Soochow: LI AUTO (02015) is firmly on the path of embracing AI comprehensively, maintaining a "buy" rating.

Soochow released a research report stating that LI AUTO-W (02015) is firmly moving towards embracing the AI road comprehensively. The company maintains its revenue expectations for 2026/2027/2028 to be 114.4/173.7/184 billion yuan respectively, with year-on-year growth of +2%/+52%/+6%. The company also maintains its net profit attributable to shareholders for 2026/2027/2028 to be 2.2/2.8/5.3 billion yuan. The EPS for 2026/2027/2028 are projected to be -1.04/1.31/2.43 yuan respectively, with corresponding PE ratios of 37/20 times for 2027/2028. The firm believes that as the company progresses in software and embodied intelligence and embraces AI implementation, it maintains a "buy" rating. Soochow's main points are as follows: - IDEAL Auto held the Livis Day software and embodied intelligence conference, redefining "embodied intelligent vehicles." - The company held the Livis Day software and embodied intelligence conference, revealing the progress of its software, base model, and self-developed chips. The core focus of this conference, according to the firm, is the restructuring of the product paradigm: the company pointed out that the industry's existing Siasun Robot & Automation, AI, smart driving, and car upgrades are transformed into "capabilities" rather than "products," and defined embodied intelligent vehicles as a single product with four attributes in one - one that serves real-life scenarios, electric vehicles that users can participate in; two, professional drivers who can reliably complete travel and daily tasks, understand roads/risks/traffic rules; three, a dedicated AI computer for large-scale design, computational power, and integrated memory and memory storage leading flagship mobile phones and computers; four, a trusted life assistant that understands user habits and replaces "functional operations" with "issuing demands." - Plans and Outlook - In the second half of 2026, there will be three OTA nodes, aligning with FSDV14 by the end of the year. The company plans to push the Mach VLA cut model to Orin and Thor platform owners in the third quarter of 2026, and achieve alignment with Tesla FSDV14 by the fourth quarter of 2026. The company explicitly stated that the focus of OTA in the second half of 2026 will be on three nodes in July, September, and December: the plan for July is to improve overall driving efficiency by about 30%, focusing on operation similar to human behavior in September, and aiming to surpass human safety and efficiency by December. - Intelligent Cockpit - Car hardware takes a step forward. The world's first launch of the Qualcomm Snapdragon 8797 (8797 Elite) platform, with approximately 550K CPU task processing capabilities, 8.1T GPU rendering, and 320 TOPS NPU; the industry's first panoramic screen with a 90Hz high refresh rate and 180Hz touch sampling rate, front display bandwidth of 23.5 G. The audio system has been upgraded to the 9.3.6 solution, peaking at 5440W, and compatible with CarPlay and Apple Music. - Base Model - Dual-wheel language intelligence + machine intelligence. On the language side, the native Agent model Mach Mind-Pro was launched, with the addition of Mind-Edge at the car end. In benchmarks like IFEval and BFCL-v4, it ranks in the top tier, utilizing token compression and hybrid RL training, achieving zero attenuation completion rate, reducing token consumption by 38%, reducing tool call redundancy by 47%, and peaking at 208 tokens/s TPS. In terms of machine intelligence, the Mach VLA era started in May 2026, with the emergence of the three-variable Scaling Law data model. - Self-developed Chips - Mach M100 in mass production, with a paper selected for ISCA 2026. The M100 is the world's first data flow architecture edge-mounted in-car AI chip, using a 5nm automotive-grade process, single-core 1280 TOPS, abandoning the Von Neumann central instruction scheduling, and moving towards a data flow arrangement that emphasizes "data is computation." In terms of performance, under the same power consumption and similar area, UniAD's frame rate is approximately 3.8 times that of Thor-U. IDEAL has achieved self-development of chips, compilers, operating systems, AI algorithms, and domain controllers in the entire stack. - Risk Warning: Passenger car price wars exceed expectations; terminal demand recovery is lower than expected; the pace of the introduction of Level 3 intelligent driving policies is slower than expected.