Hyundai bets on Nvidia’s Blackwell to go wheel to wheel with Tesla

[NEWS ANALYSIS]

Hyundai Motor’s allotment of 50,000 Nvidia’s Blackwell GPUs may prove to be a long-awaited inflection point for the Korean automaker — one that could finally propel its ambitions in autonomous driving and robotics.

Beyond the GPU supply agreement, Hyundai’s collaboration with the Nasdaq-listed tech firm stretches to a $3 billion investment in the creation of an AI-driven data center that could serve as the backbone for a vertically integrated ecosystem like that of rival Tesla’s, where in-house computing power underpins the development of both autonomous driving systems and humanoid robotics.

Each of the 50,000 GPUs is capable of delivering 1.96 teraflops of computing power, roughly 1.96 trillion operations per second. Meritz Securities estimated that the total computational capacity of the 50,000 GPUs will be nearly 100 quintillion calculations per second based on simple math.

Breaking free: Independence in driverless driving

Hyundai’s push to secure a massive number of GPUs and build its own AI data centers signals its intention to develop its own self-driving technology while making a decisive break from its longstanding reliance on third-party algorithms developed by partners like Mobileye and Motional.

For years, growing concerns over data security and technological dependence have left Hyundai lagging behind Tesla, which continues to widen its lead in real-world performance and data-driven iteration.

While Tesla’s Autopilot has already reached what many analysts classify as Level 2-plus autonomy — an advanced form of partial driving automation — Hyundai’s systems remain largely confined to driver-assistance features such as lane-keeping and adaptive cruise control. Hyundai set a goal of mass-producing Level 2-plus cars by 2027. In the 2024 autonomous tech ranking by Guidehouse Insights, Motional, a U.S. self-driving startup 90 percent owned by Hyundai, ranked only 15th globally, a sharp drop from fifth the previous year.

At the heart of the gap lies data. Training AI-driven autonomous driving systems requires vast amounts of real-world driving data, and Tesla, for instance, collects road data from over 5 million vehicles worldwide, continuously feeding it into its proprietary AI supercomputer, Dojo, which operates around the clock to refine its algorithms.

Without such real-world driving data, Hyundai will rely on its Nvidia partnership to compensate for the scarcity by generating high-fidelity synthetic environments, enabling accelerated testing and refinement of autonomous-driving algorithms using Nvidia’s Omniverse and Cosmos virtual simulation platforms, one of two components the automaker’s Blackwell-based AI factory will be built around.

The high-performance GPUs that can handle an immense computational load allow Hyundai to build an independent system capable of ingesting and learning from such massive volumes of sensor data, road conditions, and driving footage for real-time processing of data streams from vehicle sensors, high-definition maps and AI decision-making modules.

The factory will also tap Nvidia’s DGX, which is designed to provide the immense computational power necessary to train the carmaker’s proprietary autonomous-driving models.

“Very few companies today are building autonomous-driving capabilities entirely in-house. Maybe Tesla is essentially the only player doing so for now, while most other leading firms still rely on third-party algorithms,” said automotive engineering Prof. Park Chul-wan of Seojeong University.

“This is where Hyundai hopes to squeeze in. Their long-term vision is to become a mobility company that not only produces cars but also humanoid robots.”

Challenging Tesla’s Optimus

If advanced driverless driving represents the first step in physical AI, the next frontier is humanoid robots.

Although Hyundai is the largest shareholder in Boston Dynamics, a Massachusetts-based humanoid robot maker, it does not have full ownership or control of the U.S. company’s Atlas humanoid robot.

Hyundai plans to leverage Nvidia’s Isaac Sim, a next-generation robotics development platform built on Omniverse, to extend its AI systems to humanoids and robotic applications. This allows the company to simulate tasks and plan robot actions in virtual environments before deploying robots on production lines.

Atlas units from Boston Dynamics can also be incorporated as training assets during the initial stages of development.

Nvidia’s Drive AGX Thor will serve as the AI “brain” for both vehicles and robots, providing the computational power needed for real-time decision-making. Hyundai is also planning to build a robotics factory in the United States with an annual capacity of around 30,000 units.

“Hyundai’s partnership with Nvidia goes far beyond GPU procurement. The collaboration is designed to establish a continuous data pipeline from vehicle operations, which will feed into AI training at Hyundai’s data center,” said Kim Joon-sung, an analyst at Meritz Securities.

“By 2028, it is highly likely that Hyundai’s smart cars will feature high-performance inference computers developed in collaboration with Nvidia, and this cooperation will naturally extend from autonomous driving to general-purpose humanoid robotics.”

From chips to cloud

Other Korean big tech firms including Samsung and SK have also secured Nvidia chips to advance their AI technology.

Samsung Electronics secured 50,000 GPUs, committing to building an AI-driven chip factory capable of managing the entire process from chip design to production. Similarly, SK Group plans to deploy over 50,000 GPUs in an AI factory designed to function as a comprehensive industrial AI service provider, encompassing digital twins, robotics, large language models and 3-D simulation capabilities for both training and inference.

SK Telecom will conduct AI network research jointly with Nvidia, targeting the development of artificial intelligence in radio access network technology, or AI-RAN, a cornerstone for 6G mobile communications.

Naver Cloud is set to receive 60,000 GPUs, injecting high-performance computing power into its own data centers and accelerating the development of a full-stack AI infrastructure. The two will form a physical AI platform that seamlessly bridges the real and digital worlds.

Through this collaboration, Naver Cloud aims to identify industry-specific AI models for sectors such as shipbuilding, energy and biotechnology, facilitating the rapid deployment of AI technologies optimized for real-world industrial environments.