on text and images, especially in an era where language models are increasingly advanced and "information" itself is rapidly becoming commoditized. Any platform can quickly determine where to go and generate a seemingly well-structured route guide. Yet, truly judging the real world demands a system that can continuously perceive how it operates. This is precisely what Amap has been building: a city world model.

How a Map "Calculates" the Real World

Many people still perceive a map as simply roads, coordinates, and routes. But for AI, knowing a location's "where" is only the starting point. To understand a city, it must also grasp how spaces connect, what is currently unfolding in the environment, and how these changes influence human action. This is the distinction between a city world model and a traditional map. A two-dimensional road network describes streets, nodes, and reachability, while three-dimensional space adds buildings, floors, and indoor-outdoor transitions. Time allows the model to understand how the same location varies at different hours, and real-world behaviors like foot traffic, vehicle flow, searches, navigation, and visits enable the model to continuously observe how the world functions. When spatial structure, temporal changes, and actual human actions are organized into a single model, the map transforms from a mere "coordinate database" into a self-updating city world model.

Based on this world model, Amap has developed three critical capabilities for understanding the real world: three-dimensional spatial representation, dynamic perception, and spatiotemporal reasoning. The first layer, three-dimensional spatial representation, addresses the question of "what the real world looks like." From the globe to cities, roads to buildings, and outdoors to indoors, this representation aims not just to render a more realistic picture but to capture the height, distance, hierarchy, occlusion, and connectivity of the physical world. Only when these spatial relationships can be recognized and computed by machines can AI understand a location's true structure and judge how different spaces link together.

The second layer is dynamic perception, which answers "what state is the real world in right now." The physical world does not remain static after modeling is complete. Roads congest or clear, crowds gather or disperse, business hours shift, and the same street can appear entirely different by day versus night. The purpose of dynamic perception is to continuously feed these changes into the model, ensuring that AI faces not a static, outdated snapshot, but an ever-evolving representation of reality.

The third layer is spatiotemporal reasoning, which handles "how these changes will impact a journey." A real-world trip is not a simple connection of several points but an interconnected chain of time. Decisions like when to depart, which routes to take, and how long to stay at a stop, alongside whether a delay will cascade into subsequent appointments, must all be calculated within specific spatial and temporal constraints. This reasoning does not simply predict what will happen in a city; it simulates how a specific journey will unfold under given conditions, identifies potential conflicts, and offers more sensible arrangements.

Three-dimensional spatial representation provides structure, dynamic perception updates states, and spatiotemporal reasoning calculates the effects between different segments. When these three are combined, Amap's core question expands from "where is it and how do I get there" to "what does the scene look like," "what is happening right now," and "can this itinerary actually materialize in reality."

Transitioning from Wayfinding to a Holistic Lifestyle Experience

When a map is viewed as a world model, rankings cease to be a simple ordering of places and become a world simulation problem. If the internet remains confined to a two-dimensional network, it generates enormous information noise, creating discrepancies between online reviews and real-life experiences. A location might receive massive exposure through advertising or accrue numerous reviews due to tourist density. Amap's "Street Sweeping List" (扫街榜) logic bypasses this by leveraging non-fakeable real navigation and visit behaviors, offering users the most direct judgment of a place and providing a trustworthy basis for rankings.

Building on this understanding of the real world, the 2026 Street Sweeping List refines this judgment further. It now distinguishes between purpose-visits and drop-bys, as well as returning customers versus new ones, making it easier for genuinely good stores to feature prominently. For professional consumption fields like coffee shops, the list also introduces expert ratings, giving greater weight to the choices made by professionals.

If rankings are a concentrated presentation of real decisions, Amap's spatial intelligence must also enter more aspects of daily life. For instance, after a user selects a destination, another question arises: can one preview the actual place before departure? Flight Street View 2.0, powered by Amap's ABot-Earth 0.7 world model, transforms the destination into an explorable three-dimensional space, allowing users to freely adjust viewing positions, angles, and heights to achieve a sense of presence before even leaving. The "Pitfall Avoidance Guide" (避雷指南) tackles a different task: simulating your entire trip in advance. The same itinerary can yield completely different results on a Tuesday afternoon versus a Saturday night, and variables like rain or wind also affect the experience. The guide checks 15 dimensions—including business hours, itinerary pacing, parking tips, crowding predictions, and weather changes—to preemptively identify risks like timing conflicts, congestion, and store closures.

During travel and upon arrival, Navigation Live (导航Live) not only understands the user's language but also perceives their real-time location, direction, and surroundings. By using the camera to comprehend street scenes, it provides highly relevant, context-aware suggestions. Unlike traditional navigation tools, Navigation Live can see the world before your eyes and even understand spontaneous intentions. For example, in a complex old neighborhood of an unfamiliar city, if three nearly identical narrow alleys appear ahead, traditional navigation would only say "turn right in 50 meters," leading to a high chance of error. Navigation Live, however, can directly interpret the scene: "Drive into the alley to the right of the convenience store with red lanterns, avoiding the van unloading on the left." If you suddenly change your mind and say, "Forget it, I don't want to go in. Is there any local cuisine nearby without a long queue?" it can instantly combine its understanding of your current spatial position, road turning conditions, and real-time queue data to rewrite your action plan. This ability to "see the scene, understand intent, and act on the spot" makes Navigation Live more akin to an embodied intelligent agent with "spatiotemporal context, on-site perception, and end-to-end action capability."

From the Street Sweeping List providing authentic consumption decisions, to Flight Street View letting users preview destinations, to the Pitfall Avoidance Guide pre-simulating journeys, and Navigation Live accompanying users into the real world, all of these are supported by Amap's spatial intelligence. For Amap, spatial intelligence is not merely an additional technological feature for mapping and navigation; it represents a fundamental shift in its underlying capabilities, thereby expanding the boundaries of how Amap serves the real world.

An Underappreciated Capability Foundation in the AI Era

As AI competition shifts from content generation to understanding and acting upon the physical world, spatial intelligence has become a strategic focus for global tech companies. NVIDIA, Tesla, and Google are each approaching this field from different angles, such as digital twins, autonomous driving, and geographic information systems. What sets Amap's spatial intelligence apart is that it does not start from virtual simulation but is built on a long-term accumulation of real travel behaviors, covering the entire process from user decision-making to action.

Today, spatial intelligence is no longer confined to Amap's own app; it is being embedded as a foundational capability in various hardware terminals, including in-vehicle systems, smartwatches, and embodied robots. This trend has accelerated notably since last year. From the Qianwen AI glasses partnering with Amap to bring spatial intelligence directly to AR glasses, to Amap's announcement on September 10 that Mercedes-Benz became the first global automotive partner for its Street Sweeping List, Amap's spatial intelligence is emerging as a crucial gateway for AI to enter the real world.

This also means that our valuation framework for Amap needs a complete overhaul. In the old coordinate system, it was a mapping company; in the new value framework, it is a spatial intelligence company—the underlying system that enables machines to understand the real world. When Fei-Fei Li states that spatial intelligence is AI's next frontier and the entire industry seeks the ability to "comprehend the real world," people suddenly realize that Amap has been on this path all along. For this reason, in the AI age, Amap's value coordinate system must indeed be recalibrated.