On August 26, Li Auto (LI.US) released its interim results announcement for the six months ended June 30, 2026. During the reporting period, Li Auto recorded total revenue of RMB 48.65 billion, a year-on-year decline of 13.4%, while net profit swung from a profit of RMB 1.74 billion in the same period last year to a net loss of RMB 3.98 billion.
Product Portfolio Transition Triggers Financial Fluctuations
According to the financial report, Li Auto achieved total revenue of RMB 48.65 billion in the first half of 2026, down 13.4% year-on-year. Vehicle sales revenue reached RMB 45.6 billion, a decrease of 14.9% year-on-year, which the company attributed primarily to a decline in average selling price resulting from a different product mix, as well as a reduction in vehicle deliveries. On the profitability front, gross profit for the first half totaled RMB 4.644 billion, down 59.2% year-on-year; overall gross margin narrowed from 20.3% in the same period of 2025 to 9.5%, with vehicle gross margin contracting from 19.6% to 7.8%. In the first half of this year, Li Auto delivered a cumulative total of 193,472 new vehicles. Affected by factors such as narrowed gross margins, the company posted an operating loss of RMB 5.299 billion in the first half, compared to an operating profit of RMB 1.099 billion in the same period of 2025; the net loss attributable to ordinary shareholders of Li Auto stood at RMB 3.981 billion, versus a net profit of RMB 1.743 billion in the prior-year period.
Self-Developed Chip and VLA Model Rollout Anchor Embodied Intelligence
On the expense side, the company demonstrated rigid R&D investment alongside optimized operating costs: R&D expenses for the first half reached RMB 5.5 billion, up 3.3% year-on-year, with continued investment in core technology areas such as chips, intelligent driving, and large models; selling, general, and administrative expenses decreased to RMB 4.3 billion, down 17.6% year-on-year, primarily due to reduced employee compensation. Li Auto's R&D investment is primarily directed toward embodied intelligence. During the reporting period, the self-developed Mach M100 chip achieved mass production and deployment in vehicles. The chip uses a 5nm automotive-grade process and delivers a single-chip computing power of 1280 TOPS, with a compute utilization rate exceeding 80%. Li Auto stated that this marks a full-chain independent breakthrough spanning architecture design, compiler development, thermal management systems, and functional safety. More importantly, the chip has been integrated with the self-developed Mach VLA model for a unified deployment, granting vehicles comprehension and decision-making capabilities that differ from traditional ADAS. On the language model front, the company iterated its Mach Mind-Pro cloud-based large model and Mach Mind-Edge on-device model. Mach Mind-Pro covers all in-vehicle scenarios, offering cost-performance advantages for mass production across token generation speed, task completion quality, and usage cost dimensions. Mach Mind-Edge, meanwhile, is the industry's first truly mass-produced, on-device native embodied intelligence agent. It employs multimodal streaming temporal modeling to continuously understand the dynamic physical world, possesses causal reasoning and autonomous decision-making capabilities, and can proactively perceive its surroundings around the clock while invoking hardware for vehicle control and interaction. On the robot model side, the Mach VLA adopts a native 3D ViT at its foundation, integrating lidar and visual information to achieve a 50% improvement in visual range. Based on the new architecture, Li Auto unified its perception, prediction, and planning modules into a native multimodal MoE large model, aligning spatial perception, reasoning, and driving behavior within a single framework, significantly enhancing the efficiency of risk anticipation and execution decision-making. The company has explicitly stated that the capabilities of the Mach M100 chip and Mach VLA model are not limited to automobiles but can extend to broader embodied intelligence applications, supporting additional AI scenarios.
Comments