China's new-energy vehicles — battery-electric and plug-in hybrid combined — passed 60% of new car sales this year, for the first time outselling gas engines outright.
Most Western coverage of that shift is about batteries and price. Brian Gu, vice chairman and co-president of XPeng, says the real story is what the car knows how to do once it's charged.
"XPeng first of all is a technology company."
Gu joined XPeng nine years ago after a career as a technology banker, has helped run its expansion into more than 70 countries, and now runs a business that builds flying cars and humanoid robots on the same core stack as its EVs.
I listened to the full interview so you can skip it. 49 minutes of audio, 21 minutes of reading.
Here are the 10 takeaways that matter.
👤 Guest: Brian Gu, vice chairman and co-president of XPeng, and a former technology banker who joined the company nine years ago
🎙️ Host: Nicolai Tangen, Chief Executive Officer of Norges Bank Investment Management, which runs Norway's sovereign wealth fund
📰 Published: 9 September 2026 on YouTube (In Good Company with Nicolai Tangen)
🔴 YouTube | 🟢 Spotify | 🟣 Apple Podcasts | ⏱️ 49 min | ✅ Time saved: 35 min
Key Takeaways
China's new-energy vehicles passed 60% of new car sales this year, overtaking gas engines
Gu called it "a staggering number," and said subsidies are being withdrawn because the mainstream no longer needs them
XPeng designs its own battery packs and platform but does not make battery cells, and does not use CATL
The overlap between its EV supply chain and its robot supply chain is "more than 50%"
A car that can pull over or stop on a spoken command is what XPeng means by intelligence, not the battery
Its VLA 2.0 driving software takes commands through the cabin's voice interface
Vehicle development cycles have compressed to about two years, from three to five
Gu credits China's software-and-internet talent pool, ferocious local competition and fast consumer adoption
XPeng's flying car already has more than 7,000 pre-orders, pending Chinese approval
It is a car that carries a separate rotor-powered pod, which detaches, flies for 30 minutes with two people aboard, and reloads itself
China's humanoid-robot makers split into three camps — bodies, brains, and hands — and XPeng is trying to be the one that combines all three
Its own Turing chip, at 700 TOPS, has replaced Nvidia chips in its vehicles
The best training data for a robot comes from deploying it at scale, not from simulation
Gu thinks cities two decades out could have more humanoid robots walking around than humans
His biggest professional regret is leaving a chemistry PhD lab whose co-founder later won a Nobel Prize
Gu co-founded that lab with David Baker at the University of Washington; Baker won the 2024 Nobel Prize in Chemistry for the protein-folding work they set out to do together
1. Physical AI, Not A Carmaker
Tangen opened by asking Gu to define XPeng, which builds electric vehicles but has pushed into flying cars and humanoid robots on what Gu says is the same underlying technology.
Gu's own description skips the car entirely. "XPeng first of all is a technology company." He said the firm considers itself at the forefront of the "physical AI" movement — using the same core systems to drive a car, fly a vehicle and move a robot
The company was founded on a bet against the industry consensus. Nine years ago, when Gu joined, the rest of the field was pitching "electrified energy transition for automobiles" — but he said XPeng told investors from the start that electrification alone would not be the long-term differentiator
His claim is unqualified. "The long-term differentiation is not with electrification only it's actually in the intelligence," and that belief is why XPeng built what he called a full-stack approach across hardware, software and now AI
2. What Makes A Car Feel Smart
Tangen asked how a professional judges whether a car is actually intelligent, rather than just electric.
The first test is personalization. Gu said an intelligent car has to understand the driver's behavior, preferences and destinations well enough that the driver feels at ease
The second is handling. Even short of full autonomy, he said a better-trained model handles tricky road situations "smoothly" and "without causing any concerns or anxiety"
XPeng's newest software folds voice command into driving. Its VLA 2.0 release, which Gu said he test-drove that same morning, lets a driver speak an instruction into the cabin and have the car act on it directly. "I want to pull over a little bit. The car knows what to do."
He gave a second example — asking for a quick stop — and said the car plans and executes it without further input
The goal is to merge two systems that are currently separate. XPeng is working to unify its autonomous-driving intelligence with its smart-cabin intelligence into one experience
3. What Level Four Changes
Tangen asked what changes over the next ten years, and pressed Gu to define full autonomy along the way.
Level four means no human input at all. "Level four is fully autonomous without human intervention. For example, without the steering wheel," Gu said, correcting the loose way the term gets used
Gu thinks it happens in his lifetime and becomes commonplace, not niche. Once level four capability is available in "ubiquitous scenarios," he said current mobility habits get "completely transformed" — commuting stops constraining where people live, and driving time becomes time for something else
He named the industries that absorb the change. Real estate and communications both shift, in his account, once people are no longer planning their day around who is driving
Time itself gets reallocated, in his telling. With autonomy handling routine driving, Gu said people could spend more of their time on judgment calls and questions "on a more philosophical sense" rather than on tasks — he estimated 90% of time today goes to routine items
4. Range, Charging, Batteries
The conversation moved to the practical limits of battery range, how fast charging is changing, and who actually supplies the cells.
XPeng is not chasing bigger batteries. Gu said a larger pack adds to the "cost and weight of the vehicle," and that with the current battery chemistry — which he expects to remain dominant for at least five years, until solid-state cells become commercially viable — range has already reached the point of diminishing returns. Asked directly where that ceiling sits, he was blunt: "Anything more than that is waste."
What has changed is not the battery, but the stop. "Now it's maybe 5 to 10 minutes," Gu said of a fast charge that used to take half an hour to an hour — a shift he compared to the downtime of a gas-station stop
He credits the combination — not a bigger battery — with solving range anxiety. The improvement in charging network availability and speed, he said, "addresses a lot of people's range anxieties"
XPeng does not make its own battery cells, and does not use CATL. "We don't make our battery cells. So, we actually source from very high quality suppliers," Gu said, naming no supplier but confirming it isn't China's largest battery maker: "Not CATL."
Everything wrapped around the cell is XPeng's own design. The battery management system, the pack engineering and the electric platform that determines charging speed and efficiency are all built in-house, he said
5. China: New Energy Beats Gas
Gu gave the clearest market-level claim of the interview when asked where the Chinese EV market is heading.
China's electrified vehicles now outsell the internal-combustion engine. "This past month Chinese new energy penetration exceeded 60% maybe even close to 65%. That's a staggering number," Gu said, adding that this is the first time battery-electric vehicles alone have crossed the ICE segment, and that new-energy vehicles including hybrids are now more than 60% of new car sales
That threshold changed the policy conversation. Subsidies are being withdrawn, he said, because "you don't need to subsidize the mainstream product" — a reversal from a few years ago when EVs were seen as a niche purchase
The physical infrastructure is turning over with it. Gu said gas stations are starting to disappear, replaced by charging stations, as EVs and hybrids — which carry both a battery and onboard computing — pull more of daily behavior toward "intelligence" rather than fuel
6. China: "The Fittest Gym"
Asked how competitive the Chinese EV market has become, Gu reached for a gym metaphor of his own.
His description of the market is unambiguous. "China is probably the fittest gym in the world for automotive industry" — whoever survives it, he said, stands a real chance of becoming a global leader
Competing now requires a wider set of skills than it used to. Automakers used to be able to lean on tier-one and tier-two suppliers for technology; today, Gu said, a competitive product needs automotive engineering, AI and software, the new electric powertrain, and service quality all at once
His own answer to who wins is a checklist, not a name. The winners, in his account, will be the companies that combine full-stack technology with real scale and global reach — capability, not size alone
7. Norway To 70 Countries
Tangen, noting XPeng's Nordic dealer presence, asked how the global rollout is going.
The international push started small, in Norway. XPeng opened its first overseas store there around the end of 2020, and Gu called the six years since "a very interesting" and "very educating process"
The company now sells in more than 70 countries, spanning Europe, Southeast Asia, the Middle East, Australia and, most recently, Latin America
The obstacle is not the product — it's the brand. Gu said Chinese brands have historically been seen as lacking quality or recognition abroad, so entering a new market means building local brand trust and service capability from scratch, while also differentiating from "a dozen other brands" also trying to go global out of China
Volkswagen's stake in XPeng gives him a direct comparison point. Gu said XPeng is more nimble and faster at technology decisions than large multinationals, while Volkswagen and other global peers know how to sustain product development and brand loyalty across generations — a skill he said XPeng is still learning
He expects more tie-ups like the Volkswagen one. Chinese firms bring speed and local-market understanding; established multinationals bring scale and the ability to monetize that speed across wider markets — a complementarity he thinks will drive more collaboration
8. Why China Moves Faster
Tangen pushed Gu to explain "the Chinese speed" of product development directly.
Gu attributes it to three forces. China's strength in software, internet and consumer electronics feeds directly into car development; intense local competition forces faster translation of technology into features; and Chinese consumers adopt new technology faster than other markets, as they did with mobile internet and mobile payments
Development cycles have compressed sharply. Vehicle development has shortened to around two years, down from three to five, and over-the-air updates now add features on a faster cycle than the traditional automotive product cycle allows
He puts the real difference down to company DNA. "The core DNA is tech," Gu said of XPeng's founders, who came from the automotive world, the software and internet world, and increasingly AI
Being young and small is itself an advantage. At twelve years old, he said, XPeng has no department walls or decades-old processes to work around, and no layers of stakeholders to get a decision through — unlike global rivals, where a headquarters decision can take a year to reach the China market
9. Flying Cars: 7,000 Orders
Asked when cars will fly, Gu pointed to AeroHT, the company XPeng incubated and now holds a minority stake in.
The product is not the ordinary idea of a flying car. AeroHT calls it a "land aircraft carrier" — a car that carries a separate electric flying pod inside it. The pod detaches automatically, flies two people for about 30 minutes, and reloads itself back into the car afterward
It is close to a commercial product, pending regulatory sign-off. Gu said AeroHT has collected more than 7,000 indicative orders worldwide, not only in China, and is waiting on final approval from China's aviation regulator
He expects adoption to run three to five times slower than for cars or self-driving technology, given the regulatory hurdles and the education required before individuals or companies are comfortable buying and operating one
The upside, in his framing, is a market that barely exists yet. Gu said the "low altitude economy" — a term he called a Chinese buzzword of the last few years — could unlock a large chunk of otherwise untapped economic activity, in China and globally
10. Why XPeng Builds Robots
Tangen turned to XPeng's robotics push, backed by Alibaba and Tencent, and asked Gu why a car company is building robots at all.
Gu frames it as the same technology, applied elsewhere. The edge computing, in-house chips, driving models and data infrastructure XPeng built for autonomous driving are, in his telling, "naturally very transferable" to robotics, sharing the same physical-AI building blocks
China's robotics field splits into three types of company, in his read of the market. One group focuses on the body — he named Unitree and EngineAI as companies with striking motion capability, able to "run," "jump," "kick" and "perform and dance," but without matching intelligence. A second group focuses on the brain, training robots for specific tasks without investing in movement. A third builds components like dexterous hands
XPeng is trying to be the company that puts all three together into one deployable, mass-produced humanoid robot — work he said the company has now completed across the separate pieces
A humanoid shape is deliberate, not a design default. Gu's reasoning is that the built environment — offices, stores, restaurants — is built for human movement, so a human-shaped robot is the most broadly deployable form, and deployment itself is what generates the data to keep improving it
The best data, in his view, comes from real deployment, not simulation. He said the most accurate and useful training data comes from putting a general-purpose robot body into large-scale use — not from internet footage, simulation, or wearable camera collection alone
There is no off-the-shelf robot supply chain to buy from. "The supply chain for robots is still being formed," Gu said, and building an advanced robot currently means working directly with suppliers rather than picking parts off a shelf
The overlap with its EV supply chain is substantial. Gu put it at "more than 50%," though he said most shared parts are not off-the-shelf and require close joint design work with suppliers
Its own chip already outperforms what it replaced. XPeng's Turing chip, used in its vehicles, runs at 700 TOPS of compute — a figure Gu said rivals the latest Western chips and has already replaced the Nvidia chips XPeng previously used for driving
He does not see chip access as a current constraint. Gu said XPeng has "full access" to the chips it needs, beyond the memory-chip cost and supply pressure every manufacturer is dealing with, and expects more local sourcing over time
He drew a line between where China has caught up and where it hasn't. On board-level mobility chips, he said Chinese players are already at parity; a generational gap remains in GPUs and broader cloud computing
Bonus Insights
Chinese and American AI models are not really racing each other, in his account. Gu said the two ecosystems are solving different problems for different customers — US frontier labs building high-value, secured tools for governments and large companies, Chinese labs building low-cost, open, easy-to-use models for small and mid-sized businesses and emerging markets — and expects the two ecosystems to keep diverging rather than converge on one leader
His picture of a city in twenty years is dispersed and three-dimensional. Without a need to commute to a center, he expects people to live more spread out; deliveries arriving by drone or autonomous vehicle to shrink last-mile logistics; and low-altitude travel corridors layered above ground routes. He raised the possibility that humanoid robots could eventually outnumber people walking the streets
Tangen's own aside was that widely deployed household robots might lower divorce rates, by taking over chores like the dishwasher and the trash — a joke Gu did not directly engage with before returning to the robotics roadmap
Gu named who he leans on for perspective. He described Joe Tsai of Alibaba as a close friend and regular conversation partner, alongside people at Tencent, and singled out Nvidia's Jensen Huang as someone XPeng learned from during the period it relied more heavily on Nvidia's chips
A day in his life, as he described it that morning: up around 7:30, a first meeting at 8:30, two hours road-testing the new G9L and its VLA driving software around the city, then time mapping out an R&D plan for Hong Kong before ending the day with dinner and a partner meeting
His advice to young people is built entirely around his own regret. "Never afraid of change. Embrace change," he said, arguing that new professions, regions, languages, cultures and industries should feel exciting rather than intimidating
His biggest professional regret, Gu said, is a scientific career he didn't finish: a chemistry PhD he began at the University of Washington, where he co-founded a lab with a fellow first-arrival, David Baker — the professor who went on to win the 2024 Nobel Prize in Chemistry for cracking the protein-folding problem the two of them had set out to solve together.
Products, Companies & Tools Mentioned
XPeng (The company Gu co-runs — EVs, flying cars and humanoid robots built on what he calls one shared physical-AI stack)
Volkswagen (Holds a stake in XPeng and offers Gu his direct comparison point between a Chinese challenger and a legacy global automaker)
AeroHT (The XPeng-incubated flying-car company behind the "land aircraft carrier," with more than 7,000 indicative pre-orders)
Unitree and EngineAI (Named as the two Chinese humanoid-robot makers Gu says lead on body and motion capability, without matching intelligence)
Alibaba and Tencent (Investors in XPeng's robotics push; Gu also named Alibaba's Joe Tsai and Tencent contacts as people he regularly talks to)
Nvidia (Its chips were XPeng's prior go-to for onboard driving compute, since replaced by XPeng's own Turing chip)
Tesla and Figure (Gu's examples of international humanoid-robot players who have drawn attention and capital over the last two to three years)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: