Xpeng's Turing Chip: A Bold Move in the Autonomous Driving Race
April 15, 2025, 9:55 pm
Xpeng Motors is back in the game. The company is gearing up to launch its first in-house chip, the Turing, aimed squarely at the autonomous driving market. This chip is not just another piece of hardware; it’s a statement. A statement that Xpeng intends to reclaim its position as a leader in the electric vehicle (EV) sector.
The Turing chip, which has been in development since 2020, is set to hit mass production soon. It boasts a performance of around 700 TOPS (trillions of operations per second), putting it in direct competition with Nvidia’s Thor chip. However, while Nvidia has faced delays and setbacks, Xpeng is moving forward with confidence. The Turing chip is designed specifically for heavy artificial intelligence workloads, making it a powerful tool for autonomous driving.
What sets Turing apart? It features two proprietary neural cores and an architecture tailored for large neural networks. This means it can handle models with up to 30 billion parameters, a significant leap from competitors like Li Auto, which operates with a mere 2.2 billion parameters. But bigger models come with challenges, particularly latency. Xpeng has yet to disclose how it plans to tackle this issue.
The backdrop to this launch is a rapidly evolving landscape. Nvidia’s Thor was expected to be the game-changer for the industry, but it has stumbled. Reports indicate that the chips reaching partners are underperforming, with only 750 TOPS delivered. This uncertainty has left many EV makers, including BYD and Zeekr, scrambling. They had built their next-generation models around Thor, and delays could mean missing crucial launch windows.
In response, companies like Nio, Li Auto, and now Xpeng have opted to develop their own chips. Nio’s Shenji NX9031 is already in production, and Li Auto is on the verge of launching its own design. Xpeng’s Turing chip is a critical part of its strategy to regain its competitive edge.
Just a few years ago, Xpeng was a frontrunner in autonomous driving. However, it lost ground to rivals like Huawei and BYD. Now, with Turing and a revamped software stack, Xpeng aims to reassert its dominance. The company’s vision is clear: to own the entire stack—hardware, software, data, and chips—creating a tightly integrated system that it controls from end to end.
Tesla has set the standard in this arena. Its “Full Self-Driving” (FSD) system, built on innovations like bird’s eye view and occupancy networks, has kept it ahead of the pack. Analysts estimate that Tesla’s system is 6 to 12 months ahead of its competitors. In the absence of public updates from Tesla, Chinese firms are exploring their own variations. Li Auto has already commercialized an end-to-end model that handles 95% of driving scenarios.
Xpeng is also on this path. In May 2023, it launched its own end-to-end model, designed to retain information more effectively than traditional systems. The company’s head of autonomous driving recently emphasized the need for larger models and more data than what car-based chips can handle. Xpeng’s solution? Train massive models in the cloud and compress them for deployment in vehicles. This approach sidesteps the limitations of edge computing while maintaining sophistication.
Central to this strategy is Xpeng’s “world foundation model,” a 72-billion-parameter system designed for chain-of-thought reasoning. This model can understand the world, apply commonsense logic, and translate those judgments into physical driving actions. Xpeng’s R&D team has already tested models of various sizes, with the largest being 35 times the size of current mainstream systems.
To support its ambitious model goals, Xpeng has built China’s first ten-exaflop automotive computing cluster, operating at over 90% efficiency. The company is betting on the scaling law: bigger models lead to better data and performance. Xpeng plans to roll out Level 3 self-driving features in the latter half of 2025, with Level 4 trials expected by 2026.
However, the road ahead is not just about chips and software. Talent is equally crucial. Xpeng’s foundation model program is led by its North America-based AI team, while competitors like Xiaomi are also ramping up their efforts. The race is on, not just for technology but for the brightest minds in the field.
As Xpeng prepares to launch the Turing chip, it finds itself in an arms race. The stakes are high, and the competition is fierce. With Turing hitting the streets, Xpeng has given itself a fighting chance to lead once again. The question remains: will it be enough to reclaim its former glory in the fast-paced world of autonomous driving? Only time will tell. But one thing is clear: Xpeng is not backing down. The Turing chip is more than just a product; it’s a declaration of intent.
The Turing chip, which has been in development since 2020, is set to hit mass production soon. It boasts a performance of around 700 TOPS (trillions of operations per second), putting it in direct competition with Nvidia’s Thor chip. However, while Nvidia has faced delays and setbacks, Xpeng is moving forward with confidence. The Turing chip is designed specifically for heavy artificial intelligence workloads, making it a powerful tool for autonomous driving.
What sets Turing apart? It features two proprietary neural cores and an architecture tailored for large neural networks. This means it can handle models with up to 30 billion parameters, a significant leap from competitors like Li Auto, which operates with a mere 2.2 billion parameters. But bigger models come with challenges, particularly latency. Xpeng has yet to disclose how it plans to tackle this issue.
The backdrop to this launch is a rapidly evolving landscape. Nvidia’s Thor was expected to be the game-changer for the industry, but it has stumbled. Reports indicate that the chips reaching partners are underperforming, with only 750 TOPS delivered. This uncertainty has left many EV makers, including BYD and Zeekr, scrambling. They had built their next-generation models around Thor, and delays could mean missing crucial launch windows.
In response, companies like Nio, Li Auto, and now Xpeng have opted to develop their own chips. Nio’s Shenji NX9031 is already in production, and Li Auto is on the verge of launching its own design. Xpeng’s Turing chip is a critical part of its strategy to regain its competitive edge.
Just a few years ago, Xpeng was a frontrunner in autonomous driving. However, it lost ground to rivals like Huawei and BYD. Now, with Turing and a revamped software stack, Xpeng aims to reassert its dominance. The company’s vision is clear: to own the entire stack—hardware, software, data, and chips—creating a tightly integrated system that it controls from end to end.
Tesla has set the standard in this arena. Its “Full Self-Driving” (FSD) system, built on innovations like bird’s eye view and occupancy networks, has kept it ahead of the pack. Analysts estimate that Tesla’s system is 6 to 12 months ahead of its competitors. In the absence of public updates from Tesla, Chinese firms are exploring their own variations. Li Auto has already commercialized an end-to-end model that handles 95% of driving scenarios.
Xpeng is also on this path. In May 2023, it launched its own end-to-end model, designed to retain information more effectively than traditional systems. The company’s head of autonomous driving recently emphasized the need for larger models and more data than what car-based chips can handle. Xpeng’s solution? Train massive models in the cloud and compress them for deployment in vehicles. This approach sidesteps the limitations of edge computing while maintaining sophistication.
Central to this strategy is Xpeng’s “world foundation model,” a 72-billion-parameter system designed for chain-of-thought reasoning. This model can understand the world, apply commonsense logic, and translate those judgments into physical driving actions. Xpeng’s R&D team has already tested models of various sizes, with the largest being 35 times the size of current mainstream systems.
To support its ambitious model goals, Xpeng has built China’s first ten-exaflop automotive computing cluster, operating at over 90% efficiency. The company is betting on the scaling law: bigger models lead to better data and performance. Xpeng plans to roll out Level 3 self-driving features in the latter half of 2025, with Level 4 trials expected by 2026.
However, the road ahead is not just about chips and software. Talent is equally crucial. Xpeng’s foundation model program is led by its North America-based AI team, while competitors like Xiaomi are also ramping up their efforts. The race is on, not just for technology but for the brightest minds in the field.
As Xpeng prepares to launch the Turing chip, it finds itself in an arms race. The stakes are high, and the competition is fierce. With Turing hitting the streets, Xpeng has given itself a fighting chance to lead once again. The question remains: will it be enough to reclaim its former glory in the fast-paced world of autonomous driving? Only time will tell. But one thing is clear: Xpeng is not backing down. The Turing chip is more than just a product; it’s a declaration of intent.