Li Auto explores cloud inference chip as 2 chip leaders depart

Li Auto's move into cloud inference chips could cut its GPU supplier dependence, extending the automaker's chip strategy from the 1,280 TOPS Mach M100 to data center workloads. The project remains at an early stage.

"Cloud inference chips can take on some of the inference workloads that currently run on GPUs," a person in the chip industry told LatePost. These include data processing, testing and simulation for assisted driving models, as well as request handling for large language models.

The cloud chip will use the same dataflow architecture as Li Auto's assisted driving chip, which the company unveiled on May 12. The Mach M100, built on a 5nm automotive-grade process, delivers 1,280 TOPS per chip and 2,560 TOPS in dual-chip configurations, and is now in mass production in the Li L9, L8, and L6. One technical path involves reusing the vehicle-side NPU compute design, packaging multiple AI compute dies with high-bandwidth memory and high-speed interconnects to build a larger inference chip.

Li Auto registered Xinchuang Zhihe (Shanghai) Technology Co Ltd on July 13 with a business scope that includes integrated circuit chip design, a step that could point to plans to run its semiconductor business independently. The company's shares rose 2.06 percent to HKD48.44 on the Hong Kong exchange, with turnover of 2.26 million shares worth HKD109 million.

Dataflow Architecture Faces GPU Incumbent

The technical approach mirrors work at SambaNova, Groq, and Tenstorrent, which are also exploring dataflow architectures. SambaNova's core team comes from Stanford, Groq's from Google's TPU project, and Tenstorrent is led by Jim Keller, who previously headed Tesla's self-driving chip work. Compared with GPUs, dataflow architectures still need to prove themselves in model versatility, software support, and large-scale deployment.

A cloud chip is not simply a scaled-up vehicle chip. On the vehicle side, models, sensor inputs, and operating rhythms are relatively fixed, with the main goals being low power consumption, low latency, and stable execution. The cloud must handle more models simultaneously, with faster model updates and greater fluctuations in input length and concurrent requests. It also involves high-bandwidth memory, multi-chip interconnects, dynamic batching, and cluster scheduling.

Hitting a performance target with a single chip is not the hardest part — the real difficulty is making overall inference cost competitive, the person said. Whether extending the dataflow architecture from vehicle to cloud can create a cost advantage depends on model adaptation capabilities and system operating efficiency.

Chip Unit Loses Two Leaders

As the cloud project moves forward, Li Auto's head of chip software R&D, Jin Yihua, and Dai Jie, head of one of its chip front-end design groups, have left the company, LatePost confirmed through multiple channels. Both previously reported to Luo Min, head of the computing power unit, who in turn reports to group CTO Xie Yan. It is unclear whether the personnel changes will affect the cloud inference chip project's progress.

Li Auto's chip push places it alongside Chinese EV peers pursuing vertical integration. Nio set up chip subsidiary GeniTech in June 2025, which has raised nearly 3 billion yuan ($442 million) at a post-money valuation of about 8.27 billion yuan. Xpeng has also put its in-house Turing assisted driving chip into its vehicles.

The move into cloud inference chips points to deeper vertical integration into AI infrastructure, potentially lowering long-term costs and reducing dependence on external chip suppliers. Li Auto's founder and CEO Li Xiang has said the in-house chip effort is about making AI work in the physical world, not proving technical capability. With the Mach M100 already in production vehicles and the cloud chip project at an early stage, the strategic direction toward proprietary silicon is viewed positively for Li Auto's competitive positioning in intelligent driving and AI computing.

This article is for informational purposes only and does not constitute investment advice.