The first AI company's financial report of Ideal, backed by five questions

On March 12th, Li Auto released its 2025 financial report: revenue of 112.3 billion yuan, three consecutive years of profitability, with a net profit of 1.1 billion yuan. In the eyes of some investors, if measured by the standards of the automotive manufacturing industry, it’s not “rich” enough; if viewed through the lens of a tech company, it quietly holds an astonishing 101.2 billion yuan in cash.

At the table of Chinese tech companies, Li Auto has always been an “outlier.” In the past, it was the only emerging force to achieve continuous profitability, regarded as “the car company that knows how to live well.” Now, it has taken more than 50% of its earnings and invested in a high-stakes AI gamble—self-developed chips, large-scale models, incubating robot projects…

This 2025 financial report from Li Auto is like a “double-sided mirror,” reflecting the stability of manufacturing on one side and projecting the wild dreams of a tech company on the other. What kind of chess game is Li Auto playing next? The answer lies in five questions.

Why dare to say “three consecutive years of profitability”?

In 2025, Li Auto’s net profit is 1.1 billion yuan… If the outside world only looks at this single indicator, they might think that the profit scale doesn’t match the revenue.

But widening the perspective reveals several intriguing details: this is Li Auto’s third consecutive year of profit, and Li Auto is also the only domestic new force car company to have revenue exceeding 20k yuan for three consecutive years while remaining profitable. More importantly, by the end of 2025, Li Auto’s cash reserves reached 101.2 billion yuan—this number ranks first among all Chinese automakers.

During the strategic investment period, Li Auto still maintained profitability. This is the key to understanding this “contrast” financial report. The so-called “strategic investment period” is reflected in the annual R&D expenditure of 11.3 billion yuan, a record high, averaging nearly 1 billion yuan per month, with 50% allocated to AI-related investments.

Why is AI-related investment so high? Li Auto has reassessed its positioning, no longer viewing itself merely as an automobile manufacturer, but rather as a tech entity evolving towards “embodied intelligence.” Under this logic, the 1.1 billion yuan net profit demonstrates the ability to maintain financial health while undergoing large-scale transformation investments.

The 101.2 billion yuan cash reserve supports this reassessment. In the current environment of ongoing price wars in new energy vehicles and intensified technological competition, cash means strategic flexibility. It can support mass production of self-developed chips, fund technical breakthroughs in steer-by-wire chassis, or provide space for “startup-style incubation” of new businesses like AI glasses and humanoid robots.

At the earnings conference, Li Auto’s Chairman and CEO Li Xiang clearly positioned this: “In the vertical integration of embodied intelligence technology, we will invest 100%. We believe embodied intelligence has commonality in overall systemic technology, including inference chips on the edge, different models and operating systems, as well as data and training systems behind them. This is where we will focus all our efforts, as we believe this is the most fundamental basis for all embodied intelligence.”

This perhaps explains why Li Auto’s financial report shows such a “contrast”: leading revenue scale but not “super profits,” huge R&D investments but still abundant cash reserves. This is not the typical appearance of a traditional manufacturing company pursuing short-term profit maximization, but rather a tech company undergoing a technological generational leap, a typical feature during a transition period.

Where did the money go? Half was invested in invisible AI.

Regarding Li Auto’s R&D investment in 2025, CFO Li Tie stated at the earnings conference: “We believe that cars and AI are not independent. The core of our R&D investment is to build AI capabilities and integrate them into our existing business model. In our current business model, all R&D investments are integrated; AI R&D is not a separate business unit.”

The most direct manifestation of this “integration” is a self-developed Li Auto chip—MACH 100—that is about to enter mass production this year. When running large-scale VLA models, a single MACH 100 chip’s effective computing power is three times that of Nvidia Thor-U, fully enhancing the safety and smoothness of intelligent assisted driving.

Li Auto CTO Xie Yan revealed the design philosophy behind MACH 100 at the earnings conference: “Under the same silicon area, MACH 100 can provide higher effective computing power, offering greater design space for VLA algorithms. For example, we can deploy VLA models with parameter scales six times larger than the previous generation, with ten times the computational load, while still achieving higher frame rates and faster inference speeds.”

More importantly, this is a full-stack self-research effort. From chips, compilers, operating systems, to foundational models, the entire stack is self-developed, enabling Li Auto to customize AI computing power from the ground up and build a technological barrier difficult for competitors to copy.

Xie Yan used an analogy to explain this strategic layout: “When we started chip self-research in 2022, we judged that from 2025 onwards, the industry would enter an era of vertical integration of models, chips, and operating systems to gain competitive advantage. We believe the gap between chips, operating systems, and the software-hardware ecosystem will eventually resemble the gap between Apple and Android.”

The deep integration of MACH 100 with Xinghuan OS and steer-by-wire systems is bringing tangible product improvements. Xie Yan explained that this integration makes communication between autonomous driving computing, front and rear processing, and execution control more efficient, significantly reducing overall latency from sensor photon input to vehicle output.

Besides performance improvements, Xie Yan also revealed that MACH 100 offers significant cost advantages, with the BOM cost per chip substantially lower than external procurement solutions.

The value of this technological foundation has already begun to spill over externally. The Li Auto AI glasses Livis, released in December 2025, is the first validation of this AI technology transfer to consumer terminals. It proves that the technological foundation built on embodied intelligence in vehicles can be reused across products.

“We expect this year’s R&D expenses to remain around 12 billion yuan, with about half allocated to AI-related R&D,” Li Tie said.

Automobile vs. AI company, why does Li Auto see itself as a third path?

On the Chinese new energy vehicle scene, players are being rapidly categorized. On one side are traditional automakers and new forces, mostly struggling at the profit-loss line, with few achieving sustained profitability; on the other side are AI tech companies, relying on financing to burn money chasing AGI, with commercial deployment still distant.

Li Auto’s 2025 financial report presents a “middle state” that is hard to classify, combining the stability of manufacturing with the imagination of a tech company.

Compared to traditional automakers, price wars erode profits, R&D investments keep rising, and many companies struggle in a loss cycle. Even multinational giants like Volkswagen and Toyota face painful challenges such as margin compression, cost reduction, and layoffs.

Li Auto maintained its profitability during the 2025 strategic investment period, essentially maintaining its “self-sustaining” function. More importantly, its profits are not achieved at the expense of future growth. This “profitable while investing heavily” mode is rare in the auto industry.

In contrast, AI companies present a different picture. Large model startups often raise hundreds of millions of dollars, but their commercialization paths are vague, and most are in a continuous “burning money” state. Even leading AI firms struggle with “huge investments and distant returns.”

Li Auto’s uniqueness lies in its clear monetization path—selling cars. According to official statements, Li Auto’s i6 production capacity has broken through bottlenecks, with a monthly capacity of 20k units starting in March, and the proportion of pure electric models continues to rise; the hybrid + pure electric dual-drive pattern is fully established.

Beyond cars, Li Auto has expanded into other fields without blindly burning money. For new areas like AI glasses and humanoid robots, Li Xiang’s strategy is clear: “In terms of real business and product deployment, we will proceed cautiously and keep exploring. When expanding into new fields, we will start as a startup, including AI glasses and robot projects. We want to avoid the wasteful spending typical of large companies, and instead incubate products through startup teams and methods.”

Li Auto’s approach essentially combines the self-sustaining ability of its automotive business with long-term AI investments, forming a self-reinforcing evolution cycle: selling cars generates cash flow, which supports R&D, building technological barriers, which in turn enhances product strength, driving sales growth.

In the capital market, if Li Auto is viewed as an automaker, the focus is on sales volume, profit margins, and per-vehicle profit; if seen as an AI tech company, the market is willing to pay a premium for technological barriers, data assets, and future potential. Li Auto combines the “steady” of three years of profitability with the “high-growth potential” supported by hundreds of billions in cash reserves.

How does AI help Li Auto reconstruct “people” and “organization”?

Li Xiang said: “For Li Auto, AI mainly involves two aspects: one is creating AI, and the other is how to use AI. Creating AI means endowing products with life; using AI is about improving work efficiency.”

Looking back at Li Auto’s organizational changes over the past year, a clear evolution line can be seen: from traditional hierarchical management to a platform-enabled organization.

In January, Li Auto completed an internal restructuring that some might find “abstract.” The R&D teams no longer divided departments by traditional hardware and software functions but reorganized according to a new logic—Li Xiang calls it “organizing around the creation of digital humans and silicon-based humans.”

In this restructuring, Li Auto reconstructed its R&D system into four major blocks: organ systems, brain systems, software core, and hardware core. The underlying logic of this organizational change, according to Li Xiang, is: “Creating products is like creating humans.”

On the R&D side, Li Auto restructured its architecture to include chips (analogous to the heart), data sets (analogous to the lungs), and operating systems (analogous to the nervous system), breaking down previous departmental walls between hardware and software, enabling experts from different fields to collaborate closely.

The effects of this restructuring quickly became apparent. “Recently, everyone has noticed that work efficiency has improved. The training cycle for intelligent driving models, which used to iterate every two weeks, now iterates daily,” Li Xiang said.

Behind this new organizational logic is Li Auto’s rethinking of enterprise forms in the AI era. Li Xiang summarized: “AI is reversing the decline in information flow and decision-making efficiency caused by company size expansion. By introducing agent collaboration, our iteration and evolution speed has begun to resemble that of startups, and this change has already been reflected in daily operations this year. It can be said that AI has not only brought a revolution in production tools but also fostered a more vibrant organizational form.”

When company size growth becomes an obstacle to efficiency, AI and organizational change together form Li Auto’s response. The former is tool-level empowerment, the latter is mechanism-level restructuring. The combination allows a company with tens of thousands of employees to maintain the rapid evolution typical of startups.

How does L9 Livis define the “embodied intelligence flagship”?

In the second quarter of this year, the all-new Li L9 and Li L9 Livis will be launched. In Li Auto’s internal definition, Li L9 Livis is called “a flagship SUV based on embodied intelligence logic, bringing generational breakthroughs in user experience.”

Behind this somewhat awkward positioning is the first full realization of Li Auto’s hundreds of billions of yuan R&D investment over the past three years. From perception and decision-making to execution, the all-new Li L9 Livis has undergone a technological reconstruction in three dimensions.

“Whether autonomous driving or embodied intelligence, the biggest challenge is training and learning through video, and controlling the vehicle through video. Essentially, this technical system does not truly understand the physical world but is mimicking learning,” Li Xiang said.

The all-new Li L9 Livis aims to break this limitation, upgrading from the “2D camera + lidar” scheme to 3D ViT, truly perceiving and understanding the physical world like humans, rather than just watching videos.

Li Xiang explained the significance of this breakthrough: “This requires joint design and breakthroughs at the chip level, from video encoders to the entire model, providing strong computational support for encoders and models. Whether autonomous driving or embodied intelligence, I believe a major global breakthrough will happen this year.”

Accompanying this breakthrough, the VLA model will become a true “VLA”—where “Language” is no longer just for object recognition but will genuinely understand and think about the physical world like humans.

Equipped with MACH 100, the Li L9 Livis also reserves space for future evolution. Xie Yan said: “Higher local computing power allows us to provide richer intelligent capabilities beyond autonomous driving, making the car more like a robot. These capabilities will first be implemented on L9 and will continue to expand in the future.”

If perception and the brain determine how well the vehicle “sees” and “thinks,” then the body determines how accurately it “moves.”

The all-new Li L9 Livis is equipped with the world’s first “full-body steer-by-wire” chassis, including steer-by-wire steering, four-wheel steering, and the world’s first fully electric-controlled mechanical brake (EMB). According to official sources, this is the first time within a 600k-yuan price range that an 800V full-active suspension system has been installed, delivering top-tier handling and driving pleasure.

Li Xiang explained the significance: “This makes the vehicle’s response speed and safety far superior to traditional cars. The vehicle no longer relies on traditional MCU communication for control but can directly output to the control system through the model.”

In traditional cars, the driver’s intent is transmitted via mechanical or electronic signals to actuators, making the vehicle a passive tool. In the new L9 Livis, the large model can directly control all actuators via MCP, turning the vehicle from a tool into an intelligent agent capable of active response.

Li Xiang’s positioning of the Li L9 Livis is clear: “The all-new Li L9 Livis completes a technological reconstruction in perception, decision-making, and execution.” It is not just an incremental upgrade but Li Auto’s first complete work after transforming into an embodied intelligence company.