Li Auto has registered a new entity in Shanghai—Xinchuang Zhihe Technology—that covers chip design and sales, a move which suggests the company may spin off its in-house semiconductor business into an independently operated unit. One of many automaker moves to support Chinese chip independence, the registration follows closely after the May unveiling of Li Auto’s Mach M100 chip, which reached the market after four years of development.

The Mach M100 is built on a 5 nm automotive-grade process delivering 1,280 TOPS per chip, or 2,560 TOPS in the dual-chip configuration already in mass production on the L9, L8 and L6. The chip achieves an 82% utilisation rate, which Li Auto characterises as highly efficient. During the reveal event, Chief Technical Officer Xie Yan said the real test of an in-house chip programme comes down to deployment at scale, quick integration, support for future models and continued iteration without the risk of the technology becoming obsolete before it pays for itself.

Should Li Auto proceed with an independent chip subsidiary, it would mirror compatriot brand Nio, whose chip unit GeniTech has already raised nearly CN¥3bn (US$440m) across two funding rounds at a valuation of roughly CN¥8.27bn. This left the subsidiary with a 62.7% controlling stake. GeniTech made its first standalone appearance at the World Artificial Intelligence Conference in Shanghai in July, showing chips spanning driver assistance, embodied AI and agent inference.

In May, BYD revealed its own Xuanji A3, China’s first 4 nm smart-driving chip, capable of natively supporting SAE Level 3-4 autonomy and delivering more than 2,100 TOPS in a three-chip cluster. The chip claims roughly a third of the hardware cost of an equivalent Nvidia Thor-based system. Elsewhere, Xpeng has taken a similar path with its in-house Turing chip. The wider field spans a strikingly inconsistent range of process nodes, from Tesla’s 7 nm AI4 chip to Nio and Li Auto’s 5 nm silicon and Xpeng and Horizon Robotics’ 7 nm designs, making direct TOPS comparisons across companies somewhat unreliable.

The common thread across all of these programmes is a deliberate move away from dependence on Nvidia and Horizon Robotics, whose Orin and Journey chips still dominate the low- and mid-tier trims of most Chinese electric vehicles, including BYD’s own God’s Eye driver-assistance system. Owning the silicon lets automakers control cost, tune hardware to their own algorithms, and reduce exposure to any future tightening of foreign chip export rules.

That last point is precisely why smart-driving silicon has become China’s most practical battlefield for semiconductor self-reliance, arguably more so than server-class AI chips. Modern smart vehicles require 1,600 to 3,000-plus chips each, models refresh every 12 to 18 months rather than the four to five years typical of legacy automaker, and automotive-grade performance can be achieved on more mature 7 nm or 12 nm processes rather than the bleeding-edge 3 nm and below that server chips require. All of this lowers the barrier for domestic chipmakers to compete credibly against Nvidia and Qualcomm.

China’s roughly 60% share of global NEV sales gives this strategy a captive, high-volume testing ground no other country can match, though real gaps remain: Nvidia’s CUDA and Drive OS software ecosystems remain deeply entrenched, and lower-level microcontrollers, sensors and analog chips, where firms like Texas Instruments, NXP and STMicroelectronics dominate, are still heavily imported. 

Li Auto’s push toward an independent chip business—alongside Nio, BYD and Xpeng’s parallel efforts—suggests the smart-driving chip race is becoming less about any single automaker’s technical bragging rights and more about which companies end up owning the commercial infrastructure behind China’s broader chip independence goal.