August 25, 2026 - Chinese Automakers Shifting Away from Nvidia

This is the Telemetry Transportation Daily for August 25, 2026, and I'm Sam Abuelsamid, Vice President of Market Research for Telemetry. 

Over the past decade, Nvidia has dramatically shifted its business, increasingly moving away from its roots as a developer of graphics chips to power computer video games to commercial applications powering artificial intelligence. While much of the media attention has focused on the costly Nvidia GPUs that power the growing number of data centers, that doesn't tell the whole story. Nvidia has leveraged its technology for applications including health care, meteorology, robotics, and automotive. Much of the early work on automated driving systems relied on compute platforms using X86 CPUs and Nvidia GPUs, and Nvidia has provided a series of development platforms that provide an excellent starting point for ADS developers. 

As has frequently been the case recently, Chinese automakers have been among the first to adopt Nvidia's latest chips, the Orin and now Thor, although they are now being used by European brands like Volvo and Mercedes-Benz, with GM joining in 2028. But there are a couple of challenges with using Nvidia's chips in automotive applications. Because the basic architecture of Nvidia's chips is being used for so many different applications, they have become increasingly complex and often contain components an automaker might not need. That leads to them being costly and power-hungry. Nvidia chips remain an excellent tool for training AI models, but sometimes they can be too much for running inference or using the models. 

For Chinese automakers, the increasingly fractious trade situation between the Trump administration and the rest of the world is also making it more challenging to use Nvidia silicon. As a result, we are seeing more companies moving away from Nvidia to develop their own custom chips for in-vehicle inference for driver-assist systems. Tesla was among the first companies to use Nvidia for ADAS in 2016, but it switched to custom inference chips in 2019. More recently, Nio, XPeng, and BYD have moved to custom chips, while others have adopted silicon from Horizon and Huawei. Rivian will be making the switch from Nvidia to an in-house chip about the end of this year. The latest to announce custom chips is Xiaomi, with the first 3nm chip produced in China and 160 GB of unified memory.  

As long as the build-out of AI data centers continues to put pressure on the availability and pricing of SoCs, RAM, and storage, we'll probably continue to see companies at least taking a look at more cost-effective and energy-efficient alternatives for inference in vehicles. 

Thanks for listening.