NextFin News - The rapid evolution of large language models (LLMs) that has redefined the tech landscape over the past year is failing to translate into a faster arrival for autonomous heavy-duty trucks on public roads. Despite the global frenzy surrounding generative AI, leaders of China’s self-driving industry are signaling that the "intelligence" powering chatbots like ChatGPT or DeepSeek is fundamentally disconnected from the physical requirements of navigating a 40-ton vehicle through highway traffic.
James Peng, CEO of Pony.ai, dismissed the notion that linguistic breakthroughs accelerate vehicle deployment during a press briefing last week. Peng, a former Google engineer who has spent nearly a decade positioning Pony.ai as a frontrunner in both robotaxis and trucking, argued that the skills required for language processing and physical driving are non-transferable. "The world’s best linguistics expert doesn’t mean he’s a good driver," Peng stated, emphasizing that the relevance of LLM advances to autonomous driving is "absolutely zero."
This cautious stance is echoed by Julian Ma, CEO of Inceptio Technology, a startup focused exclusively on long-haul autonomous trucking. Ma, who founded Inceptio in 2018 with a focus on the commercial viability of "Level 3" and "Level 4" systems, confirmed that his company is sticking to its original timeline for a mid-2028 commercialization milestone. While Inceptio has already deployed trucks with human "safety monitors" across thousands of kilometers of Chinese highways, the leap to fully driverless operations remains tethered to a rigid data-collection schedule rather than software breakthroughs in other AI sectors.
The divergence between "cyber AI" and "physical AI" stems from the type of data required. While LLMs are trained on the vast corpus of human text, autonomous driving requires "world models"—simulations that understand the physics of the real world. Ma noted that Inceptio expects to reach 5 billion kilometers of real-world truck driving data by late 2028. Only then, he argues, can AI extrapolate that experience into the 50 billion kilometers of simulated data necessary to safely remove the human driver from the cab. This data-heavy approach suggests that the bottleneck for autonomous trucking is not a lack of "intelligence," but a lack of high-quality, real-world exposure.
The skepticism from Chinese leaders contrasts sharply with the more aggressive rhetoric often seen in the United States. Tesla CEO Elon Musk has frequently linked the success of "Full Self-Driving" (FSD) to the company’s broader AI and compute capabilities, recently suggesting that unsupervised FSD could arrive as early as late 2026. However, Musk’s history of shifting deadlines—including a January 2026 admission that Tesla needs 10 billion miles of data for safe deployment—aligns more closely with the data-centric caution expressed by Peng and Ma than his optimistic public forecasts might suggest.
The stakes for China’s logistics sector are immense. With a shortage of long-haul drivers and rising labor costs, the commercialization of autonomous trucks could fundamentally alter the economics of the world’s largest freight market. Yet, the insistence on a 2028 horizon by industry leaders suggests that the "AI summer" will not provide a shortcut. For now, the path to driverless freight remains a slow, kilometer-by-kilometer grind, unaffected by the ability of machines to write poetry or code.
Explore more exclusive insights at nextfin.ai.
