The "Ultra" Revolution of the All-New XPeng P7

Author | Wang Xiaojun

Editor | Chai Xuchen

Three self-developed Turing AI chips, an 800V high-voltage platform, 5C supercharging capability, and a full range of dual-chamber air suspension—these configurations, once found in million-level luxury cars, are now standard on a domestically produced electric coupe starting at 219,800 yuan.

On the evening of August 27, XPeng Motors' intelligent manufacturing center in Guangzhou was brightly lit. The all-new XPeng P7 was officially launched, with four models introduced: the 702 Long Range Ultra priced at 219,800 yuan, the 820 Ultra Long Range priced at 239,800 yuan, the 750 Four-Wheel Drive High Performance Ultra priced at 259,800 yuan, and the 750 Four-Wheel Drive Peng Wing Ultra priced at 301,800 yuan. The main differences among the first three versions lie in their range, while many core functions are standard across the entire lineup.

When XPeng Motors Chairman and CEO He Xiaopeng announced the pricing of the new P7, the audience erupted in enthusiastic applause. The cheers and applause on site also proved the competitiveness of the all-new XPeng P7. Soon, this competitiveness was reflected in the orders. Within 7 minutes of officially opening for orders, the number of large deposits exceeded 10,000, setting a new record for XPeng.

Setting aside the momentum and returning to the product itself, the all-new P7 is also a new manifestation of XPeng's more than ten years of experience in car manufacturing and smart technology accumulation. Nowadays, industry evaluations of XPeng's new car are no longer limited to the previous technical dimensions but give high ratings across every dimension.

In terms of design, since its debut, many people from both inside and outside the industry have been impressed by the car's aesthetics. The all-new P7 adopts "The Next Design," an original design with extreme futurism, showcasing an ultra-wide sporty stance. The original five-surface design eliminates complex feature lines, naturally integrating functions such as light groups, sensors, and tailgates into the surface stitching, achieving a unity of aesthetics and functionality.

The body proportions also meet the standard level of high-performance coupes, with a length-to-height ratio of 3.57, a width-to-height ratio of 1.40, a wheel width ratio of 0.28, and a wheel height ratio of 0.52. At the press conference, He Xiaopeng also compared several indicators with models like the Porsche Panamera.

Beyond aesthetics, the updated XPeng begins to tell its story from more dimensions.

Before the official launch, XPeng took the all-new P7 to the Zhejiang International Circuit, where it set a lap time of 1 minute 38.767 seconds, becoming the fastest mass-produced coupe under 500,000 yuan, even surpassing some million-level models. This demonstrates its strength in driving control and performance.

At the press conference, He Xiaopeng explained that the performance is attributed to the perfect combination of supercar-level mid-mounted electric drive layout, 50:50 axle load distribution, dual-chamber air suspension + AI intelligent chassis. Additionally, the vehicle passed the 90 km/h moose test with a braking distance of 33.16 meters, with Brembo calipers and a mixed tire setup providing track-level safety redundancy In addition, the all-new XPeng P7 series is equipped with dual-chamber air springs and DCC intelligent damping as standard, allowing for millisecond-level adjustments in stiffness and height, balancing comfort and handling. Furthermore, the new car is equipped with front narrow and rear wide tires, featuring a front 245 and rear 275 sports tire combination, providing solid grip to turn these achievements into reality.

Beyond challenging the track, XPeng also conducted a 24-hour endurance test on the all-new P7. At the launch event, He Xiaopeng showcased his communication with Lei Jun regarding the endurance test.

Before the car's launch, it completed a 24-hour endurance test under the witness of authoritative institutions, achieving an average speed of 220 km/h over a distance of 3961 km, with a total of 32 charging sessions, each lasting 11-13 minutes, setting a new benchmark for industry endurance and charging efficiency.

This reflects XPeng's capabilities in the three electric systems.

The all-new XPeng P7 is the first to feature the 858 high-performance battery pack, integrating an 800V high-voltage platform, 820 km of range, and 5C ultra-fast charging AI battery. Among these, the 800V 5C ultra-fast charging system is one of the car's biggest highlights. It can recharge 525 km in just 10 minutes, with a SOC charging from 10% to 80% taking only 11 minutes. Notably, this system can automatically adapt its boost technology to public charging piles, requiring only 22.5 minutes to complete a 10%-80% charge even on a 180 kW State Grid pile.

In terms of intelligence, the entire P7 series is equipped with three self-developed Turing AI chips previously featured in the G7 Ultra version, achieving an effective computing power of 2250 TOPS, which He Xiaopeng believes meets the computational needs for future L3-level assisted driving.

With the support of high computing power, the all-new XPeng P7 can efficiently drive local VLA large models, creating an AI intelligent driving assistant that can think. The vehicle possesses reasoning and prediction capabilities closer to those of human drivers, allowing it to handle complex road conditions with ease.

Also due to the support of computing power and other aspects, the all-new XPeng P7 is the first to feature a slippery road AES, enabling emergency steering avoidance in low-traction conditions. Additionally, the enhanced AEB's working range covers speeds up to 130 km/h, combined with pre-activated smart airbags and ECALL automatic rescue.

At the cabin level, the all-new XPeng P7 has achieved a leap from "intelligence" to "wisdom." The vehicle is equipped with a VLM cabin all-round butler and an 87-inch AR-HUD, with a three-axis dynamic central control screen that supports multi-angle and multi-mode adjustments. This allows the central control screen to sway left and right and nod up and down, responding to commands and even expressing emotions like a partner.

Interestingly, the "Road Rage Relief" feature introduced during the debut sparked heated discussions among netizens. This feature combines HUD and projectile launching, allowing drivers to "fight monsters while driving." After customizing the steering wheel buttons, drivers can click to trigger the "monster-fighting" function, throwing toy packages at surrounding cars. Some netizens previously expressed concerns that such a feature might distract drivers and affect driving safety.

However, He Xiaopeng stated at the launch event that every new feature has faced such doubts upon its release, but the industry is progressing and has considered these issues, implementing sufficient safety guarantees In fact, for XPeng, this is not an ordinary press conference, but a concrete manifestation of XPeng's comprehensive upgrades and iterations in products, operations, and organization.

August 27 is also a day of great significance for XPeng Motors. On August 27, 2017, He Xiaopeng officially joined XPeng Motors; on August 27, 2020, XPeng Motors was listed on the New York Stock Exchange; and last year on August 27, the XPeng MONA M03 was launched. These three dates mark new milestones in XPeng's history.

This time, XPeng, leveraging its advantages, gave the new P7 a grand presentation. The day before the press conference, the XPeng Technology Park was officially lit up, with seven new P7s neatly arranged. On the day of the press conference, a congratulatory message appeared on the big screen of the New York Stock Exchange, celebrating XPeng's five-year anniversary and wishing for the successful launch of the new P7.

Facing this highly anticipated new car, He Xiaopeng's goal is to "maintain third place and strive for second" in the segmented market. With the delivery of this car, achieving a monthly sales volume of 40,000 units for XPeng does not seem difficult. This also means that achieving quarterly profitability in the fourth quarter is not in doubt.

Below is a conversation with XPeng Motors Chairman and CEO He Xiaopeng, and Nick, the product head of the XPeng P7 model (edited without altering the original meaning):

Q: Why does the new XPeng P7 come standard with Ultra across the entire range? Why was this choice not made for the G7?

Nick: (We) have our own positioning considerations for each model.

First of all, the P7 model has a totemic significance for XPeng Motors, and we believe that "the best" should be placed on this model.

Secondly, from the beginning until now, the XPeng P7 is a crucial model for the XPeng brand. We believe it must be very simple and pure, representing the forefront of brand technology, so we chose a very straightforward "full series Ultra," including full air springs and many configurations that are part of the entire series concept. We feel the need for a model that represents the brand's tone, including our totem and overall development situation. Therefore, this time we chose a very direct strategy that represents the highest level of technology.

Q: In this industry, when sales are poor, everyone turns to making full-size SUVs. However, XPeng has continuously launched three successful sedans, each one more successful than the last, from MONA to P7 to P7+. There must be a mature methodology or understanding of the product behind this. What do you think is XPeng's biggest advantage and unique entry point in making sedans?

Nick: Our understanding of sedans starts with aesthetics first, matched with corresponding technological attributes. Whether it's MONA, including this year's MONA M03 MAX, P7+, or the new P7, these three products all have: first, a fantastic design, and second, excellent technological attributes in this segmented market.

The M03 has almost unique intelligent capabilities in this segment, and the same goes for the P7+. Our Ultra capabilities, along with the upcoming VLA+VLM large models, are also excellent vehicles in this segmented market. So overall, the first is aesthetics, and the second is technological attributes; we have excelled in both areas in this segmented market, which is why we can see that all three of our sedans have performed very well In this niche market, both the volume and overall efficiency are very good.

He Xiaopeng: Every car we launch has a summary, but we haven't concluded that there is one methodology that can guarantee success; each model has its gains and losses. This year is a major transition year for XPeng, while 2026 and 2027 will be years of new products. In the next financial report, we will inform everyone how many new products will be launched next year, which will exceed everyone's expectations. We will have a layout for product lines ranging from over 100,000 to under 2 million, including full-size SUVs. We are working hard to learn and think about which aspects we can improve. I believe there are many commonalities, such as how we are currently enhancing aesthetics and advancing autonomous driving to the next generation, approaching L4. These are all things we are working on. In the future, everyone will see us making progress in design, technology, quality, and product development. I hope we continue to uphold a constant PDCA attitude.

In the automotive field, no new car companies have been evergreen in the past three years, and we still have many shortcomings. Therefore, we are constantly identifying our weaknesses and addressing them, while also finding our strengths and making them even stronger. This is what we are currently doing.

Question: To some extent, XPeng is the company that resembles Tesla the most globally in terms of products. XPeng has everything Tesla has, and in some experiences, XPeng even does better. However, looking at the capital market now, the market capitalization of the two companies has a difference of 50 times. Do you feel wronged in your heart? Is XPeng underestimated?

He Xiaopeng: There is a significant valuation gap between Chinese and American technology companies, roughly a 7-fold difference, which is somewhat similar to the difference in exchange rates. Chinese companies, including Chinese e-commerce companies, previous search companies, IM (communication) companies, and AI companies, this is a very important point.

Secondly, I believe that our power in certain technology indices has not yet been reflected compared to our competitors. For example, I mentioned in the financial report that XPeng will also launch RoboTaxi capabilities next year. In fact, we have only recently announced these capabilities, and I believe that as time goes by, give us another six months, everyone will see our progress and completely different thinking in this area, and at that time, everyone will have a higher recognition of XPeng.

Question: This year is a major year for XPeng to refresh its models, and next year and the year after may be major years for new launches. With the high frequency of new models appearing, is there a plan and strategy for the old models to exit? How is the management of the supply chain and the allocation of production resources handled for the new models to achieve the effect of many new models being launched and delivered simultaneously?

He Xiaopeng: In the next two years, XPeng will have many all-new models, many of which will be different from the current cars, with layouts that have not been seen before in this positioning and category. For example, the G6 will also have a 2026 annual facelift next year, and we have reanalyzed that the 2026 annual facelift does not need to change that much; instead, it should shift from "physical value" to "emotional value." In the future, we will see us exerting greater power in the emotional value facelift, making all car owners find it interesting. This is what we are currently working on The money spent is not much, but it's very cool.

The core of the second question is to turn "platformization" into "modularization," and then have very good overall planning of production specifications, as well as good systems and processes, and effectively supervise and manage this system to solve the second problem.

Q: Is XPeng's production capacity sufficient? When can it really be elevated to the next level?

He Xiaopeng: As we move forward, car manufacturing companies, including new companies like ours, will have stronger production capacity because our modularization is getting better and better. This is something I am very confident about. As I mentioned at the beginning, the goal for the P7 is to be among the top three in its class, and we are indeed preparing production capacity with that goal in mind. You can see that this factory currently has the P7, and there are actually many other models as well. This factory is quite large, and I believe there will be more models in the future. We think the production capacity here is sufficient, and we are very confident about this.

Q: When you set this pricing strategy gradient, with 20,000 as a gradient, what considerations were behind it?

He Xiaopeng: In terms of commercial pricing, we still return to what we just talked about, which is the ultimate experience. For example, why do we equip the entire series with Ultra and dual-chamber air suspension? We actually hope to achieve the ultimate experience overall. Today, I also mentioned that we discussed the price of the all-wheel drive for a long time. Initially, the price we discussed was significantly higher than it is now, but ultimately we felt that this car, if you drive it with all-wheel drive, is just too cool; it’s really comfortable, powerful, and enjoyable. At that point, we thought, should we lower the price of the all-wheel drive a little bit? Eventually, we painstakingly managed to lower the price of the all-wheel drive. Because we adjusted the price of the all-wheel drive, we also made some adjustments to the overall pricing, so it’s quite an interesting point. I believe the sales of the all-wheel drive version will be higher than we might have imagined; I am very confident.

Q: In today’s press conference, you spent a lot of time talking about the car, such as handling and chassis, while the previously emphasized AI intelligence seemed to be less. This is somewhat atypical; why is that? What are your considerations?

He Xiaopeng: As I said today, if you drive this car, it’s the one I’ve driven the most among all the cars I’ve tested recently. In the past, I would drive many cars for a while using autonomous driving, and then switch to manual driving for a while, but once I get in this car, I don’t want to let go. This is an ultimate sports sedan, and I think it still needs to be driven manually. So today, you saw me spend a lot of time discussing its driving control, chassis, handling, and acceleration, which brings that adrenaline rush feeling.

(Everyone) will gradually see our advancements in intelligence by the end of this year or next year because we are actually crossing this era and hoping to enter the next one. Internally, we are all challenging ourselves, asking why we can’t use Thor or Orin to advance to the next generation by next year. I said that by next year, everyone will see that we need to move to the next generation, and in between is like a generation and a half; it’s not high enough, but it’s also not very low, which makes it quite painful and struggling. After making cars for so long, I later decided to prioritize hardware, putting the hardware in first, and the software providing a version that is consistent with the current version's capabilities, which is sufficient but not yet at the ultimate level of usability But the later time will provide everyone with an extremely useful experience. Once I have basically developed those software capabilities, I will specifically report and share with everyone. There are many features that you will only feel are very useful after using them; it's quite difficult to imagine at this moment.

Q: Currently, among the VLA, very few can reach the mass production stage. Li Auto said they invested 6 billion in AI this year to develop this thing. However, yesterday Yuanrong (Yuanrong Qixing) also announced their version, and Zhou Guang from Yuanrong said that a few hundred million could also achieve this. How much has XPeng invested in this? If the investment in VLA does not reach a certain level, will there be an insurmountable limit?

He Xiaopeng: XPeng's investment in VLA this year is also close to 5 billion, and spending several billion a year is necessary. A few hundred million might be enough to create a micro VLA, but it is essentially an end-to-end logic. The relationship between VLA and end-to-end can be considered as a kind of end-to-end. Secondly, end-to-end can actually be seen as a primarily two-dimensional structured information, a logic and model guided by data end-to-end. However, VLA is a logic that includes multi-dimensional space, including time dimensions, and is directly related to unstructured movement.

The real VLA may take some time for everyone to achieve mass production. The reason I have been saying recently that the real VLA will be at least ten times stronger than all current first-tier players (including XPeng) is that this is the true effect of VLA.

Q: VLM was pushed out two months later than VLA because, from the perspective of the overall functionality of VLM, it has a cross-domain integration nature. So I want to ask, if VLM is pushed out two months later, what impact does it have on the efficiency of the model that VLA pushed out first, or have we addressed this issue?

He Xiaopeng: VLM requires independent hardware. From a certain perspective, I believe today's VLA is strictly speaking VA, where L (Language) is very weak. Although it can see the text on the road, its reasoning logic for language is significantly reduced due to computing power.

VA is a fast model, but VLM is a slow model. The VLM we are currently integrating in the field is more like a VLT; it can derive tasks. Due to the combination of time and space, it can infer a set of tasks, such as "please get me a bottle of water." Getting water is a group of tasks. VLM and VLT are both reasoning-based slow models in the digital world, so it is a slow model, not a fast model. Therefore, VLA and VLM ultimately need to be combined; it is more like a fast "smart small brain" and a relatively slower "smart big brain" combination.

I believe the final addition of A and L will gradually be seen in XPeng by the end of this year or next year. It actually consists of two models but will gradually integrate. Just like the human brain has three parts: the brain, the cerebellum, and the spinal cord, automotive integration also divides into these three types, with different Hertz, different functions, and different areas of the brain. Many people ask why you can't put them together? In fact, putting them together is actually not energy-saving; the energy consumption is poor and not scientific, or follows a non-biological logic. So ultimately, I believe that within a car, there are three different brains for electric control, movement, and thinking.

Q: With the technological evolution of various companies in VLM and VLA, it seems that the level of assisted driving capabilities has been raised by everyone. In this case, will there be a phenomenon of homogenization or convergence among the solutions of various companies? How will XPeng maintain its own characteristics?

He Xiaopeng: I believe there are still many layers, but how to layer them, I will share my views on this issue with everyone within this year.

XPeng's VLA has been in development since October last year. We have never said that actually developing this function is not something that can be thought of today, worked overtime tomorrow, and completed the day after. I also don't believe that in the vast field of AI, it can be done well with very little money. There may be shortcuts, but in the long run, the foundational work must be done well. So I believe that next year, in 2026, China's automated assisted driving will make a huge leap, and at that time, the current first camp may all drop to the second camp, and a new first camp will emerge.

Q: This year, there has been significant development in the computing power of domestic large models, and GPUs have also made great progress. I would like to ask how our reserves in GPU are? What layouts do we have for training large models on domestic GPUs?

He Xiaopeng: We are moving quite fast; we are currently using both NVIDIA's and domestic GPUs. Over the past year, we have made significant changes to the cloud-based GPU and local inference computing power, namely Turing, as well as the cloud-based VLA model and local VLM model. By the end of this year, we will see the preliminary version of the VLA and VLM encirclement, so I am quite confident in the next AI competition. Ultimately, we need to grow quickly and be profitable. Because future AI investment will definitely be a huge investment, I have mentioned this year that R&D expenses for future car companies will need to reach 50 billion a year, with possibly over 30 billion coming from AI and only over 10 billion from hardware. So from another perspective, this is a process that everyone will have to sprint in global competition in the future, so AI investment is very important.

Q: You previously said that if you want to harm a friend, you should advise him to build a car. Lei Jun said you strongly advised him to build a car. I would like to ask why you strongly advised him to build a car?

He Xiaopeng: Everyone may not have seen the premise. What I initially said was that many buddies want to start a business, meaning they currently do not have a company, or they have sold their previous company and want to start anew in the automotive field, which I believe is very difficult. However, I talked to Lei Jun in 2017, and I believe Xiaomi is suitable for building cars because Xiaomi originally has the software gene of the internet and the hardware gene of the internet. These two genes have a certain degree of relevance to cars, and it is relatively high; I won't go into the specifics Question: On today's 8.27 anniversary, do you personally think your sense of achievement outweighs the pain of entrepreneurship?

He Xiaopeng: Because making cars is a never-ending marathon, like a Muay Thai match, any opponent can attack you from any angle with any move. You first have to be strong enough, and even if someone really knocks you down, you have to be able to get back up. Seeing people kick you every day feels different from when I was in the internet industry. In the internet, there are many hundred X battles, and once you get through them, you feel very comfortable, with a high happiness index. (But making cars faces) natural barriers, network effects, ecological effects, and first-mover effects. There’s no possibility of a double logic; many of the logics that existed before are gone. I advise you to listen to me: making cars is not a happy thing.

Question: For a long time, you have said that XPeng is not just a car company, but also wants to be an AI company. However, is there a paradox here? Because technology companies have high gross margins that support their investments, with net profit margins often at 20% to 40%, allowing them to invest heavily. Car companies, on the other hand, cannot do this; for example, the profit margin of Chinese car companies is only 4% to 5%. How can you support the investment in an AI company with such low profits?

He Xiaopeng: Before the era of digital industry technology companies, I believe the highest global research investment came from physical industry technology companies. The largest investment in China is from Huawei, which initially focused on hardware or hardware combined with software. If you ask whether Huawei is a technology company, I definitely think it is; last year, its research expenses were close to 180 billion (179 billion). In the past decade, digital companies (in R&D investment) will increase. I believe that the vast majority of digital companies are like the American Seven Sisters, such as Apple and Tesla, which are hardware combined with software companies. Therefore, I do not believe that only software or only digital technology companies can be called technology companies; physical companies can also have very strong technology companies, so they can also perform well. Moreover, physical companies have a new effect, which is the scale effect of hardware, something that does not exist in the digital world. Future technology companies will have both types, and the second type is harder to achieve because it requires not only good software but also good hardware, while simultaneously strengthening user operations and customer sales. This capability is a greater challenge than that faced by pure digital world giants, and of course, it is also a bit more arduous