(Image Source: Lin Zhijia)

AsianFin— As China's race toward full autonomy intensifies, DEEPROUTE.AI CEO Zhou Guang took a hard stance on the limitations of Level 3 autonomous driving.

Zhou called it the "lower bound" of intelligent driving, and warning that prioritizing safety at the cost of usability could undermine consumer adoption.

Speaking in an extended post-event interview following DEEPROUTE.AI's major announcement at the 2025 Shanghai Auto Show, Zhou described Level 3 (L3) autonomy — which requires human oversight — as inadequate if it can't deliver a meaningful improvement in driving experience.

"If you're stuck in traffic for an hour and only move 50 meters, even if the system is extremely safe, no one will want to use it," Zhou said. "You can't blindly chase safety at the cost of intelligence. That defeats the purpose of autonomous driving."

Zhou argued that smart driving systems must strike a balance between safety and intelligence. "Imagine turning on L3 so you can take a nap, but after an hour, the car hasn't gone anywhere. You'll lose confidence in the system," he said. "If the car just stops to ensure absolute safety, that's an extreme scenario — and one that ultimately fails the user."

DEEPROUTE.AI announced a strategic partnership with ByteDance's Volcano Engine to jointly develop AI-powered vehicles with Chinese automakers. Volcano Engine will provide computing infrastructure to accelerate training and deployment of DEEPROUTE.AI's VLA (Vehicle-Level Agent) model. The company expects over five VLA-equipped smart vehicles to enter mass production this year.

Founded in 2019, DEEPROUTE.AI was one of the first Chinese startups to bring AI into the autonomous driving space. The company collects large volumes of real-world, desensitized data to develop an AI "brain" with advanced human-like cognition. Its focus includes "No-Map" solutions and end-to-end deep learning models, with the ultimate goal of creating artificial general intelligence (AGI) for the physical world.

Zhou, a graduate of Tsinghua University, previously worked at Texas Instruments and Baidu's U.S. autonomous driving lab before returning to China to launch DEEPROUTE.AI.

The company now has over 800 employees — more than 80% in R&D — and has raised over $500 million across six funding rounds. Investors include Alibaba, Fosun RZ Capital, Yunqi Partners, and Lightspeed China. In its latest funding round in November 2024, it secured a $100 million strategic investment from Great Wall Motors.

Looking ahead, Zhou said DEEPROUTE.AI will focus on three key priorities: scaling mass production, advancing its VLA model, and expanding its Robotaxi footprint.

Zhou also singled out Tesla's FSD V13 as the benchmark for intelligent driving globally. "The system predicts your predictions," he said, arguing that such anticipatory intelligence is more important than traditional metrics like MPI (miles per intervention). "A strong autonomous system must be trained with takeover intentions in mind. If it's filled with ineffective actions, users won't want it — no matter how safe it claims to be."

Still, Zhou acknowledged the importance of explainability in AI driving systems. As VLA models evolve to include natural language capabilities, they'll be better able to articulate decisions, which will improve user trust and perceived safety. "Given recent high-profile incidents, I believe safety must remain the number one priority. Scaling without safety is reckless and unacceptable," he said.

Zhou also offered a candid view on China's increasingly crowded autonomous driving market. "There are already 20 to 30 companies competing, many of which started with just a few million RMB. But very few will survive," he told AsianFin. "We're entering the second half of the EV race. The first half was electrification. The next five years will be all about intelligent driving."

For Zhou, the endgame remains Level 5 (L5) fully autonomous driving. "That's the ultimate goal for every company in this space — but there are different paths to get there. The system must eventually serve the vehicle holistically. That's the only way to reach true autonomy."

As Moore's Law and Huang's Law show signs of slowing chip advancements, Zhou believes foundational AI models will play a greater role in the next stage of industry evolution. "As the tech matures, AI will become more grounded and widely applied — not just in driving, but in platforms like MovingAgent and FastAgent," he said.

更多精彩内容，关注钛媒体微信号（ID：taimeiti），或者下载钛媒体App