Li Auto’s Spinoff Plan: Li Auto, Nio, and BYD Challenge Nvidia and Horizon Robotics
According to Automotive World analysis, the common thread among Chinese EV makers is a deliberate pivot away from Nvidia and Horizon Robotics, with smart-driving silicon being the most practical arena for semiconductor self-reliance. The analysis notes that this shift is driven by the need to reduce dependence on foreign suppliers, control costs, deepen hardware-software integration, and mitigate risks from potential export restrictions. However, it also points to persistent challenges, such as the strong entrenchment of Nvidia’s software ecosystem and continued heavy imports of microcontrollers and analog chips.
Li Auto confirmed its 5nm Mach M100 chip delivers 1,280 TOPS per unit (2,560 TOPS dual-chip) with 82% utilization. In a reported parallel move, Automotive World notes that Li Auto registered Xinchuang Zhihe Technology in Shanghai for chip design and sales, indicating a potential spin-off of its semiconductor unit.
According to Automotive World, Nio’s GeniTech has raised nearly CN¥3bn ($440m) at a valuation of ~CN¥8.27bn, with Nio holding a 62.7% controlling stake. The same outlet reports that BYD’s Xuanji A3 is China’s first 4nm smart-driving chip, supporting SAE Levels 3-4 autonomy and delivering over 2,100 TOPS in a three-chip cluster at roughly one-third the hardware cost of an equivalent Nvidia Thor system.
Li Auto Registers Xinchuang Zhihe, Hinting at Potential Chip Spin-Off
According to recent filings reported by Automotive World, Li Auto has registered a new entity—Xinchuang Zhihe Technology—in Shanghai, chartered specifically for IC design and sales. Industry sources view this registration as the strongest signal yet that Li Auto is poised to spin off its semiconductor division into a standalone subsidiary, granting it financial independence to pursue external ADAS/AD contracts outside the parent company.
On the technical specification side, Li Auto has released granular data for the Mach M100 architecture that is highly relevant to energy efficiency and thermal management. The chip is confirmed to be fabricated on a 5nm FinFET node, enabling high transistor density. Peak performance is rated at 1,280 TOPS (INT8) in a single-chip configuration, scaling to 2,560 TOPS in a dual-chip redundant architecture for Level 4 autonomy. The most significant mechanical-adjacent metric disclosed is the silicon utilization rate of 82% at full load. This is benchmarked against the broader ADAS industry, where utilization rates frequently sit between 50–65% due to memory wall and pipeline inefficiencies. Higher utilization directly reduces idle current draw and thermal dissipation, meaning fewer parasitic watts are drawn from the 800V architecture to cool the compute stack, effectively extending vehicle range during compute-heavy autonomous operations.
Detailed Spec Table: Li Auto Mach M100
| Specification | Mach M100 (Single) | Mach M100 (Dual) | Industry Avg. (Comparable) |
|---|---|---|---|
| Process Node | 5nm FinFET | ||
| Peak TOPS (INT8) | 1,280 | 2,560 | ~500 |
| Silicon Utilization Rate | 82% | ~55% | |
| Effective Throughput | 1,049 TOPS | 2,099 TOPS | ~275 TOPS |
| Energy Efficiency | High (Minimal parasitic wattage/TOPS loss) | ||
Pros and Cons of the Spin-Off Structure
| Pros | Cons |
|---|---|
| Allows Mach M100 to be offered to third-party OEMs, distributing R&D amortization. | Introduces potential supply chain friction and transfer pricing conflicts between Li Auto EV and Xinchuang Zhihe. |
| An independent entity can move faster in the competitive chip-design talent market. | Li Auto loses absolute direct control over its core ADAS hardware roadmap. |
| High utilization rate makes the chip exceptionally attractive to range-conscious EV manufacturers. | Reliance on a pure-play foundry (likely TSMC) exposes the spin-off to capacity and geopolitical risks standard in the fabless model. |
Li Auto CTO Xie Yan has underlined the challenge, stating: “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.” This comment highlights the strategic imperative behind the potential spin-off: to achieve the scale and continuous iteration that can only come from external sales and independent operation.
How the Spinoff Impacts the EV Chip Supply Chain
Automotive World analysis suggests the Xinchuang Zhihe registration is less an internal reorganization and more a strategic weapon aimed at the broader supply chain. The common thread among Chinese EV makers is a deliberate pivot away from Nvidia and Horizon Robotics; smart-driving silicon is widely viewed as China’s most practical battlefield for semiconductor self-reliance. By spinning off the chip unit, Li Auto allows the Mach M100 architecture to be sold to rival OEMs without internal channel conflict, directly pressuring incumbent suppliers on cost and lead time.
This impact is amplified by sheer market volume. China commands roughly 60% of global NEV sales, and a modern smart vehicle can carry between 1,600 and 3,000+ individual chips. This demand density sustains a highly competitive 12–18 month hardware refresh cycle for ADAS computing. An independent Xinchuang Zhihe, unfettered by Li Auto’s internal vehicle production schedules, can ruthlessly pursue this rapid cycle, offering the domestic supply chain a high-throughput, fabless alternative that deepens the existing shift away from reliance on Nvidia and Horizon Robotics.
What’s Next for Li Auto’s Chip Ambitions
Despite the strategic clarity offered by the Xinchuang Zhihe registration, it is not yet confirmed whether Li Auto will definitively proceed with the spin-off of its chip unit. Based on available source intelligence, the outcome remains unknown.
Further complicating any market projection, direct TOPS comparisons across different process nodes are unreliable due to fundamental architectural differences that cannot be captured by a raw TOPS figure.
Crucially, it remains unknown whether the chip unit can achieve the necessary scale and continuous iteration before its core design risks technological obsolescence.
Image Credit: Automotive World / Stewart Burnett
Source: Automotive World
Original Image: Link
