Think Tank Affiliated with Xinhua News Agency: Can Tesla's FSD Succeed in China?
This week, a major news story about Tesla was its long-awaited update to its autonomous driving software in China on Tuesday, adding urban navigation capabilities. This has been enthusiastically hailed by many Chinese Tesla owners as Tesla's FSD finally "entering China."
According to a notification on Tesla's app, the new features include automatic lane changes based on speed and route, as well as traffic light detection at intersections to determine whether to turn. The in-car camera will also monitor the driver’s attention.
However, this move disappointed Chinese Tesla owners, who stated that the feature did not live up to Elon Musk's promises. Many Chinese social media users expressed frustration, saying that the update failed to deliver the features Tesla had been promoting for years. Some even pointed out that competitors like BYD offer similar driver assistance functions at a lower price—or even for free. One Tesla owner in China said he paid RMB 56,000 ($7,800) for the FSD software when purchasing a long-range Model 3 in 2019, only to feel frustrated by the lack of functional updates.
Due to insufficient training on Chinese roads and inadequate adaptation to local traffic regulations, these features are less advanced than those available in the United States. Nonetheless, these functions are similar to Tesla's "Full Self-Driving" (FSD) system, though FSD is a more advanced driver-assistance technology developed using generative AI, capable of handling more complex traffic conditions, whereas the current autopilot features mainly assist with standard driving scenarios.
"It’s clear that Tesla had no choice but to release a system with known limitations... Tesla is struggling to catch up with the intelligent driving capabilities of Chinese automakers, making its high pricing even less justified."— A Tesla owner interviewed in China
The delayed launch of a full FSD system in China has left Tesla lagging behind domestic competitors such as Huawei, Xiaomi, and BYD, which have already introduced dozens of electric vehicles equipped with advanced intelligent driving software, capable of handling China’s complex urban traffic conditions.
While Tesla charges customers nearly $9,000 for a limited version of FSD, Xiaomi's SU7 electric sedan offers a highly competitive alternative. The SU7 has already surpassed the Tesla Model 3 in monthly sales in China and provides an enhanced intelligent driving system, including urban navigation, free of charge.
Earlier this month, BYD also began offering advanced autonomous driving functions for free on most of its models.
Tesla’s latest autonomous driving update is roughly equivalent to BYD’s mid-tier "God’s Eye" system, which uses advanced computing and LiDAR to generate a 3D representation of the vehicle’s surroundings, helping it avoid obstacles. This system is available on BYD's premium Denza brand, as well as higher-end models from its Dynasty and Ocean series.
Musk has previously stated that Tesla is seeking regulatory approval to fully introduce FSD in China.
According to industry standards, both autonomous driving and FSD are classified as Level 2 autonomous driving technologies, meaning they require human supervision and intervention when necessary. Automakers are only required to seek regulatory approval for Level 3 and above autonomous driving technologies.
Currently, China only requires automakers to register Level 2 updates with industry regulators, meaning Tesla's FSD does not yet require full government approval.
Musk had previously hoped to transfer Tesla's China-collected driving data overseas for training Tesla’s autonomous driving AI. However, cross-border data transfers are not that simple. Now, Musk and his engineering team are exploring alternative solutions, one of which is synthetic data. Tesla has reportedly been using publicly available Chinese street videos from the internet to train its AI software. Additionally, Tesla is planning to build data centers in China to train the algorithms necessary for fully autonomous driving.
A day before Tesla announced its FSD software update, Liaowang Institute—a Think Tank Affiliated with Xinhua News Agency—published a commentary on the challenges facing Tesla’s FSD in China, highlighting that its prospects remain highly uncertain.
Can Tesla's FSD Succeed in China?
At noon on the 18th Beijing time, Elon Musk live-streamed the release of Grok 3, the latest chatbot developed by his AI company, xAI, on the X platform. Musk revealed that Tesla users will soon be able to use Grok 3 while driving.
As AI models like DeepSeek are integrated into vehicles, AI is set to become the core task of smart cars in the coming years. Just as Tesla led the global wave of electrification over the past few years, the new battleground is AI—whoever dominates AI will dictate the fate of the automotive industry.
Tesla's challenges in 2024 have been evident. For the first time in a decade, its global sales have declined year-over-year. In China, on the first working day after the Lunar New Year holiday, Tesla China responded with major price cuts, signaling fierce market competition. Chris McNally, an analyst at Evercore ISI, bluntly stated that Tesla’s current valuation has severely overestimated the market's expectations for its AI business. Meanwhile, Nicholas Colas, co-founder of DataTrek Research, pointed out that over 90% of Tesla’s market value is based on expectations for future development, which remains highly uncertain. As a publicly traded company navigating the rapidly evolving EV market, Tesla must now reinvent its AI narrative both in China and globally.
Grok 3 inevitably reminds the public of Tesla’s Full Self-Driving (FSD) system, which has long been touted as the company’s flagship AI product. Over the years, Musk has repeatedly promised the imminent realization of FSD, yet has failed to deliver. While a regulatory-compliant trial version of FSD has been launched in the U.S., its adoption rate remains unclear.
Last May, Musk made a highly condensed 44-hour visit to China, which the industry widely speculated was aimed at paving the way for FSD’s entry into the Chinese market. Tesla’s global deployment roadmap for 2025 Q1 includes plans to introduce FSD in China and Europe. The company has also built a data center in China and complied with four key automotive data security requirements, seemingly signaling that FSD is getting closer to entering the Chinese market. However, recent foreign media reports suggest that Tesla’s FSD launch in China may be delayed. The bigger question remains: Can Tesla’s AI narrative truly resonate in China?
Data security is the first major hurdle Tesla must overcome. Typically, an autonomous driving system consists of three components: perception (gathering real-time environment data via cameras, radars, or LiDAR), planning (using this data to create a driving route), and execution (carrying out real-time driving decisions). Tesla’s FSD employs a "Neural Network Path Planning" approach, which differs from traditional rule-based machine learning. The former trains AI by mimicking human drivers' behavior, rather than simply teaching predefined driving rules such as "stop at red lights" or "stay in the center of the lane."
According to Dhawal Shroff, head of Tesla's AI team, their training model is based on massive amounts of real-world driving data collected from Tesla vehicles. Tesla FSD requires at least 1 million video clips to reach an acceptable performance level, and 1.5 million clips would be even better. With over 7 million Tesla vehicles worldwide (including 1.7 million in China), the company has a strong data foundation for AI training. However, in China, Tesla faces two major challenges: Will Chinese Tesla owners willingly share their data? In April 2024, China’s auto industry regulator mandated strict rules on in-car data collection, ensuring compliance with privacy standards. Tesla’s vehicles are set to "default to not collecting in-cabin data" unless explicitly enabled by the driver. Thus, Tesla cannot freely collect in-car data unless drivers voluntarily opt in.
At the same time, Tesla’s AI training overlaps with regulated mapping activities. In China, mapping is strictly regulated due to national security concerns. While Tesla claims it does not require high-definition maps, its data collection and video-based AI training could still be classified as mapping activity, subjecting it to strict regulatory scrutiny. Without sufficient Chinese driving data, Tesla’s FSD may struggle to "learn" how to drive in China's complex road conditions.
Even if Tesla legally collects and stores data in China, AI training must be conducted locally, requiring high-performance GPU chips. Tesla's U.S. "Cortex" supercomputing cluster is powered by over 100,000 NVIDIA H100 and H200 chips. These chips are now subject to U.S. export bans on AI processors. In October 2023, the U.S. government restricted exports of NVIDIA’s A100, H100, H200, and L40S chips to China. In January 2024, the Biden administration introduced the "AI Diffusion Interim Final Rule," further tightening AI hardware exports to China.
Tesla has developed its own AI chip, the D1, which powers its Dojo supercomputer—the core training platform for FSD. The D1 chip is manufactured by TSMC using a 7nm process. While Tesla aims to replace some NVIDIA hardware with Dojo, it remains heavily dependent on NVIDIA. Musk’s position on in-house chips has been inconsistent: In 2021, he emphasized Tesla's goal of achieving full AI hardware self-sufficiency. By mid-2023, he publicly stated: "If NVIDIA could provide us with enough GPUs, we might not need Dojo." This raised doubts about Tesla’s long-term AI chip strategy. Without a viable alternative, Tesla’s ability to train AI models in China could be severely limited.
The third challenge Tesla faces is safety. Some early FSD beta users have reported concerning behavior. A CNBC interview with a beta user revealed that Tesla's FSD made unsafe lane changes, cutting off other vehicles. Tesla’s machine learning approach mimics human driving, but if it learns aggressive driving habits, it may fail to meet safety standards. Unlike other automakers, Tesla relies solely on cameras for FSD. Most companies use a combination of sensors: cameras, LiDAR, millimeter-wave radar, and ultrasonic sensors, to ensure redundancy in decision-making.
A purely vision-based system may struggle with low visibility conditions. Tesla has faced multiple U.S. investigations over FSD-related crashes, including cases where its vision system failed to recognize obstacles. Given China’s stricter automotive safety regulations, Tesla may need to modify FSD’s approach before launching it in the country.
Tesla’s AI ambitions in China are far from guaranteed. Regulatory compliance for data collection remains unclear. Chip supply constraints may limit Tesla’s AI capabilities. Safety concerns over Tesla’s AI-driven autonomy remain unresolved. As Tesla seeks to define the future of AI-powered vehicles, it must overcome these challenges to prove its AI story can succeed in China.