Core Viewpoint - The article discusses the optimization of AI model architecture design on a given autonomous driving chip using a mathematical framework, allowing engineers to predict the optimal model structure without extensive training of multiple models [1][2]. Group 1: Hardware Co-Design - The paper emphasizes hardware co-design, which integrates AI model architecture parameters with hardware characteristics, rather than traditional software and hardware collaboration [2]. - The core philosophy of the model design is to maximize intelligence while ensuring response speed and memory usage constraints are met [4][5]. - The research team evaluated 1,942 candidate architectures on the NVIDIA Jetson Orin platform, ultimately selecting 170 representative models for training, achieving a 19.42% reduction in perplexity compared to Qwen2.5-0.5B under the same latency conditions [2]. Group 2: Mathematical Framework - The first formula defines the ultimate optimization goal, focusing on minimizing the model's error rate while adhering to strict constraints on response time and memory usage [3][4]. - The second formula serves as a model capability predictor, allowing engineers to estimate the performance of any model architecture without actual training, thus accelerating design iterations [11][12]. - The third formula connects model architecture with hardware characteristics, helping engineers identify performance bottlenecks in specific chips [17][18]. Group 3: Performance Metrics - The article highlights that the inference process of large language models on current edge computing chips is often limited by memory bandwidth, rather than peak computational power [22]. - The fourth formula presents a counterintuitive conclusion that, in memory-constrained edge devices, wider models should have higher sparsity, suggesting a trade-off between model width and expert activation rates [24][25][26]. Group 4: Chip Characteristics - The TOP2 Mach 100 chip is designed specifically for AI model inference, optimizing memory bandwidth and effective computational power rather than peak performance [27]. - The self-developed compiler plays a crucial role in the hardware-software synergy, enabling flexible support for new computing modules without being limited by vendor-specific hardware [28].

此外，品牌已投入使用4054座理想超充站，拥有22447个充电桩。 截至2月28日，理想汽车在全国布局有539家零售中心，覆盖160个城市；售后维修中心及授权服务中心 548家，覆盖223个城市。 观点网讯：3月1日，理想汽车公布2026年2月交付数据，当月共交付新车26421辆。 免责声明：本文内容与数据由观点根据公开信息整理，不构成投资建议，使用前请核实。 ...

Group 1 - In February 2026, Li Auto delivered 26,421 vehicles, a 4.5% decrease from January 2026's 27,668 deliveries, and a 0.6% increase compared to February 2025's 26,263 deliveries, indicating stagnation in year-over-year growth [1] - For the full year of 2025, Li Auto's total deliveries were 406,300 vehicles, representing an 18.8% decline year-over-year, and only achieving 58.05% of its initial target of 700,000 vehicles [1] - Li Auto's cumulative delivery as of December 31, 2025, reached 1,540,215 vehicles, making it the first new force brand in China to surpass the 1.5 million delivery milestone [1] Group 2 - In 2025, NIO and Xpeng reported significant sales growth, with NIO delivering 326,000 vehicles (up 46.9%) and Xpeng delivering 429,400 vehicles (up 125.94%), while Li Auto was the only company among the "new forces" to experience a decline in sales [2] - Li Auto's Q3 2025 revenue was 27.4 billion yuan, down from 42.87 billion yuan in the same period last year, with a gross profit of 4.5 billion yuan, a 51.6% decrease year-over-year [2] - The net loss for Li Auto in Q3 2025 was 624.4 million yuan, contrasting with net profits of 2.8 billion yuan in Q3 2024 and 1.1 billion yuan in Q2 2025 [2] Group 3 - Li Auto's OTA 8.3 system was officially launched before the Spring Festival, enhancing user experience with three core systems: VLA driver model, smart cockpit, and smart electric systems [3] - The new generation of Li Auto L9 and its ultimate version, Li Auto L9 Livis, are set to launch in Q2 2026, aiming to redefine flagship SUV products in the era of intelligent vehicles [3] - As of February 28, 2026, Li Auto has established 539 retail centers and 548 service centers across 223 cities, with 4,054 charging stations and 22,447 charging piles in operation nationwide [3]

理想汽车：2026年2月交付26,421辆汽车。 ...

Core Insights - Li Auto reported a new vehicle delivery of 26,421 units in February 2026, indicating strong sales performance [1] - As of February 28, 2026, the total cumulative delivery of Li Auto vehicles reached 1,594,304 units, showcasing significant growth in the company's delivery capabilities [1] Company Summary - The new vehicle delivery figure for February 2026 reflects the company's ongoing expansion and market penetration efforts [1] - The cumulative delivery milestone of over 1.5 million vehicles highlights Li Auto's position in the electric vehicle market and its ability to scale operations effectively [1]

Group 1 - The core viewpoint of the article highlights the expansion of Li Auto's retail and service network across China, indicating significant growth in its operational footprint [1] - As of February 28, 2026, Li Auto has established 539 retail centers covering 160 cities [1] - The company has also set up 548 after-sales service and authorized service centers, which span 223 cities [1] Group 2 - Li Auto has invested in 4,054 Li Auto Supercharging stations nationwide, providing a total of 22,447 charging piles [1]

Core Viewpoint - Li Auto Inc. continues to strengthen its position in China's new energy vehicle market, achieving significant vehicle deliveries and expanding its infrastructure and product offerings [1][2][4]. Group 1: Vehicle Deliveries and Performance - In February 2026, Li Auto delivered 26,421 vehicles, bringing its cumulative deliveries to 1,594,304 as of February 28, 2026 [1]. - The company powered over 1.45 million charging sessions during the Spring Festival travel peak, with a total charging volume exceeding 42 million kWh [2]. Group 2: Infrastructure and Services - As of February 28, 2026, Li Auto operated 4,054 super charging stations with 22,447 charging stalls across China [3]. - The company has established 539 retail stores in 160 cities and 548 servicing centers in 223 cities [3]. Group 3: Product Development and Innovations - Li Auto rolled out OTA update version 8.3, enhancing the VLA Driver large model, smart cockpit, and smart electric functionality [2]. - The company plans to launch the all-new Li L9 in the second quarter of 2026, expanding its product lineup [2][4]. Group 4: Company Overview - Li Auto is a leader in the new energy vehicle market, focusing on premium smart electric vehicles and extended-range electric vehicles [4]. - The company emphasizes innovation in product, technology, and business models to provide safe and convenient services to families [4].

Core Insights - The report released by Li Auto indicates a significant increase in charging activity during the 2026 Spring Festival, with a total of 1.45 million charging sessions recorded over 10 days, resulting in a total charging volume of 42.37 million kWh, marking a new high for the company [1] Group 1: Charging Activity - Li Auto users accounted for 530,000 charging sessions, while users of other brands' electric vehicles contributed 920,000 sessions [1] - The charging volume at Li Auto's high-speed service station reached 27.04 million kWh, representing 15.9% of the total charging volume at high-speed service stations nationwide [1]

Group 1 - The article highlights the integration of AI subtitles in live streaming, enhancing user experience by providing simultaneous translations in both English and Chinese, allowing for a more coherent viewing experience [1] - It discusses the convenience of watching live streams on WeChat Video Accounts, where users can easily access and share content without complex operations [2] - The article mentions that top participants in the live stream will respond to questions in the background, facilitating deeper discussions about the actual business operations and long-term fundamentals [3]

Core Viewpoint - The core viewpoint of the article is that Li Auto is shifting its sales assessment from a sole focus on sales volume to emphasizing user reputation and service quality, through the introduction of the "Store Partner Program" aimed at enhancing store management capabilities in response to increasing market competition [1][2]. Group 1: Store Partner Program - The "Store Partner Program" is not a franchise model but an incentive mechanism for frontline store managers, aimed at improving store operational capabilities [2]. - The program represents a shift in the sales power structure, decentralizing authority to store levels, with store managers now responsible for overall store performance [4]. - This change allows frontline stores greater autonomy in decision-making and direct participation in profit sharing, with the goal of placing knowledgeable individuals in key positions to manage stores effectively [5]. Group 2: Market Context and Competition - The initiation of this program is closely related to changes in the market environment, where competition has shifted from product differentiation to channel and operational efficiency [6]. - The competition among car manufacturers is increasingly about store management capabilities rather than just product features [7]. - Li Auto aims to enhance the cohesion and effectiveness of its frontline teams through this program, allowing outstanding individuals to share in the operational success [7]. Group 3: Store Network Optimization - Li Auto is also optimizing its offline store network, having closed underperforming stores and opened new comprehensive stores in certain cities [8]. - The new store strategy emphasizes matching suitable managers with store locations, ensuring that operational responsibilities are aligned with the right personnel [8]. Group 4: Future Product Cycle - The launch of the "Store Partner Program" is aligned with Li Auto's upcoming product cycle, with several new models expected to be released by 2026 [8]. - The goal of the organizational adjustment is not only to boost sales but also to ensure that frontline stores possess sustainable operational capabilities, making them crucial connection points between the brand and users [8][9].