Li Auto is advancing the first round of external financing for its chip subsidiary at a pre-money valuation of about 15 billion yuan ($2.2 billion), Chinese media outlet LatePost reported on Monday, a move that would mark the next stage of the automaker’s push to build an independent semiconductor business.
The company aims to raise several billion yuan, according to the report.
A successful raise would value Li Auto’s chip business, before any new money, above the post-money valuation of about 10 billion yuan ($1.5 billion) that Nio‘s rival chip unit reached in its first round in February.
Corporate Structure
Corporate records show Li Auto has built a new holding structure for the chip business since mid-July, starting with a Hong Kong company, Mach Intelligent Cores Limited, incorporated on July 14.
On August 4, that company established a wholly owned Shanghai subsidiary, Shanghai Mach Zhixin Intelligent Technology, with Xie Yan, Li Auto’s Chief Technology Officer, as legal representative.
Days later, the Shanghai unit set up a further subsidiary, Shanghai Mach Xin Technology, whose legal representative is Wang Yang, Li Auto’s Joint Company Secretary.
The Hong Kong holding company suggests the business is being reorganised to accept offshore investment.
The new structure sits alongside a separate company Li Auto registered on July 13 in Shanghai’s Zhangjiang area, Xinchuang Zhihe (Shanghai) Technology, wholly owned by its onshore unit Shanghai Li Auto Technology, with Wang as legal representative and a business scope covering chip design and sales.
It had been reported as the vehicle for the chip programme. No source has explained how the two structures relate.
Li Auto’s management has already approved a spin-off of the chip business and has been pitching the M100 to robotics companies, 36Kr reported on September 18. Xie said in May that Li Auto’s chips were for its own products.
By mid-June he said selling them externally was not ruled out and that robot makers had shown interest.
A 200-Person Team
Li Auto‘s in-house chip team has about 200 employees focused on AI computing architectures, chip design and related software development, LatePost reported.
The team is developing two categories of internally designed chips: vehicle-side processors and cloud-side inference chips.
Human resources staff have already begun discussing employment adjustments with chip-team members, one employee told the outlet, a signal that contracts may shift to the new subsidiary.
Xie Yan, who heads Li Auto’s Systems and Computing Group, has been the driving force behind the Mach M100 programme.
When he joined in July 2022, the chip team was still in its early stages.
Rather than starting with smaller, lower-risk silicon, Xie pushed the team to aim directly at parity with the highest-end products from Nvidia at the time.
The team chose a dataflow architecture over the more common GPGPU approach, a technical bet that Li Auto says delivers an operational efficiency rate above 82%.
In June, Li Auto granted a combined 35 million stock options to three executives as part of a long-term incentive plan. President Ma Donghui received 15 million, and Xie and Chief Financial Officer Li Tie 10 million each, at an exercise price of $14.38 per ADS.
The options vest in 20% tranches for each HK$200 billion of Hong Kong market value, with full exercise at HK$1 trillion ($127.5 billion). Li Auto was worth about HK$115 billion when the options were granted.
Mach M100
The fundraising comes with the Mach M100, Li Auto’s first in-house chip, already in mass production since May.
Built on a 5-nanometre automotive-grade process, a single M100 delivers 1,280 trillion operations per second. It has been fitted to the new L9, including the Livis flagship with two chips, and to the redesigned L8 and L6.
Xie told analysts in August that chip supply was sufficient to meet demand.
Li Auto is now developing a cloud-based inference chip, also using a dataflow architecture, that could potentially handle some inference workloads currently running on GPUs.
Industry insiders cited by LatePost said using a dataflow design for cloud inference is technically feasible, but the challenge is pushing the inference cost of the entire solution to a level that is competitive within the broader AI computing industry.
Funding Pressure
Expanding from vehicle-side applications to cloud computing will demand further research-and-development spending.
Li Auto lost 3.97 billion yuan ($593 million) in the first half, against a profit of 1.74 billion yuan a year earlier.
Its vehicle margin fell to 9.4% in the second quarter from 19.4% a year earlier, during a refresh of its whole L-series line-up. It still held 87.5 billion yuan ($13.1 billion) in cash at the end of June.
Bringing in external capital for the chip business could ease the burden on the parent company’s balance sheet while the core car operation recovers.
China’s semiconductor sector raised about 162.7 billion yuan ($24.3 billion) in the first half of the year, LatePost said, with AI computing among the largest categories attracting investment.
A separate equity incentive plan for the chip business could also help retain talent.
LatePost previously reported that Jin Yihua, head of software R&D for Li Auto’s chi-p team, and Dai Jie, head of the first front-end design group, had left the company.
A larger number of core technical staff have departed to join or launch embodied-AI ventures. Qin Dong, who ran the chip team’s SoC department, also left earlier this year, according to Leiphone.
Peer Comparisons
Nio’s chip subsidiary, Anhui Shenji Technology, closed a first round of 2.257 billion yuan ($337 million) in February at a post-money valuation of about 10 billion yuan. Nio retained 62.7%, external investors took 27.3% and 10% went to an employee incentive vehicle, according to LatePost.
Shenji has since begun licensing its chip technology and formed a joint venture with Axera.
Following the round, Nio retained 62.7%, external investors held 27.3% and 10% was reserved for an employee incentive vehicle.
Shenji has since begun licensing its chip technology to third parties and presented an expanded three-chip lineup at Shanghai’s WAIC conference in July.
Xiwang, an inference-chip company spun off from SenseTime’s large-chip division at the end of 2024, has completed seven financing rounds totalling about 4 billion yuan ($597 million), with a valuation exceeding 10 billion yuan.
XPeng‘s Turing chip has won a production contract with Volkswagen, and the company is looking to sell its technology to other carmakers. It has not disclosed plans to fund a standalone chip entity.
