Cars. They used to be really simple. They were vehicles that you drove from A to B. You turned the key, and away you went until you needed petrol. You filled up and kept driving. For entertainment, you listened to the radio, cassette, CD or eight-track. You got your car serviced. Engine oiled, tuned. Really simple. But now? It’s a very different story.
Here are just some of the recent comments that have been made about the latest version of cars by technology leaders. A new “computing space”. A new “paradigm for transport”. A machine “you can have a natural conversation with”. And even a “special kind of robot” in a world “where screens are the new horsepower”.
Interestingly, the latter comment was made by a very senior research scientist at a mobile processor developer, and while one could argue how fanciful the proposition is, it is a revelation of just how connected the robotics, automotive, connectivity and IT industries are and will be.
Not only is a car now a network on wheels, but it’s also a hub of an array of information, entertainment, AI, navigation, safety and even driving applications. Oh, and potentially able to fly – something that will likely be made possible by the XPeng brand.
It may actually be a little unfair to simply describe Guangzhou-based XPeng, founded in 2014, as a car company. From the perspective of actual products, the company regards itself as working in the field of intelligent mobility with the mission of exploiting the full potential of mobility from vehicles on the road to even vehicles in the air. Moreover, it very much stressed that it does not just move people from A to B. Instead, it says it wants to “inspire new life experiences” and offer better, more innovative ways to move.
The manufacturer also regards itself as a global physical AI company, dedicated to bringing artificial intelligence (AI) into the physical world to reshape future mobility and smart living. Indeed, speaking in August 2026, XPeng vice-president Jacky Gu said the company could be justifiably described as a “physical AI-driven next-generation mobility and robotics company”.
Physical AI has rapidly established itself as the cornerstone of the modern manufacturing world. Indeed, research in July 2026 found that over three-quarters (77%) of manufacturers expect physical AI to have a significant or transformational impact on warehouse operations, while 75% anticipate the same across assembly and manufacturing.
Long-term investment
The study also stressed how just as enterprise-wide deployment of physical AI was ramping up, manufacturers were committing to long-term investment in a fast-growing area. As manufacturers increase their AI investments, the focus is shifting from standalone automation projects to larger physical AI ecosystems as part of a people-first transformation. That is, instead of replacing the workforce, manufacturers are using intelligent systems to help employees work more safely, efficiently and productively while supporting workforce redeployment where needed in a human-plus-AI operating model.
This fits in with the XPeng philosophy as it goes about its business. The company’s physical AI footprint took its first steps in 2017 with the G3 APV car, followed by the P7 XNGP vehicle in 2022; the P7+ AI car in 2024; GX VLA 2.0 in 2025; and the current flagship G9L, set for launch later this year.
Through in-house R&D, XPeng has developed a full-stack physical AI architecture spanning AI chips, world foundation models, and integrated software and hardware applications. This unified technology foundation of XPeng powers an expansive product portfolio of smart electric vehicles, Robotaxis and humanoid robots, advancing the deployment of physical AI at scale.
At the heart of all of the technical AI offer is XPeng’s self-developed processor, the Turing chip, designed specifically to support large-scale AI models in physical products. It forms part of XPeng’s full-stack technology architecture, bringing together computing hardware, AI models, software and product systems.
The reasons for going self-developed in this sphere were clear for XPeng. Rather than adapting its AI technology around a general-purpose third-party processor, the company was confident that its approach meant it could develop the chip alongside the models and applications it is intended to run.
From a basic technological perspective, the chip comprises a 40-core processor with a domain-specific architecture designed for neural networks, including two neural processing units developed in-house, representing the processing “brain”. There is also an integrated neural network based on a domain-specific architecture (DSA).
This all forms the bedrock of support for AI models containing up to 30 billion parameters – what is regarded as the maximum operable parameters for large models – and what XPeng assured would be computing performance equivalent to three conventional high-performance automotive chips.
The number of chips used, and system configurations, varies by product. In terms of where it will be applied, the Turing chip is designed as a multi-platform processor for AI vehicles, AI robots and flying cars, thus support XPeng products across different formats, including intelligent standard auto vehicles plus the Iron humanoid robot range and the futuristic Aridge flying-mobility family, supported by the Land Aircraft Carrier X3 vehicle.
For example, in “normal” XPeng automotive vehicles, the chip is designed to provide the computing capacity needed to run supported AI applications, including intelligent-driving and smart-cockpit systems. This, says the company, will enable large AI models to be deployed directly in production vehicles and supports faster model inference.
XPeng stresses that energy and power is a critical foundation of physical AI with higher performance empowering physical AI to achieve in cars longer range, higher speed, greater torque and superior agility of vehicles. Gu describes this motion control as the “cerebellum” of physical AI, determining how gracefully, efficiently, intelligently and safely physical AI behaves in the real world.
As regards the tangible benefits that such technology advances, the company believes its advances in the field have seen maximum cornering grip improve by as much as 20% compared with previous vehicles, while cornering understeering has fallen by a similar amount. In terms of driving safety, XPeng claims that cars can now drive at a continuous obstacle avoidance speed 5% faster than before, with the breaking distance at 100km/hour falling by 3%.
As much as they seem totally futuristic, the Iron robot range is designed to coexist with people: not as a gadget, but what is hoped to be a “helpful presence”, as the company looks to make artificial general intelligence (AGI) practical across general scenarios. In addition, it was keen to stress the importance of the humanoid form so that the machine could fit into everyday life applications and use cases. Moreover, this likeness was not just about form, but also about proportion, structure and tactile quality.
As is now becoming typical in the automotive IT and connectivity arena, the result of this is that processing AI workloads in the vehicle can also reduce reliance on cloud connectivity and allow information to be processed closer to where it is generated. That is, continuing the current trend of taking processing to the edge and on-device AI.
XPeng notes that the balance between onboard and cloud processing depends on the application and product configuration. Yet despite this variation, what is a standard aim across all vehicles is to provide an enriched global intelligent cockpit experience with driving safety incorporated into the chip architecture. To that end, the Turing chip includes an independent safety island for real-time monitoring, and is designed to meet automotive functional-safety and reliability standards, including ISO 26262, AEC-Q100 Grade 2 and Asil-D requirements.
As well as the Turing chip, XPeng also develops in-house its XOS in-car operating system, which is designed to connect the vehicle interface, voice assistant, infotainment, connected services and supported vehicle controls, providing the software foundation for an intelligent cockpit.
For optimal utility, XOS is designed to evolve through over-the-air updates. It can, for example, refine interfaces, introduce new functions and adapt the experience for different markets, with availability varying by model, hardware, software version and local regulation. The latest global release, XOS 5.8.0, introduced updates across comfort, safety, entertainment and personalised vehicle scenarios. XOS 5.8.0 can currently support 21 languages and more than 700 individual capabilities. Scenarios can be created through the in-car display or the official XPeng app, while preset modes cover areas including comfort, entertainment, safety and energy saving.
A key part of these personalised experiences is the ability for the intelligent cockpit to allow drivers and passengers to interact with the vehicle through voice commands as well as through the central display. The voice system’s AI supports continuous conversation and control of supported vehicle functions such as not needing to repeat a wake phrase for every request. The system is also built to respond to consecutive instructions while distinguishing what are considered “relevant” commands from unrelated speech.
Interpreting intent
To ensure capability, XPeng has integrated large AI models into XOS with the intention of allowing the system to interpret the intent behind a request rather than relying solely on predefined commands. To make the experience more streamlined, the cockpit experience uses XPeng’s AI assistant, Xiao P, which, depending on the vehicle and software configuration, can support a range of applications. These encompass natural and continuous conversation; contextual and indirect requests; control of supported cabin functions; reminders and scheduling; information and general questions; navigation, entertainment and charging-related functions; and select visual or environmental understanding.
XOS also allows users to create combinations of vehicle settings for different situations and build personalised scenarios. This means several cabin settings can be activated together rather than adjusted individually each time. A key potential use case cited is the system interpreting words such as “I’m cold” as a request to adjust the climate controls. It can also support requests such as setting a location-based reminder or waking an occupant after a specified period.
Such capability is seen as being able to move the experience beyond conventional voice control towards an assistant that can understand context and carry out the appropriate supported action. XPeng believes that together, these systems create a cockpit that is conversational, configurable and capable of evolving after a vehicle has been delivered.
Gu revealed in August 2026 that its Smart Electric Platform Architecture (Sepa) platform was in the process of an evolution from intelligence to AI iteration, and ultimately towards embodied AI. Sepa integrates core hardware, electrical systems, infotainment and full autonomous driving software to lower production costs and speed up software upgrades.
Version 2.0 of Sepa was unveiled towards the end of 2023, and offered an electric powertrain supporting advanced driving assistance and its AI engine capable of achieving 77 trillion operations per second (Tops). Currently, Sepa 3.0 is configured for dual-energy and AI-based intelligent driving. Its Turing chips are capable of 3,000 Tops of effective computing power with Level 3 autonomous driving capability and power consumption down by 10% compared with the previous version of the platform.
AI based on the Turing chip also drives a vision-language-action (VLA) model and infotainment vision-language model (VLM) offering cross-domain convergence. The claimed net result of a Sepa 3.0-based architecture is five times smoother and safer driving compared with Level 2 autonomy, and 25% higher overall driving efficiency.
Gu also confirmed that Sepa 3.0 will feature in the forthcoming G9L range, which is described by XPeng as a new-generation SUV purpose-built for families. G9L is being attributed by XPeng as offering safer, smarter and sleeker driving with what Gu said would be “world-first breakthrough” innovations. In addition to a raft of AI-based infotainment technology and services, the new car range will boast AI-powered full-body adaptive seating, with 91 pressure sensors with a minimum sensing resolution of 2g.
Drive-by-wire steering wheel
Among the many new features in the vehicle is a first-generation drive-by-wire steering wheel that has a geometric configuration paired with satellite buttons. It also has a second-generation technology island, with a 17-inch control screen and an AR heads-up display with floating console.
Explaining how the company designs cars such as the new G9L with AI, connectivity and infotainment technology very much up-front, XPeng creative design director Joan Melenchon stressed how AI enhances a lot of functionalities that in the past the company never thought about.
“It influences because many devices that are required by the AI for the autonomous capabilities are not only affecting the exterior and the interior, not only through screens but also with sensors and rudders,” he said. “Nowadays, you can also guide and move through gestures. You can move doors and contacts that way. This all requires a lot, especially [in terms of] sensors and cameras that we try to hide as much as possible to not to impact [customers] psychologically and physically.
“Even the design has shifted from us in the past, when we were talking about a car with four wheels and a computer inside. Now, it’s the other way around. It’s a computer with four wheels. And this also affects the way we design it, the way we think the user experiences it and interactivity in the interior. And this defines a lot of new interior layouts, especially with [fewer] physical restrictions and a lot more flexibility in the interior.”
Melenchon does not believe this brave new world has made his job harder, nor easier: just more challenging and more fun: “With these new functionalities, it makes you think in a creative mindset to always face new challenges. I think what is important is what makes our customer’s life easier and then we need to provide solutions to make a customer’s life easier.”
Software development
From a software development perspective, the next part of the Sepa evolution will look to offer embodied AI and cross-domain integration. XPeng promises Sepa 4.0 will deliver “significantly” improved performance, offering in the region of over 3,000 Tops’ worth of processing power. Explaining this to Computer Weekly, Gu said: “We have some very common updates starting from the [3.0]. Maybe in the future it could be two times the current capability, [but] it’s not confirmed. We just see it will be over 3,000. [With the car], minimal human intelligence [required] is the target, but it will take some time.”
As part of the software evolution, XPeng has already confirmed that global delivery of its VLA 2.0 intelligent driving system is set to begin in 2027. It said XPeng asserts that the new VLA will “redefine the future of intelligent driving”, and that unlike traditional modular autonomous driving stacks, VLA 2.0 integrates perception, reasoning and action in a unified AI foundation model designed to improve cross-scenario generalisation while reducing reliance on high-definition maps and rigid rule-based systems.
The system is designed to interpret complex real-world environments and respond with driving behaviours that resemble those of experienced human drivers. With VLA 2.0, XPeng aims to enhance the comfort and confidence of intelligent driving, creating a smoother, more enjoyable experience for every journey.
Crucially, it is engineered to identify complex scenarios – such as erratic vehicles, accident scenes and uneven road surfaces – and respond proactively to support safer, more dependable driving performance. It is also the company’s first software version designed to achieve full autonomous driving, which XPeng expects to arrive in the next three years.
Technologically, the VLA 2.0 is said to differ with traditional models that operate through a sequential vision-language-action pipeline, in which visual perception is translated into language-based reasoning before generating vehicle actions. By contrast, VLA 2.0 is based on an end-to-end vision-to-action architecture constructed to enable the automotive system to translate perception directly into driving decisions, removing intermediate translation layers to improve both efficiency and response speed.
Full-scenario VLA 2.0 capabilities are intended for true “driving-anywhere”, covering campus lanes, rural dirt roads and non-navigable routes. The system takes into consideration complex conditions such as narrow lane passage and pothole avoidance, and starts from a standstill.
Like with the Turing chip, the AI foundation model behind VLA 2.0 is designed to extend beyond passenger vehicles. The same underlying architecture will be applied across multiple embodied AI platforms, including Robotaxi fleets, the Iron humanoid robots and the modular flying vehicle systems.
That said, the system is essentially still in development. Early testing is said to have demonstrated “robust” performance in dense urban environments and mixed-traffic conditions. In addition to its safety and stability, XPeng claims VLA 2.0 has improved driving efficiency by 23%. In measured tests during Guangzhou’s evening rush hour, its traffic efficiency was noted as having exceeded that of traditional L2 intelligent driving systems and Robotaxi models, achieving performance comparable to experienced human drivers.
XPeng’s first Robotaxi equipped with the VLA 2.0 has begun public road testing, and trial operations are scheduled to commence later this year. International road testing is also on the slate for the rest of 2026, with global deliveries of VLA 2.0 planned for 2027. Volkswagen will serve as the first customer for VLA 2.0 in the Chinese market.
Looking further into the future – or, more accurately, up – is the Aridge division, which XPeng claims is the world’s leading creator of low-altitude mobility services and the current largest flying car company – the latter a rather select group, as one could imagine.
The division has the stated mission of popularising low-altitude flights, enabling mass markets to enjoy flying. That is starting with the world’s first mass-produced modular flying car, launching products for scenarios ranging from personal flying to urban air commuting and end-to-end three-dimensional transportation.
‘Three steps’
The business has a “three steps” product strategy for its aviation business. That is: a modular flying car with a land aircraft carrier; a high-speed-long range tilt-rotor flying car; and an electric vertical take-off and landing (eVtol) flying car. The former is intended to facilitate leisure getaways around urban areas, while the high-speed, long-range tilt-rotor flying car is designed for efficient commuting between cities and surrounding areas, aiming to unlock a new era of advanced air mobility. The latter is described as no less than “facing the future of 3D transportation”, delivering door-to-door and end-to-end travel.
Operating from three facilities – test bases at Anhui Mingguang, Guangdong Chishui and Guangdong Huangpu – the division works from the basis of an intrinsic belief that personal mobility will, in its words, “extend from the ground to the skies”. The key, says XPeng, will be applying core technologies in electric vehicles and aviation with the intended result of building “safe, intelligent and accessible” low-altitude products. The XPeng team boasts expertise across multiple domains of flying car development including overall design, industrial design, aerodynamics, payload, structure, strength, integrated avionics, intelligent driving and powertrain systems.
In the manufacturing, four core areas are vertically integrated. First, a powertrain system, based on three-iterated generation of technology that is claimed to offer “industry-leading” performance metrics. A composite materials unit – which is set to move into a 10,000-unit-capacity smart factory by the end of 2026 – deals with the industrialisation of 151 key components.
Flight control and navigation teams have worked on developing in-house a full-stack flight control and navigation system that is said to be triple-redundant. It also features intelligent flight envelope protection, low-battery automatic landing, and AI-based flight time estimation technologies.
Finally, the aerial autopilot system team boasts over 100 specialists in intelligent piloting. This is said to have resulted in the world’s first 3D low-altitude flying system based on “one-click” take-off return and smart landing.
Progress assessment
Assessing progress on the strategy (maybe dream to use a better word for the latter two vehicles), the Traveller X2 modular flying car with associate land aircraft carrier is very much a real prospect.
The craft made its first manned flight as far back as June 2021, and four months later, XPeng closed Series A financing of over $500m – the largest single round in the sector in Asia. In October 2022, in what was a massive boost to commercial viability of both the manufacturer and individual crafts, the Traveller X2 completed its first overseas public flight in the key market of Dubai, while an eVtol flying car prototype achieved its first successful flight.
January 2023 saw the Traveller X2 granted a special flight permit in China, and in Autumn that year, XPeng successfully tested the craft’s self-developed multi-parachute rescue system and launched the Land Aircraft Carrier modular flying car package. The application for the Land Aircraft Carrier Type Certificate was accepted by Chinese regulators in March 2024.
By the summer of 2024, XPeng had signed an investment and cooperation agreement with the Guangzhou Development Zone leading to breaking ground on a flying car manufacturing base in Huangpu in October of that year. A month later, XPeng conducted the global-public maiden flight of the Land Aircraft Carrier.
These activities resulted in the company raising $250m in Series B financing in July 2025, and the signing in September that year of a strategic cooperation agreement with the Ras Al Khaimah Transport Authority in the United Arab Emirates with the Land Aircraft Carrier gaining a special flight permit in the country. October 2025 saw the official launch of the Aridge brand and the successful trial production of the Land Aircraft Carrier air module at the company’s mass-production facility.
Somewhat understatedly, XPeng says it is looking at a “promising future” for its Aridge, Iron and auto vehicles. While concept robotics and aviation could be seen as very niche, the complete opposite applies for its automotive vehicles. Evidence for this comes right next to Guangzhou. China’s new energy vehicle penetration – a definition encompassing battery-electric vehicles, plug-in hybrids and fuel-cell hydrogen cars – passed the 50% tipping point two years ago. All eyes now will be on the Paris Motor Show in October for the products that could shape the core future.
