DeepSeek CEO Outlines Path to Level 5 Autonomy: Data-Driven AI and Safety Standards Lead the Way

Maxwell Zhou, CEO and Founder of DeepSeek.ai, one of China's most advanced providers of smart driving solutions, recently shared his insights on the path to fully autonomous driving during the BEYOND Expo 2025 in Macao. Zhou defined full autonomous driving as reaching Levels 4 and 5, where cars can navigate without human intervention, regardless of location or environmental conditions. DeepSeek's Approach and Business Model DeepSeek's business model focuses on providing energy and entertainment AI solutions to OEMs (Original Equipment Manufacturers), enabling them to enhance their vehicle offerings. The company has already delivered over 50,000 cars with its AI technology to customers since August 2024, and expects to surpass 200,000 deliveries this year, expanding to over ten car models. Zhou envisions DeepSeek becoming the "physical agent AI" for cars, aiming to launch fully autonomous driving fleets using innovative data-driven technologies. Unlike many competitors that heavily rely on HD maps, DeepSeek's approach is map-less. This technology allows their cars to operate nationwide, addressing the challenge of maintaining accurate HD maps across vast regions. Zhou highlighted that even with the limited use of HD maps, human drivers often still require navigation aids, emphasizing the practicality and scalability of DeepSeek's data-driven approach. Technological Breakthroughs and Future Strategies Zhou discussed the rapid pace of technological change in the industry, particularly the shift from map-dependent to data-driven and reasoning-based systems. DeepSeek is incorporating language reasoning into its autonomous driving technology, leveraging pre-training and post-training methods similar to those used in advanced language models like GPT-4. This ensures that their AI systems are not only intelligent but also aligned with safety standards, driving with human-like decision-making capabilities. Commercial Viability and User Acceptance To make autonomous driving commercially viable, Zhou emphasized two key factors: cost control and user acceptance. Over the past five years, DeepSeek has reduced the cost of local driving technology from $100,000 to a few thousand dollars, making it feasible for widespread adoption. Zhou also stressed the importance of ensuring high performance to gain user trust, noting that a subpar product can deter consumers. DeepSeek's data-driven intelligence approach aims to achieve both cost efficiency and robust performance, driving continuous improvement through extensive data collection. Safety and Regulation Safety remains paramount in the development of autonomous vehicles. Zhou acknowledged the public's concerns and emphasized the role of strong regulations in fostering a safe and competitive environment. DeepSeek adheres to the highest safety standards, utilizing advanced post-training techniques like Proximal Policy Optimization (PPO) to enhance the reliability and safety of its AI systems. According to Zhou, the company achieves one disengagement per 100 kilometers in urban areas and one per 1,000 kilometers on highways, which is currently classified as Level 2 autonomy. He believes that transparency in system capabilities and strict adherence to regulatory guidelines are essential to build consumer confidence. International Expansion Despite originating in China, DeepSeek aims to expand its reach internationally. The company has already established an office in Europe and is testing its technology in various overseas markets. Zhou mentioned the importance of complying with local regulations and laws, particularly regarding data sensitivity and privacy. DeepSeek's map-less, data-driven approach enables easier international deployment, and the company is currently targeting European markets due to the concentration of car manufacturers in the region. Competition and Market Evolution Zhou observed that the autonomous driving landscape has become more concentrated over the past few years. While there were over 50 companies three years ago, only a handful remain today. This consolidation is partly due to strong regulations that ensure a higher technological baseline. Zhou believes that companies prioritizing safety and innovation, like DeepSeek, are better positioned to succeed in this market. He anticipates continued advancements in AI, particularly in transformer-based technologies, will drive the industry forward. Vision for Level 5 Autonomy Looking ahead, Zhou is optimistic about achieving Level 5 autonomy, the pinnacle of autonomous driving, within the next five years. The recent breakthroughs in large language models and their reasoning capabilities, coupled with advances in pre-training, post-training, and safety alignment, are bringing this vision closer to reality. DeepSeek aims to contribute significantly to this journey, leveraging its expertise in AI to develop smarter and safer autonomous vehicles. Industry Insiders' Evaluation Experts in the tech and automotive industries have praised DeepSeek's innovative approach and rapid progress. The company's focus on safety, cost control, and scalability sets it apart from competitors. DeepSeek's map-less, data-driven technology is seen as a game-changer, particularly in markets where maintaining HD maps is impractical. Industry insiders believe that DeepSeek's international expansion, adherence to regulatory standards, and collaboration with OEMs position it well to play a crucial role in the realization of fully autonomous driving. Company Profile DeepSeek.ai, founded by Maxwell Zhou, has quickly risen to prominence in the autonomous driving sector. With a Ph.D. in AI and over two decades of experience in robotics, Zhou has led the company to achieve unicorn status within two years. DeepSeek's pioneering technology and strategic business model have attracted significant attention and partnerships, solidifying its position as a leader in the race toward autonomous driving.