China’s autonomous driving sector is undergoing a strategic recalibration, shifting from a resource-heavy 'arms race' of raw computing power toward high-efficiency, algorithm-led solutions. For years, the industry was defined by a rush to stack expensive sensors and high-performance chips, often measured in thousands of TOPS (Tera Operations Per Second). However, a new wave of 'disruptors' is proving that sophisticated urban navigation is possible with far less hardware redundancy than previously thought.
At the forefront of this shift is Yu Qian, CEO of QCraft (Qingzhou Zhihang), who argues that the industry’s 'second half' will be won by those who can squeeze maximum performance out of modest hardware. His company recently demonstrated full-scenario urban navigation using only 128 TOPS of computing power, a feat that challenges the prevailing wisdom that Level 4 autonomy requires massive, power-hungry chipsets. This approach emphasizes 'algorithmic minimalism' over 'hardware redundancy,' aiming to make advanced smart driving features accessible to the mass market.
The transition to domestic silicon is another critical pillar of this evolution. As geopolitical tensions complicate the supply of high-end international chips, Chinese firms are increasingly turning to home-grown alternatives. Yu emphasizes that the success of domestic chips cannot rely on consumer patriotism alone; rather, these components must deliver an 'over-the-horizon' experience that matches or exceeds global benchmarks to win genuine market share.
Technologically, the focus is moving toward integrated 'world models' and reinforcement learning, moving away from the rigid, rules-based systems of the past. This evolution allows vehicles to navigate complex urban environments with more 'human-like' intuition. By focusing on 'inclusive' smart driving that bridges the gap between electric and traditional internal combustion vehicles, the industry aims to enter an era where high-level autonomy is no longer a luxury feature but a standard safety expectation.