XPeng has revealed a new trump card

Author | Wang Xiaojun

Editor | Chai Xuchen

The internal competition in the automotive industry continues, and each company must keep playing its cards.

On November 6, this was XPeng's sixth Technology Day. For the past five years, it has been held on October 24, Programmer's Day. This year, for the first time upgraded to AI Technology Day, He Xiaopeng moved the main event to his alma mater, South China University of Technology.

On the lawn of South China University of Technology, XPeng showcased all its models, along with the futuristic flying car "Tomorrow Starship." Returning to his alma mater, He Xiaopeng presented several new strategies to respond to changes in the industry, signaling a restart of his ambitions.

In the gymnasium of South China University of Technology, He Xiaopeng began with the story of coding "hello world" from his dormitory, discussing the belief in technology changing the world. The current explosive transformation of AI once again presents XPeng with an opportunity to change the industry.

Returning to the core business, He Xiaopeng spoke about range extension technology, flying cars, the "Turing" chip, humanoid robots, advanced AI driving, and AI operating systems. Each topic seems worthy of its own press conference.

Compared to last year, XPeng's situation this year is completely different. At that time, XPeng was in a relatively low phase in terms of finance and sales, and after organizational changes, it was still in a phase of morale recovery.

This year, the MONA M03 has become a hit, allowing XPeng's delivery volume to continuously surpass 20,000 units, sweeping away past gloom and boosting internal morale. He Xiaopeng, having regained confidence, finally wore a long-lost smile.

After experiencing success again, He Xiaopeng is not content to be limited to car manufacturing and entering the range extension track; instead, he aims to tell a bigger AI story to the market. Among them, chips have become a key hardware component for XPeng's continuous technological upgrades in the next phase, making it an important part of this press conference.

At the end of August, XPeng officially announced the successful tape-out of its chip. At this press conference, XPeng also brought new developments regarding the chip. In October of this year, XPeng's Turing AI chip successfully ran the latest version of intelligent driving functions, completing as many as 2,791 function verifications in just 40 days.

The XPeng Turing AI chip has become the world's first chip that can be applied simultaneously in AI cars, AI robots, and flying cars. It is custom-designed for large models, featuring a 40-core processor that supports the operation of large models with 30 billion parameters, achieving the effectiveness of mainstream chips "one chip equals three chips."

Based on the iteration of hardware capabilities, XPeng officially upgraded its autonomous driving brand to "Turing AI Driving." This system centers around AI large models and includes self-developed cloud and vehicle-side large models, AI chips for large model development, and underlying architecture. He Xiaopeng also introduced that in the future, XPeng's intelligent driving products in China and globally will be uniformly named "NGP," no longer distinguishing between highways and cities.

In addition to the interconnected chips, XPeng also has some more technological side businesses, such as Robotaxi, flying cars, and AI robots.

At the end of July, He Xiaopeng revealed plans to enter the Robotaxi field through collaboration. He continued to tease that the currently developed "Ultra" model will present a new form of Robotaxi to the industry. It will be equipped with multiple XPeng Turing AI chips and the XPeng Cang Hai platform, boasting up to 3000T of vehicle-side computing power and L4-level intelligent driving capability, with the future steering wheel shifting from "mandatory" to "optional." Riding the tailwind of the low-altitude economy, XPeng's "million-dollar luxury car"—the split-type flying car—will make its global debut at the China International Aviation and Aerospace Exhibition in November 2024, with pre-sales officially starting in December.

This time, XPeng's humanoid robot has also made more progress.

This AI robot is named Iron, featuring a 1:1 human-like posture and over 60 joints, capable of simulating human standing, lying down, and sitting. The various technologies used in the XPeng AI robot are derived from the same source as the AI cars, such as the XPeng Turing AI chip, which provides it with an intelligent brain capable of independent thinking and even reasoning.

Although the robot did not walk a few steps as the audience expected, the XPeng AI robot has already entered factory work. In the future, the XPeng AI robot will also be deployed in sales stores, offices, homes, and other scenarios.

Clearly, AI has become the core of He Xiaopeng's grand narrative, linking AI cars, Robotaxi, AI robots, flying cars... illuminating the galaxy of XPeng's imagination.

However, back to reality, if XPeng wants to realize the above vision, it must break the shackles of sales and scale the heights of production.

In the past, He Xiaopeng was a staunch follower of pure electric vehicles. This time, after gaining insight into new market changes, XPeng has made a new choice by officially launching the XPeng Kunpeng Super Electric System.

Under this system, "Kun" represents the super range-extending system, utilizing next-generation range-extending technology, while "Peng" represents XPeng's new generation of pure electric systems. He Xiaopeng introduced that the XPeng Kunpeng Super Electric System has a pure electric range of 430 km and a comprehensive range of over 1400 km. This system is based on an 800V high-voltage silicon carbide platform, equipped with 5C supercharging AI batteries, hybrid silicon carbide coaxial electric drive, a silent range extender, as well as AI battery doctor and AI power functions.

The reason for entering the range-extending track is that XPeng wants to sell cars to more countries and regions, as many areas globally do not have the convenient charging infrastructure found domestically. The range-extending route provides more energy replenishment options in addition to building supercharging stations.

Currently, XPeng is not yet profitable, and significant investments in various business areas have continuously tested its financial situation. Although delivery volumes have reached new highs, industry competitors have already moved on to the next challenge. XPeng is now entering the range-extending track, but this is already a mature field with strong competition.

He Xiaopeng also admitted that XPeng is on an upward slope in the fourth quarter of this year, and 2025 may be a leveling off. The elimination rounds after 2026 will be even more intense, and XPeng needs to continue evolving to remain a player at the table. Surviving healthily is the only way to become a leader in the AI field.

Below is the transcript of the dialogue between media such as Wall Street Journal and XPeng Motors Chairman He Xiaopeng, Head of Autonomous Driving Li Liyun, and Head of Assembly Gu Jie (edited):

Q: How do you think XPeng's overall state compares to the past, one year and two years ago? What is your personal perception?

He Xiaopeng: Speaking of changes, in February and March of this year, I told our executive team that I believe XPeng will be on an upward slope in the fourth quarter. We need to put in more effort. I think the fourth quarter of this year to the end of next year will be a period of stabilizing and consolidating after reaching a height, laying the foundation for the next phase of growth I think this is a challenge. We need to build a solid foundation and a good system, focusing on management of the foundation, quality, and talent. Looking ahead, it may be around 2026 or 2027 when we see the basis for the knockout stage. The knockout stage does not necessarily mean the advancement stage; during the knockout, there will still be rankings for first, second, third, and fourth places. The competition in China's automotive technology and globalization will last for many years, and we are far from the time to relax.

Q: What are XPeng's future talent development plans? How will you promote XPeng's development?

He Xiaopeng: Over the past 24 months, we have done a lot of reflection, especially in talent management. I think the first step is to improve compensation, the second is to globally attract good talent, and the third is to enhance management methods to activate them. In recent months, I am particularly happy that we have good talent joining XPeng, including those in autonomous driving, power, and marketing.

For example, BYD has a million employees, and my management level is far from that height. XPeng will also be a company with tens of thousands of employees in the future, and we look forward to and welcome more people to join. Just now, a student from the robotics team told me that after today's press conference, he can help recruit more global talent in robotics, autonomous driving, and power.

Q: We just launched the Kunpeng Super Electric. I would like to understand the strategic considerations behind it and its special significance for XPeng's product overseas.

He Xiaopeng: We have been thinking about this for many years. First, we must achieve technological differentiation to obtain different customer value. I believe there are several areas where pure electric vehicles do not perform well. First, in some countries where there is not enough electricity, their infrastructure development may be slower than we imagine.

The second point is that companies like XPeng, which can solidly achieve long range, know that achieving a maximum range of 800 km with pure electric is sufficient. To reach a range of 1000 km, for example, our current Super Electric aims for a range of 1400 km, we still need to change the energy approach. Today, we may use oil + electricity, and in the future, we may use hydrogen + electricity to achieve a range of 1700 km. This presents a significant opportunity in broader areas where infrastructure is not well developed.

In the future, XPeng will gradually adopt a "dual energy" approach, transforming electric into "electric plus super electric," combining two energy methods for our customers.

Q: Today, many people asked about the range extender. There is a saying that pure electric vehicles are dancing with shackles, while range extenders have removed the shackles. What are your thoughts on this statement? Also, what are the differences or advantages between starting with pure electric models and then moving to range extenders, versus starting with range extender models and then moving to pure electric?

He Xiaopeng: I do not quite agree with the first description. I believe that super electric and pure electric cater to two different user groups, with 90% of their characteristics being different. In our multiple surveys, we have reached the same conclusion, which is even more evident in globalization. For example, European users may prefer pure electric, and I think both are feasible.

From my perspective, I think it is a bit easier to develop super electric from pure electric. Because transitioning from pure electric to super electric, those original components can be handled by traditional partners, so I believe it will be slightly simpler. However, both have high technical barriers and difficulties Question: Today's press conference was packed with information, with five major areas advancing simultaneously. XPeng has made significant investments in its current models, with MONA investing 4 billion, and the push for AI also requires substantial computing power investment. This year, there will be a significant increase in personnel, but XPeng is still in a loss-making state. How will the company strive to improve product gross margins in the future?

He Xiaopeng: First, all companies that want to excel in AI, including AI cars and robots, need a lot of "bullets"; otherwise, they will definitely not succeed. Second, relying solely on scaling up to achieve higher profits may have been valid in the past, but it is not applicable at this stage. A larger scale does not necessarily mean better gross margins.

So, over the past two years, we have done a lot. We avoid unnecessary spending and increase spending where it is needed, focusing on platformization. As I mentioned earlier, with super electric vehicles and dual-energy cars, there are many platformization efforts. Why have we achieved AI Eagle Eye in autonomous driving today? If we only consider China and specific regions, it is still possible to excel in autonomous driving.

However, if we consider doing well across the entire country and globally, using previous logic, the testing costs, data costs, and certification costs are very high. Therefore, we must think from the planning, scheduling, architecture, and platform levels.

Question: Our models priced above 40,000 euros have become the best-selling in Europe this year. Will the EU's anti-subsidy tax affect us? How do you think Chinese brands can better enter the European market?

He Xiaopeng: First, the Chinese Ministry of Commerce has been organizing all Chinese automakers to respond, and we will certainly wait for the government's unified planning. All manufacturers are waiting for the results of the unified negotiations between the Ministry of Commerce and the EU, and we will definitely follow and implement them.

All Chinese manufacturers should see that ultimately, Chinese cars will go global and achieve win-win situations with local markets. Whether it's Volkswagen or Toyota, they sell well globally, but most of the time, they produce and sell locally. So from another perspective, when Chinese car companies and other manufacturing enterprises enter the global market, they should achieve win-win situations with local markets to help themselves gain greater value. I often say that openness is both a mindset and a win-win situation. I hope that in the next ten years, more Chinese manufacturers will go global and become globally recognized brands.

Question: Nio's Firefly is specifically designed for overseas markets. The range extension technology we are promoting, including several upcoming models, is quite competitive in Europe. Do we have any special plans for overseas markets?

He Xiaopeng: We will not have an independent brand specifically targeting the global market; we will launch some of our Chinese models in the global market. The XPeng G9, G6, and X9 models will be launched in certain countries, with the X9 being more popular in Asian countries, so we will only launch it in Asian countries. There will also be cars for both China and the global market in the future.

Question: Recently, we saw that the XPeng MONA M03 sold exceptionally well. Was this surge in orders anticipated? What do you think has made the XPeng MONA M03 so popular with consumers?

He Xiaopeng: The sales performance definitely exceeded my expectations, thanks to customer love and the hard work of our team. Therefore, we are working hard to increase production capacity, hoping to accelerate our production capacity faster than our competitors Our most important task moving forward is to strive to turn every car into a hot seller. We will do our utmost for tomorrow's P7+.

Q: In the past, the focus on XPeng has always been on intelligent driving. This year, starting with MONA, there has been frequent mention of XPeng being in the first tier of pure electric energy consumption management. Can you share XPeng's technical philosophy on energy consumption management? How are these philosophies reflected in the products?

Gu Jie: XPeng's energy consumption performance is evident in the market; not only are our internal test data very good, but user feedback is also excellent. Before defining product energy consumption and range capability, we clearly know what our goals are. The best brand for pure electric energy consumption management is Tesla, which is well recognized in the industry, and we have always regarded Tesla as our target to catch up with.

From the G9 in 2022 to the G6, X9 in 2023, and now the P7+, we continuously optimize the efficiency and energy consumption of various subsystems using platform solutions and the SEPA2.0 architecture, achieving very low overall vehicle energy consumption.

Q: XPeng models will no longer differentiate between Pro and Max for standard intelligent driving, and there will be no need for subscriptions or fees. Is this a significant cost pressure? What work has been done behind the scenes to support this?

Li Liyun: After XPeng shifted to an end-to-end large model, each version in the future will allow users to feel the evolution of capabilities more quickly. Under the entire platform approach, XPeng will equip all models with advanced intelligent driving, and under the vision of technological equality and inclusiveness, XPeng will achieve standard intelligent driving across the entire lineup without differentiating between Pro and Max.

He Xiaopeng: Previously, we believed that advanced intelligent driving was a value-added service that required payment. If advanced intelligent driving becomes a standard service, you will find that there will be new possibilities for value-added services in the future, meaning it serves as the foundation for new value-added services. Therefore, the standard for AI cars is to be equipped with high computing power and advanced intelligent driving as standard; these are fundamental, otherwise many capabilities cannot be realized.

Q: Currently, the industry has both end-to-end + VLM visual language model solutions and companies proposing visual language action model solutions (VLA) that focus more on vehicle-side models. What are XPeng's thoughts on vehicle-side model capabilities in the next generation of end-to-end technology? How to establish stronger scene understanding capabilities?

Li Liyun: First of all, we do not comment on technical concepts and competitors' solutions. During the 1024 Technology Day and today's AI Technology Day, we showcased our true automated driving "end-to-end four-part series" and how we will transition from extreme assisted driving to automated driving and driverless driving. Our logic is actually the same as that of the world's most advanced AI technology company, OpenAI; we call it "heroes see the same thing."

The core of future competition lies in the cloud. XPeng has also stated that the current vehicle-side computing power is limited. We must first explore stronger capabilities in the cloud and then achieve a dimensionality reduction process on the vehicle side. This is our logic and methodology for maintaining a leading position.

He Xiaopeng: The Scaling Law actually applies to automated driving as well; it raises the upper limit of technology and, to some extent, supplements a certain lower limit. However, rules can also effectively control the lower limit of intelligent driving, meaning that both end-to-end and rule-based solutions can achieve good intelligent driving Currently, XPeng's end-to-end foundation model (cloud-based large model) is also used in robotics, sharing the same technical origin. The cognitive tasks for robots are far more complex than those for the small brains of autonomous driving, so both end-to-end and rule-based approaches could be valid. Ultimately, it depends on who can navigate the long-term journey most smoothly, steadily, and achieve the best results. As long as one reaches the destination, all roads lead to Rome.

Q: XPeng often mentions the desire to remove LiDAR when achieving autonomous driving. Why not retain LiDAR for high-premium and high-spec models, but instead go all-in on removing LiDAR?

He Xiaopeng: In fact, having too many choices is not beneficial for customers, companies, or the industry, as more choices mean slower upgrades.

From our perspective, in the field of end-to-end intelligent driving, consistently using camera solutions will inevitably surpass LiDAR solutions in a relatively short time. However, for high-level autonomous vehicles, I believe LiDAR is still necessary from a backup redundancy perspective when there is no human takeover.

Q: Is XPeng completely shifting to a pure vision approach? Both pure vision and human driving can have errors or misjudgments; how can safety be ensured? Tesla was the first to adopt pure vision; how can we compete long-term with companies like Tesla that adopted pure vision early on?

Li Liyun: The end-to-end large model primarily aims to maximize the visual capabilities of cameras, and we still have active sensors. Human drivers can also experience false brakes and sudden stops; at this stage, AI driving cannot completely avoid false brakes or sudden stops. AI is different from humans; it is less likely to be distracted or influenced by emotions. We have better data and engineering systems that allow us to train AI more effectively.

In this process, we still have engineering ethics and rules that can provide good support and supplementation. Although achieving zero false brakes and zero sudden stops may not be entirely feasible, there are indeed times when the environment changes dramatically, such as when something suddenly rushes out. However, we believe that with the support of large models, AI's intelligent driving capabilities will continue to improve.

Q: We expected a robot to appear at the press conference; why wasn't a robot arranged to take a few steps today?

The robot did not make an appearance today because I did not want to put too much pressure on the team. It is easy for a robot to take a few steps, but achieving many capabilities well requires support from different teams. Part of the XPeng robot is being developed by the power team, while another part is being handled by the team under Liyun. These capabilities need to be integrated, and I believe that a more comprehensive integration will be achieved at some point next year