He Xiaopeng sent Wu Xinzhu to the door of NVIDIA's house

The hand was extended

Ren Yafei, reporter of "China Entrepreneur"

Editor|Yao Yun

Source of header image | Visual China

Today is Wu Xinzhu's first day at NVIDIA.

On the afternoon of August 24, the day before reporting to the new owner, Wu Xinzhu, former vice president of autonomous driving of Xpeng Motors, forwarded a Weibo by He Xiaopeng on his personal Weibo. He Xiaopeng mentioned in Weibo, "It turns out that Huang Sect Master has 2 hours to share the summary and outlook for the second day of each financial report, and often confesses his specific mistakes this quarter, which is really not easy!" The next opportunity is at AIGC and AUTO, and our deeper collaboration is about to begin. At the same time, the Weibo picture is a group photo of He Xiaopeng, Huang Jenxun and Wu Xinzhu.

Wu Xinzhu took this official announcement to forward and said: "Thank you Xiaopeng for personally sending Lao Huang here. According to Lao Huang, he still worked for Xiaopeng later, but he didn't have to pay his salary. "The cooperation between Xpeng Motors and NVIDIA is obvious.

It is understood that Wu Xinzhu will serve as the head of automotive products at NVIDIA, and in general, the business leaders can report directly to Huang.

On the same day, NVIDIA announced its financial results for the second quarter of this year, and the company's revenue in the quarter reached a record high of $13.51 billion, a year-on-year increase of 101%. Specifically, the data center business set a revenue record of US$10.32 billion, which became a "strong support" for NVIDIA during this period; In the gaming business, the other pillar of NVIDIA was stabilized with revenue of $2.49 billion, while in the automotive sector, the automotive business, which originally accounted for nearly 5%, fell to only 1.9% in the second fiscal quarter, and compared with the more than 100% growth momentum in the previous two fiscal quarters, the year-on-year growth rate in the fiscal quarter was only 15%.

For the unfavorable automotive business, Nvidia CFO Colette Kress said that the main reason is the overall decline in automotive demand, especially in the Chinese market.

Nvidia has always had high hopes for the automotive business, and Huang Jenxun has publicly stated: "In the future, the potential market size of NVIDIA will reach $1 trillion, and specific to market segments, the revenue of the automotive industry will account for 33%, reaching $300 billion." ”

From 2015 to 2022, relying on its long-term accumulated technical advantages in the field of intelligent driving, NVIDIA has successively launched Tegra series, Paker, Xavier, Orin and other high-computing power chip products, and has made a large wave of friends, Tesla, Mercedes-Benz, Land Rover, Volvo, NIO, Xpeng, etc. are included in NVIDIA's "circle of friends".

But Nvidia's ambitions are by no means just to take on the role of selling chips. Danny Shaoiro, vice president of its automotive division, once said: "NVIDIA is not just a technology supplier, from software to hardware, we want to do more for the automotive industry to make the cars of the future safer and more energy-efficient." ”

Previously, NVIDIA had directly taken a cut from Mercedes-Benz's sales by giving priority to providing high-computing power chips to Mercedes-Benz, so as to be more deeply involved in the operation of the main engine factory. For car companies, in order to gain the initiative in the dialogue with NVIDIA, they undoubtedly need to have the same chip design capabilities as Tesla, and this ability is almost difficult for most traditional car companies to achieve in a short period of time.

New NVIDIA

"As a technician, what I want to see most is the implementation of technology." Wu Xinzhu revealed his vision when he officially announced his joining Xpeng Motors.

At the beginning of 2019, Wu Xinzhu, who was the senior engineering director of Qualcomm's autonomous driving, joined Xpeng Motors as vice president of autonomous driving.

Xinzhu Wu graduated from Tsinghua University with a bachelor's degree in electrical engineering, and later received his master's and doctoral degrees in electrical engineering from the University of Illinois at Urbana-Champaign. After graduation, Wu Xinzhu joined Qualcomm in 2006 and served for 13 years, becoming a senior engineering director, holding 160 US patents, and is the soul of Qualcomm's autonomous driving business.

According to "LatePost", Xpeng Motors began to develop its own software after He Xiaopeng joined in 2017, but the early stage of research and development in intelligent driving was not smooth. It was not until Wu Xinzhu joined, reorganized the technical architecture, and integrated the team, that Xiaopeng's road of self-research was on the right track. On the P7 model launched in April 2020, Xpeng replaced the supplier's solution with a self-developed algorithm, more than one year earlier than ideal and nearly two years earlier than NIO.

After Wu Xinzhou joined, Xpeng Motors was gradually labeled as a "technical school" in the field of autonomous driving business. Xpeng's advantages in the field of autonomous driving have been transformed into market competitiveness and increased sales. Xpeng P's sales in July were close to 10,000 units, and it was also the first pure electric new force model to achieve mass production of 100,000 units.

Wu Xinzhu, who labeled Xiaopeng as "intelligent driving", is exactly the talent Huang Renxun wants.

Although NVIDIA is good at building hardware and making great progress in the field of autonomous driving computing, there are obvious shortcomings in the creation of services and ecosystems. In addition to the head car companies have sufficient self-research capabilities, most car companies still hope to have a complete packaging solution, and this is what NVIDIA is not good at, but Wu Xinzhou is good at.

For example, the product power is weaker than NVIDIA's horizon, and it uses heavy investment in ecology to enhance competitiveness. Horizon not only provides perception algorithms, but also helps customers solve basic engineering problems first, so that the workload of customer development is greatly reduced, and the adaptation cycle of algorithms and software is as low as possible. Therefore, car companies, Tier 1 and Horizon cooperate, from the beginning of the project to SOP, the progress can be very fast.

Now, with the practical experience and rich landing experience of intelligent driving software engineering obtained in Xiaopeng, Wu Xinzhu, who joined NVIDIA, can also create supporting software based on its excellent chips.

Source: Visual China

NVIDIA's car "circle of friends"

The usual new job involves two demands, one is the needs of the individual and the other is the needs of the organization. And this time Wu Xinzhou joined NVIDIA, there may be a third demand, the in-depth cooperation between Xpeng Motors and NVIDIA.

Source: Medium Attempt Library

According to Huang's plan, in the future, the automotive business will be listed alongside data centers and games as NVIDIA's three pillar businesses, which means that the automotive business must reach at least the same magnitude as the game and data center business. In the second quarter of 2023, NVIDIA's data center, gaming and automotive businesses generated revenues of $10.32 billion, $2.49 billion and $253 million, respectively, and the automotive business still has a long way to go to cross the $10 billion threshold.

Since 2015, NVIDIA has entered the field of in-vehicle SoCs and in-vehicle computing platforms to provide basic computing capabilities for autonomous driving.

At present, around the computing power competition of autonomous driving chips, NVIDIA has achieved the "top configuration" in the industry. NVIDIA releases an automotive-grade SoC chip almost every two years, and continues to increase the level of computing power. The Xavier chip computing power released in 2020 is 30TOPS, the Orin computing power released in 2022 is 254TOPS, and the Thor computing power is directly increased to 2000TOPS.

It is worth mentioning that NVIDIA released a new generation of automatic driving chip Thor, which is not only an automatic driving domain chip, but also an intelligent cockpit domain chip, which can be used for multiple purposes with one "core" to achieve cabin and drive integration and become the central computing unit of the car. The Thor chip has two obvious features, one is the large computing power, which is twice that of Altan, 8 times that of Orin, and the other is the function that can integrate a variety of automotive chips. The industry generally believes that with the release of the Thor chip, it is a sign that NVIDIA challenges the intelligent cockpit chip track represented by Qualcomm, Thor is planned to be mass-produced in 2024 and launched in 2025, and Extreme Krypton will become Thor's first customers.

At present, NVIDIA, including the third-generation Orin self-driving chip, has achieved a good market share. Fro & Sullivan statistics show that in 2022, NVIDIA's shipments will account for 82.5% of the global market share of high-computing power autonomous driving chips.

Relying on the long-term accumulated technical advantages in intelligent driving computing power, NVIDIA has also established a huge "circle of friends" in the automotive field.

Guosen Securities pointed out in the research report that NVIDIA's automotive customers can be roughly divided into three categories: first, new forces in car manufacturing, including NIO (ET5, ET7), Xpeng Motors (P5, P7, G9), Li Li (X01), WM (M7), SAIC Zhiji, R Auto, FF, Lucid Group, etc.; Second, traditional car companies, including BYD, Mercedes-Benz, Jaguar Land Rover, Volvo, Hyundai, Audi, Lotus, etc.; The third is autonomous driving companies, including GM Cruise, Amazon Zoox, Didi, Volvo Commercial Vehicles, Kodiak, Tucson Future, Plus, AutoX, Pony.ai, WeRide, Yuanrong Qixing, etc.

The key to NVIDIA's ability to quickly get a large number of customers is that for companies that want to develop intelligent driving at L3 and above, there are no more chips to choose from in the market.

Stretching out his hand again, the Yellow Sect Master wanted to be soft and hard

Huang is adamant that cars will be Nvidia's next multibillion-dollar business. However, NVIDIA started with hardware, and there is no shortage of more experienced car manufacturers in the automotive industry, and autonomous driving, which represents the future trend, also needs "soft power" that closely matches "hard power".

This puts forward requirements for NVIDIA's automotive ambition: to control both the brain of autonomous driving and the nerves of autonomous driving.

The relationship between the self-driving chip, the platform and the model is not easy to understand, NVIDIA gives an image metaphor: the car is the body, the self-driving platform is the nerve, and the autopilot chip is the brain. To achieve this goal, NVIDIA cannot rely only on chip hardware.

Huang Yefeng, a senior industry analyst in the semiconductor industry, said that from the general direction, there are two parts related to self-driving cars, one is the data center, which needs to process data, train AI, simulate digital twins, and build virtual worlds; The second is the AI computer on the car body, which processes various sensor data to perceive the driving environment and realize automatic driving.

The end-to-end platform for NVIDIA's autonomous driving development and deployment is called Nvidia DRIVE. From the perspective of the NVIDIA platform layer, DRIVE is mainly about Nvidia AI and Omniverse (the universe). At present, NVIDIA's three major platforms are Nvidia HPC, AI and Omniverse.

It is understood that Omniverse is a virtual simulation platform that can reflect the real size of the object and can abide by the laws of physics. It can help car companies to conduct realistic simulation before product mold opening or factory construction, helping customers maximize the exposure of various potential problems in the early design stage.

Omniverse has already cooperated with a number of car companies:

Volvo Car Corporation and General Motors use Omniverse USD Composer to connect and unify their asset workflows, which also enables the assembly of automotive parts into digital twins in a virtual environment; Mercedes-Benz and Jaguar Land Rover automotive engineers use Driver Sim generation synthetic data in Omniverse to train AI models to validate active safety systems through virtual NCAP driving tests, and the former also uses Omniverse to build, optimize and plan assembly flows for new models; Toyota uses the Omniverse to build a digital twin of its own factory.

Wu Xinzhu completes the stack

In 2021, NVIDIA's automotive business revenue was US$566 million, accounting for 2% of the company's total revenue; In 2022, the revenue of the automotive business increased to US$903 million, accounting for about 3%; In the second quarter of 2023, the proportion was only 1.9%.

This is far from the two goals set by Huang Jenxun. One is a software company in the field of smart cars, and the other is positioned in the field of car autonomous driving to achieve 30% market share.

Like Qualcomm, compared with the revenue of data centers and games, NVIDIA's automotive business accounts for a small proportion of revenue. Although the autonomous driving chip Orin will start large-scale mass production and delivery in 2022, and the automotive business has grown, the scale effect has not appeared. The reason is that in the past, chip manufacturers were more likely to provide hardware development reference designs for downstream customers to help them quickly provide overall software and hardware solutions for automotive companies. Today, software is becoming the new barrier to entry.

To this end, in 2021, NVIDIA acquired DeepMap, a high-definition map startup, in an attempt to expand its ability to provide full-stack solutions by integrating the latter's technical solutions.

The turning point came in cooperation with Mercedes.

In June 2020, Mercedes-Benz announced that it will cooperate with NVIDIA to develop fully autonomous vehicles in the next decade, leveraging their respective experience in high-performance computing and high-end automobile manufacturing to jointly create a new software-defined vehicle, and NVIDIA is responsible for providing software and hardware full-stack solutions, jointly developed with Mercedes-Benz. The partnership is considered by NVIDIA to be the "largest single-business model transformation" in the company's history, that is, the two parties can share the revenue from future user purchases of features and subscription services.

Public data shows that NVIDIA will receive $8 in revenue for every $1 sold of hardware, software and other related system providers. Huang also stressed that "the business model of automakers will fundamentally change." By 2025, many automotive companies are likely to sell cars at prices close to cost, primarily through software to provide value to users. In Huang's view, the most important factor for future users to buy new cars is software that is constantly iteratively upgraded and continuously enhanced.

Some industry insiders believe that NVIDIA saw the opportunity of artificial intelligence many years ago and invested heavily in developing a complete artificial intelligence stack, including software, services and hardware. The smart car track is no exception. With the landing of technical cooperation between Xpeng Motors and Volkswagen Group, NVIDIA saw the opportunity, but also felt the risk.

It is reported that due to the lack of full-stack software capabilities, the current cooperative development process between NVIDIA and Mercedes-Benz is not smooth, and the project has been delayed. NVIDIA and Mercedes-Benz had to bring in solution providers that could provide Orin platform development. In addition, Mercedes-Benz projects in China are even seeking help from other suppliers. Mercedes-Benz is continuously strengthening its local software research and development capabilities in China. It is reported that in order to ensure the strengthening of Mercedes-Benz's R&D capabilities in China, by the end of 2023, the size of Mercedes-Benz's R&D team in China is expected to reach 2,000 people, nearly double that of 2020.

This time, Huang Renxun recruited Wu Xinzhu under his command, perhaps because he valued his experience in full-stack systems, especially software mass production.

For NVIDIA, which has a market value of more than a trillion dollars, it is not satisfied with supplying parts. At the 2023 Snowflake Summit, Jensen Huang said that artificial intelligence is ushering in the era of software 3.0 (relying on data, algorithms and computing engines). Huang believes that unlike the "hardware-sales-software" strategy of the past, an emerging software landscape will emerge as the company shifts to a "software-sales-hardware" strategy. Nvidia wants to sell more software that only runs on its GPUs.

Resources:

"NVIDIA urgently needs Wu Xinzhu", BusinessCars

"Analysis of NVIDIA's Development Status and Market Size in 2022", Guosen Securities

"Dialogue丨Xiaopeng Wu Xinzhu: How the opponent plays cards will not disrupt our rhythm", LatePost

News Hotline & Submission Email: [email protected]