Source | YuanSight

'My current feeling can be summed up in eight words: after suffering comes happiness, dreams come true.' In a recent New Year's Eve livestream, Xiaomi founder Lei Jun reminisced about the year's triumphs and tribulations. Besides setting fresh goals for Xiaomi's smartphone and automotive divisions, Lei Jun's recent focus on expanding the AI business has garnered significant attention.

According to media reports, Lei Jun is rumored to have personally lured '95s genius girl' Luo Fuli with a salary in the tens of millions. Luo Fuli, from the popular large model community DeepSeek, is a key developer of its open-source large model, DeepSeek-V2.

Tech leaders focusing on large models are not limited to Lei Jun. Li Xiang, the founder of Li Auto, who has recently 'resurfaced' on the internet, also frequently discusses AI.

At the recent 2024 Li Auto AI Talk, Li Xiang proclaimed that cars will evolve from transportation tools of the industrial era to space robots in the AI era. He further stated that he aspires to be more than just the CEO of an automotive company but the CEO of an AI company.

In the current era dominated by large models, for Lei Jun and Li Xiang, focusing solely on smartphones or automotive manufacturing may no longer fully satisfy their ambitions.

01

AI Fuels Intelligent Driving Competition

As early as the beginning of 2023, Li Xiang mentioned in an internal letter that he hopes Li Auto will become a globally leading AI company by 2030.

At the recent 2024 Li Auto AI Talk, Li Xiang reiterated that Li Auto invests over 10 billion yuan annually in R&D, with nearly half of this investment allocated to AI, including self-developed end-to-end and VLM (Vision Language Model) base models.

According to Li Xiang's vision, the ultimate goal is to transform cars into 'family members' where humans do not need to give commands, and the car can make decisions autonomously, functioning as a 'silicon-based family member'.

'Li Auto's mission is to create a mobile home, a happy home, so I refer to it as a silicon-based family member. I don't need to give it any more instructions, nor do I need to assign it tasks. It is our family member, even an important organizer of the family. It not only understands me but also understands my children and friends around me,' said Li Xiang.

As a relative newcomer among new-energy vehicle brands, Li Auto has aggressively pursued intelligent driving by leveraging AI large models.

In 2024, Li Auto commenced development of mapless NOA, fully rolled out in July; in October, it announced the full rollout of its next-generation dual-system intelligent driving solution combining end-to-end and VLM; in November, Li Auto introduced parking space-to-parking space intelligent driving; and by the end of December, it announced an end-to-end, full-scenario upgrade integrating end-to-end and VLM for highways and cities.

From a purely utilitarian perspective, focusing on intelligent driving undoubtedly boosts Li Auto's car sales.

Recently, Li Auto shared a set of data: in February 2024, the delivery volume of Li Auto vehicles equipped with AD Max accounted for only about 20%, exceeding 50% in the second half of the year.

Traditional intelligent driving systems typically comprise multiple independent modules, such as a perception module for recognizing roads, vehicles, pedestrians, etc., a decision-making module that formulates driving strategies based on perception results, and a control module that translates decisions into actual vehicle operations (like acceleration, braking, steering, etc.). These modules require complex information exchange and coordination, and each module needs to be designed and optimized individually.

In contrast, end-to-end intelligent driving systems simplify this process by directly utilizing technologies like deep learning to enable the model to learn how to directly output vehicle control commands from raw sensor data (like camera images, LiDAR point cloud data, millimeter-wave radar data, etc.).

In simpler terms, after integrating AI large model technology, the intelligent driving system can learn from numerous human driving video clips from diverse scenarios and independently summarize a set of driving methods based on human experience.

Since then, with the continuous 'real-world implementation' and popularity of end-to-end solutions from automakers like Tesla, Huawei, and XPeng, competition in the intelligent driving sector has intensified.

Under this trend, Xiaomi has also included end-to-end solutions in its agenda. On December 31 last year, Xiaomi Motors announced the rollout of the pioneer version of its end-to-end, full-scenario intelligent driving system, supporting features like automatic vehicle exit from parking spaces and automatic parking spot search upon arrival at the destination.

In terms of talent and computing power closely related to large models, Xiaomi has been continuously increasing its investments. Besides Lei Jun personally offering high salaries to attract talent, Xiaomi is also focusing on computing power. Last December, media reports revealed that Xiaomi is building its own GPU cluster with tens of thousands of cards.

With the advent of the AI era, for Xiaomi, which simultaneously handles smartphone and automotive businesses, the significance of early layout in large models may outweigh that for other smartphone manufacturers.

02

Converging with Musk's Path

Last May, Musk's large model company 'xAI' completed a financing round of $6 billion, valued at $24 billion post-investment, making it the world's second-largest AI unicorn within a year of its establishment.

Tesla has always been a crucial lever for Musk's AI business.

In 2022, Tesla's autonomous driving department proposed to Musk that a system similar to ChatGPT could be tailored for cars to plan optimal routes; in November 2023, when xAI officially released the open-source large model Grok-1, Musk planned to integrate it into Tesla vehicles.

However, Tesla is just one of the 'tools' Musk uses to pursue his AI dreams. When discussing the primary application scenarios for AI, Musk stated that autonomous vehicles and general-purpose humanoid robots will be key.

'Once we have these humanoid robots and autonomous driving technology, we can actually manufacture anything. Theoretically, if humanoid robots can work intelligently and their numbers are unlimited, there are no practical limits to the scale of the economy,' said Musk.

In 2022, Tesla introduced the Optimus robot at AI Day. According to the plan, this humanoid robot will be used to perform repetitive and high-risk tasks. In December 2024, Tesla released a daily video of the latest Optimus robot, showcasing its ability to complete balance tests like going up and down slopes.

After the robot's market launch, Musk painted an ambitious picture - the total market value of Tesla will reach ten times that of Apple's current value, exceeding $30 trillion, with Optimus robots playing a crucial role.

Like Musk, car manufacturing is also becoming a 'side business' for Li Xiang and Lei Jun.

During the aforementioned interview, Li Xiang stated that Li Auto will definitely venture into the humanoid robot industry in the future, with a '100% probability of making robots.' However, he also noted that the current priority is not to develop humanoid robots but to achieve Level 4 autonomous driving.

Li Xiang believes that Li Auto's goal for next year is to become the number one in China's spatial intelligence field, ranking in the top three in linguistic intelligence, large language models, and the services provided. 'Li Auto will train computing power according to corresponding requirements. We need to truly compete with leading enterprises, not just within the automotive industry,' he said.

Unlike Li Xiang, who currently only considers robots as a planned business, Xiaomi has already ventured into the robot business.

Since 2023, Xiaomi has been unusually active in robotics, first establishing Xiaomi Robotics and then building Beijing's first research and development, manufacturing, and application demonstration project for general-purpose humanoid robot complete machines in Yizhuang, with a planned investment of 2 billion yuan.

03

AI is Sexier Than Car Manufacturing

It has been said that the money Musk earns from Tesla subsidizes SpaceX to realize his dream of pursuing Mars. However, it must be noted that besides car manufacturing, AI, robots, and aerospace businesses have also brought tremendous imagination to Tesla.

In recent years, whether in terms of business growth rate, stock price, market value, etc., an important reason why Tesla has been ahead in the automotive industry is that investors in the secondary market do not just view Tesla as a hardware company but as a technology company or even an AI company.

Judging from the current competitive environment, Tesla obviously has reasons to increase investment in emerging businesses like AI.

Tesla is facing enormous pressure for growth in its automotive business. In 2024, Tesla produced a total of 1.7734 million vehicles and delivered 1.7892 million, a 1.1% decrease from the 1.81 million vehicles delivered in 2023. This is the first annual sales decline for Tesla since 2015.

In China, Tesla sold 657,000 vehicles in 2024, a year-on-year increase of 8.8%, but its advantage in the new energy vehicle market is gradually being eroded by other domestic brands.

Under these circumstances, Musk needs to find new growth points to maintain a steady cash flow and meet investors' expectations of Tesla as an AI company.

Lei Jun and Li Xiang are also focusing on AI, each with their own strategic considerations.

Although Li Auto comfortably sits at the top of new-energy vehicle sales with a cumulative sales volume of over 500,000 vehicles in 2024, its 'doll-like' product model in recent years and the weak sales of its pure electric vehicle Mega have been criticized.

Meanwhile, as electric vehicles now aim to 'install fuel tanks,' competition in the extended-range sector is intensifying, and with the future of pure electric models uncertain, focusing on intelligence is undoubtedly the best route for Li Auto to break through.

When recalling his entrepreneurial experience, Li Xiang said he chose a very specific field within the mobile internet ecosystem. 'Although we did well, to some extent, you might miss the forest for the trees,' he said.

AI seems to be the forest Li Xiang is aiming for. Li Xiang plans to achieve 100% autonomous driving within three years for smart cars. When asked if Li Auto could become a company like Apple if objective conditions are met, Li Xiang replied, 'Definitely.'

Nowadays, with smartphone manufacturers like Apple, Huawei, OPPO, and vivo having already placed early bets on AI, for Xiaomi, which has always wanted to prove itself as an internet company rather than just a hardware company, the AI large model business can not only enhance software services but also drive the IoT sector and further empower the 'human-car-home ecosystem'.

Following Musk's path is also a constant test of the financial strength of successors.

According to reports, Microsoft and OpenAI plan to invest over $100 billion by 2028 to build a data center project. Li Xiang has also pointed out that achieving the goal of 100% autonomous driving within three years requires sufficient funds to recruit top talent, purchase adequate computing power, master VLA (Vision Language Action) capabilities, and have over 5 million cars on the road.

It is foreseeable that the AI competition may evolve into an arms race that only giants can afford.

Some images are sourced from the internet. Please inform us if there is any infringement for removal.