Chinese Automakers Race to Build Their Own Chips as Domestic Substitution for Autonomous Driving Chips Accelerates
Miles Bennett
Chinese automakers and AI labs are ditching Nvidia chips in parallel — BYD, NIO, and XPeng have launched in-house alternatives while DeepSeek V4 now runs on eight domestic chips. Nvidia's window in China is closing from both ends.
Why are autonomous-driving companies dropping Nvidia?
Autonomous trucking startup Zelostech told CNBC it plans to switch to multiple chip suppliers within one to two years, ending its sole reliance on Nvidia.
Its finance director Shi Yunjian said domestic chips will cost far less than the two Nvidia Orin chips currently installed per vehicle.
This means → the first driver of the switch is not politics — it is cost. Domestic chips are now cheap enough for companies to absorb the switching risk.
What have BYD, NIO, and XPeng done?
BYD last week unveiled its own driver-assistance chip, joining NIO and XPeng in the in-house camp.
NIO CEO Li Bin said the company no longer buys chips — it rents compute powered by a mix of processors.
XPeng's joint vehicle with Volkswagen runs on XPeng's in-house "Turing chip"; VW separately partnered with Horizon Robotics for driver-assistance systems in China — neither uses Nvidia.
In plain terms = the top automakers are not "considering alternatives." They have already replaced Nvidia.
How far along is the AI-model ecosystem?
MiniMax, Kimi's latest models, and DeepSeek V4 are all compatible with domestic chips.
A Goldman Sachs report dated May 5 noted DeepSeek V4 runs on eight domestic chips, including those from Huawei and Alibaba's T-Head unit.
Goldman expects "the shift to domestic chips will accelerate from 2026 through 2028."
This means → the model ecosystem is not waiting for chips to mature — it is actively feeding real training data back to domestic chip makers.
What signal did Huawei's latest move send?
Huawei disclosed last week that its chip R&D has adopted new scientific methods and plans to integrate the results into future products.
This reflects a recovery signal: years after U.S. sanctions, Huawei is still iterating on self-developed chips.
How much time does Nvidia have left?
Kevin Xu, founder of hedge fund Interconnected Capital, estimates Chinese firms will still need Nvidia chips for the next three to five years.
But he argues Beijing is motivated to compress that timeline — domestic chips can only improve through large-scale use in real commercial settings.
His core logic: "The more Nvidia chips enter the ecosystem, the more diluted the drive for domestic substitution becomes."
In plain terms = Nvidia's window is not measured in "how many years remain." Each additional chip sold slows domestic replacement by one step — so Beijing will actively narrow that window.
Where does Nvidia stand in China right now?
Nvidia's revenue from mainland China and Hong Kong has been shrinking steadily.
CEO Jensen Huang visited Beijing in May alongside President Trump and began a partnership with robotics startup Unitree — yet overall Chinese demand for Nvidia chips keeps falling.
This means → high-level visits and point partnerships cannot reverse a structural trend: the demand side is systematically walking away from Nvidia.
Content is for reference only, not financial advice.