China’s EV giants develop in-house driving chips to replace Nvidia

In the past twelve months, China’s four largest electric vehicle manufacturers have all introduced their own autonomous-driving chips, a strategic pivot driven by rising costs, US export restrictions, and the rapid mainstreaming of smart-driving features. This move directly challenges the dominance of Nvidia and Horizon Robotics, the two primary suppliers of driving-compute hardware in China’s automotive sector.

The most recent announcement came from Li Auto, which unveiled the Mach M100, a 5-nanometre chip designed for its new L9 Livis SUV. A single unit delivers 1,280 trillion operations per second (TOPS) with an impressive 82% utilisation rate. Earlier, BYD introduced the Xuanji A3, a 4nm chip now in mass production. Three of these chips working in concert exceed 2,100 TOPS, enough to support Level 3 and Level 4 autonomy. Nio has been shipping its own NX9031 since the ET9 sedan launched in early 2025, rolling it out across its main brand and its mass-market Onvo sub-brand. Xpeng is deploying its Turing chip, claiming up to 2,250 TOPS per unit.

Two forces are driving this shift. The first is geopolitical. US export controls have tightened access to Nvidia’s most advanced chips, making self-reliance a strategic necessity. The second is economic. Smart-driving technology has moved from a luxury perk to a mass-market expectation in less than a year. By 2025, 67.6% of passenger cars in China were equipped with driver-assistance systems, according to Horizon Robotics. The share with mid-to-high-level systems, including urban navigate-on-autopilot, nearly doubled to 42.6% from 21.6% the year before.

The cost pressure is most acute in the budget segment. In cars priced below 200,000 yuan ($29,545), which account for about 65% of China’s passenger-car sales, the penetration of advanced driver-assistance systems surged from 5% at the start of 2025 to over 50% by year-end. BYD’s God’s Eye B system, which includes urban NOA, is now available across all its models for 12,000 yuan. Goldman Sachs reported a take-up rate of more than 60% on models like the entry-level Seagull, effectively bringing the floor price of a BYD with urban NOA to 78,800 yuan. Goldman also noted that Nvidia and Horizon Robotics posted gross margins of 75% and 65% respectively in 2025. For carmakers producing millions of vehicles annually, those margins represent a significant cost they believe they can reclaim by designing their own chips.

Huatai Securities estimates that Nio’s switch to its in-house NX9031 chip cut per-vehicle costs by 10,000 yuan. Beyond unit savings, custom chips offer tighter hardware-software integration, faster iteration, and greater control over product roadmaps. In a market where autonomous-driving capability is rapidly becoming the primary differentiator, these advantages are critical.

Nvidia remains the global benchmark with its Drive Thor platform, but its position in China, its largest automotive market outside the US, is eroding. BYD, the world’s largest EV maker by volume, no longer relies on Nvidia for its smart-driving stack. Nio has replaced Nvidia silicon across its entire lineup. The competitive intensity in Chinese EVs now extends beyond batteries and pricing into the chip architectures that power autonomy. Jensen Huang has warned that Chinese companies running advanced AI on non-Nvidia hardware would be a “horrible outcome” for America. In China’s autonomous-driving sector, that outcome is already unfolding.