As NVIDIA’s market cap surged to $4 trillion, making it the world’s most valuable company, some automakers courted by CEO Jensen Huang are actively distancing themselves from the chip giant.
NVIDIA has faced setbacks in automotive innovation, particularly with Mercedes-Benz and General Motors.
In June 2023, Mercedes CEO Ola Källenius and his CTO participated in a cross-country U.S. test drive covering thousands of kilometers. They evaluated vehicles equipped with NVIDIA’s and Chinese startup Momenta’s ADAS systems on a round trip between Los Angeles and San Francisco. The conclusion: NVIDIA’s performance fell short of Momenta’s. Later that year, Mercedes requested urban NOA demonstrations in Shanghai. Momenta’s system achieved near-zero disengagement, while NVIDIA’s showed “sudden braking and aggressive acceleration inconsistent with human driving habits.”
This year, GM executives internally labeled NVIDIA’s ADAS solution “very scary” after evaluation.
Per 36Kr reports, Mercedes has transferred ADAS development for multiple China-market models from NVIDIA to Momenta. Jaguar Land Rover, another NVIDIA software client, is also seeking alternative ADAS suppliers. “NVIDIA staff in China have largely ceased engaging with automaker projects.”
While automakers historically raced to adopt NVIDIA’s latest chips—starting with the new Thor “God of Thunder” SoC—NVIDIA now risks losing major Chinese clients in bulk.
The threat extends beyond specialized ADAS players like Huawei, Horizon Robotics, and Momenta. Chinese EV makers are launching in-house automotive chips: NIO and XPeng already have self-developed chips in production; Li Auto’s ADAS chip enters mass production next year; Xiaomi founder Lei Jun confirmed imminent rollout of its automotive silicon.
Nvidia’s New Chip Hit by Delay
Notably, the repeated delays and delivery failures of NVIDIA’s Thor chip have provided the final push for automakers to accelerate the deployment of their self-developed chips.
For example, Li Auto postponed the release of its upgraded EREV L-series models from March to May this year. According to 36Kr sources from a core supplier, this delay was directly caused by NVIDIA’s failure to deliver Thor chips on schedule. Meanwhile, upon detecting signs of further Thor delays in early 2025, XPeng decisively halted development for the Thor platform and urgently redirected resources to adapt its in-house “Turing” chip.
Sources indicate that Li Auto has expedited the deployment of its self-developed chips, advancing mass production by several months to Q1 next year.
This situation stems from NVIDIA’s strategic positioning: as an entity deriving less than 2% of revenue from automotive business, its priority lies in conquering technical challenges rather than ensuring delivery timelines.
His misalignment manifests in resource allocation. Multiple automaker engineers revealed to 36Kr that NVIDIA allocated insufficient resources to address Thor’s delivery issues, noting: “We even had to implement engineering contingencies at the domain controller level to mitigate chip design flaws ourselves.” Significantly, CEO Jensen Huang rarely addresses automotive matters in his routine email communications.
Top Carmakers Crack In-House Chip Barrier
Developing chips is an adventure for any automobile manufacturer. The research and development of a chip, whether it’s by NIO, Li Auto, XPeng or others, has almost taken a full four years.
William Li once publicly wrote to recall, “The most thrilling moment was in 2023. At the critical juncture when the front-end design of the chip was about to be completed, an important partner suddenly decided to end its business in China.” NIO’s back-end chip design encountered severe challenges. In the end, it managed to build its own back-end design team and apply for an account with TSMC, advancing step by step until the tape-out stage.
Because of this, NIO’s chip team has an unusually complete scale, with more than 600 people covering front-end design, back-end design and testing, which is close to the configuration of a standard chip company.
Many engineers commented to 36Kr that the architecture design of NIO’s Shenji chip is even more reasonable than that of NVIDIA’s Thor. In fact, there’s no need to blindly worship NVIDIA, because “in the field of autonomous driving chips, everyone is actually on a similar starting line.”
It is reported that the cost of the first self-developed chips of NIO, XPeng and Li Auto is basically between 300 million and 400 million US dollars. And the investment is still increasing continuously. Li Auto and others are already preparing for the research and development of the second chip.
The impenetrable ecological barriers such as CUDA and NVlink established by NVIDIA in the GPU market seem difficult to be transformed into a moat in the automobile field. Many partner automobile manufacturers are waiting and seeing whether NVIDIA’s automotive chip or software business will be abandoned.
Discover more from ChinaEVHome
Subscribe to get the latest posts sent to your email.
