Published June 25, 2026 · By Apick Lion
You're driving through Beijing. First time in China. You pull up to an intersection, red light ahead. You glance at your phone — Amap, the local navigation app — and it tells you exactly how many seconds until the light turns green.
Twelve seconds. Eleven. Ten.
You look up at the traffic light. There's no countdown display. No visible sensor. No obvious connection between the app and the infrastructure.
How did it know?
This is not a demo. This is not a pilot. This is a feature that 300 cities in China use every day — over 2 billion times a day, according to Amap's public disclosures. And it's something Google Maps — with all its engineering talent, all its data, all its resources — has not yet deployed at scale.
Not because Google isn't smart enough. Because the problem isn't about intelligence. It's about something else entirely.
Amap (Gaode, in Chinese) launched its traffic light countdown feature in May 2022. It was simple at first: when you approached an intersection, the app showed you a countdown timer for the red light. No hardware. No connection to the traffic signal system. Just a number on your phone that matched reality with surprising accuracy.
By 2025, the feature had evolved into something much more sophisticated. The "Red Light AI Pilot" added start reminders, lane recommendations, green-light passage guidance, and even predictions for multi-cycle waits during heavy traffic. It covers every traffic-light scenario in urban driving.
Today, Amap's intelligent traffic light system provides countdown services more than 2 billion times per day. It covers nearly 500,000 traffic intersections across mainland China, Hong Kong, Macau, and Taiwan. The system calls on Beidou satellite positioning over 450 billion times per day — figures sourced from Amap's official disclosures reported by Beijing News, People's Daily Online, and Economic Reference News (Xinhua).
2 billion countdowns. Every day. That's not a prototype. That's infrastructure.
Here's what most people assume: Amap must be connected to the traffic light system. It reads the signal timing data directly from the city's traffic management center.
That assumption is wrong.
Amap's countdown feature does not rely on signal hardware integration. It doesn't plug into traffic management systems. It doesn't install sensors at intersections. It uses AI to predict — not read — traffic light timing.
The math works like this:
Every driver running Amap generates a continuous stream of position and speed data. When thousands of cars stop at the same intersection, the system detects a pattern: cars decelerate, stop, wait, then accelerate. By analyzing the distribution of start times across thousands of vehicles over multiple cycles, the system infers the underlying signal timing.
It's essentially reverse-engineering the traffic light cycle from driver behavior alone.
The system uses AI models to process this fragmented, noisy data. It accounts for time of day, traffic volume, and historical patterns. When traffic surges in one direction, the system can even predict signal phase adjustments up to 30 seconds in advance. The time error is reportedly kept within a fraction of a second.
Amap further refined the technology using a visual spatio-temporal modeling approach that incorporates timing perception, allowing the system to directly observe intersection dynamics in real time. Amap began patenting related technology as early as April 2019.
The result: a system that simulates and predicts traffic light timing at nearly half a million intersections, with no hardware investment, no government data integration, and no ongoing infrastructure maintenance.
Traffic light countdown prediction is widely regarded as a technically difficult challenge for global navigation services. The difficulty isn't just technical — it's structural.
Google Maps has immense data. It has billions of Android devices generating location data. It has world-class AI talent. So why doesn't it have this feature?
Data density. Amap's system works because China has an extraordinary density of navigation users. Hundreds of millions of drivers using the app simultaneously, generating enough data at every intersection to infer signal timing with statistical confidence. In the U.S., the density of navigation users per intersection is significantly lower. The math becomes significantly less reliable with sparse data.
Traffic signal fragmentation. Traffic light systems in the U.S. and Europe are fragmented across thousands of municipalities, each with different vendors, different protocols, and different levels of data openness. There is no unified standard. Amap built its system on top of a relatively consistent national traffic infrastructure — not because it was connected, but because the underlying patterns were predictable enough to model.
The hardware alternative. The prevailing approach to "smart traffic" in many Western markets has been hardware-first: install sensors, cameras, and roadside units; build a connected infrastructure; then deliver services on top. This approach requires significant capital investment and is typically limited to cities that can afford the deployment. Amap's approach — pure software, pure data, zero hardware — can be deployed more broadly in markets with sufficient user density.
The two approaches reflect different market conditions: the hardware-first model requires upfront infrastructure investment, while the data-driven model can scale with user adoption without hardware deployment.
The Amap traffic light feature is not a breakthrough in model architecture. It's not a new training technique. It's not a larger parameter count.
It's a breakthrough in data engineering — taking noisy, unstructured data from millions of everyday users and turning it into a service that improves millions of other users' daily experience.
This is what Chinese AI deployment looks like at scale. It's not about building the biggest model. It's about solving the most everyday problems with the data you already have.
Amap didn't wait for perfect infrastructure. It didn't wait for government data-sharing agreements. It didn't wait for hardware vendors to install sensors. It used what was already there — millions of phones, millions of trips, millions of data points — and built a service that works today, at scale, for free.
That's not a technical advantage. It's a deployment philosophy advantage.
Markets with high navigation-user density can pursue a data-first approach and use what exists. Markets with fragmented infrastructure often default to a hardware-first model.
One approach can build broad coverage quickly in dense markets. The other creates targeted deployments in specific cities.
2 billion countdowns per day.
500,000 intersections covered.
450 billion satellite positioning calls per day.
These numbers are not marketing. They are production metrics from a system that has been running for years. And they represent something that no amount of AI model improvement can replicate: real-world data density.
You cannot train your way to this. You cannot fine-tune your way to this. You can only deploy your way to this — by getting a product in front of users, collecting data, improving the service, and repeating the cycle.
Google Maps has been mapping roads for two decades. Amap has been mapping behavior for four years — and turned that behavioral data into a real-time traffic light countdown service used across hundreds of cities.
The next time someone tells you that AI is about models and benchmarks, tell them about the 2 billion countdowns.
That's what AI looks like when it actually works.
This article is an editorial analysis comparing product features under different regional market conditions. It does not constitute a comparative assessment of overall product quality, capability, or corporate performance. Feature availability depends on regional data density, infrastructure conditions, and market-specific factors.
Data sourced from Amap (Gaode) official disclosures as reported by Beijing News, 163.com, People's Daily Online, Upstream News (cqcb.com), and Economic Reference News (Xinhua News Agency), 2022–2026. Figures for daily countdown services, Beidou positioning calls, and intersection coverage are drawn from Amap's public disclosures as reported in the above outlets.