Cars to become large-scale AI agents, says Horizon exec

AI technology is moving from supporting role to the very core of the automotive industry, ushering in the era of "AI-defined cars", said Horizon Robotics President Chen Liming.

"The industry has shifted from hardware-defined to software-defined, and now we are entering an AI-defined era," Chen said at the World New Energy Vehicle Congress held in Haikou, Hainan province, on Sunday.

He said in this next phase, intelligent cockpits and autonomous driving will no longer develop separately.

Instead, the focus will be on deeper integration—what he described as "cabin–driving fusion"—and, ultimately, on vehicles evolving into large-scale AI agents.

"Cars will become the first mass-deployed AI devices operating in the physical world," he said.

Chen highlighted three areas where AI is reshaping the sector. First, AI-driven development is replacing the traditional software 1.0 paradigm, forcing companies to rebuild quality standards and processes around data-centric software 2.0.

Second, breakthroughs in model training and inference are improving the performance, generalization and safety of intelligent driving, pushing the industry closer to Level 3 and Level 4 autonomy.

Third, industry relationships are shifting from a linear supply chain to a dense network of deep strategic partnerships that span capital, technology and product development.

These advances are driving new demands for computing power. "As algorithms evolve and models grow larger, relying on a single vehicle or a single chip is no longer enough," Chen said.

He sees tighter coordination between vehicles, infrastructure and the cloud as essential, with cloud-based training and inference complemented by low-latency decision-making at the edge. That, in turn, raises the bar for both cloud and on-board chips in terms of performance and energy efficiency.

On strategy, Chen said carmakers face a balance between full-stack in-house development and collaboration with ecosystem partners.

Early in a technology cycle, companies with the capability often prefer full-stack R&D to control product innovation.

But as technologies mature and scale becomes critical, he expects most automakers to shift from "full-stack self-research" to what he called "full-stack controllability", relying on partners for cost and performance advantages.

"Roughly 20 percent of OEMs may continue full-stack development, while the majority will leverage ecosystem scale," he said.

Safety, however, remains the non-negotiable baseline. "Safety is the '1'; every other function is a zero that follows," Chen said. "You can never trade safety for cost. All cost optimization must be built on a foundation of safety."