Part IThe Story
The Man Inside the Suit
The robot had a zipper. On the morning after XPeng's AI Day in Guangzhou, a night He Xiaopeng later said he did not sleep through, that zipper was the thing its maker most wanted the world to see.
The trouble had started on the evening of November 5, 2025, when a humanoid called IRON walked onto the stage. It was 178 centimeters of synthetic skin over an engineered skeleton, with a curved display where a face would be. Its gait was modeled on a catwalk, and it moved with a sway of the hips so fluid that the Chinese internet decided, within hours, that it was a person in a costume. The suspicion was itself a kind of review. One comment, which He would read aloud on camera the next morning with a visible smile, put it plainly: "The robot walks more smoothly than Tesla's. It's impossible. It must be a person."
So he recorded a video. He walked behind the machine, unzipped its back, showed the radiator where a spine would be, and let it walk again. Later, during a live broadcast, the company took the demonstration further. Engineers peeled IRON open layer by layer, skin, then muscle, then skeleton, and cut into the leg. It kept walking. A Chinese automotive outlet called it "the first Turing Test moment in the history of embodied intelligent robots," which was generous. It was also a strange sort of triumph: the test was passed by failing to be human, and the proof required an act that looked, from a certain angle, like vivisection.
The robotics team did not want to do it. "It was very difficult for me to persuade our robot team," He told the interviewers who caught him afterward, still running on no sleep. "They originally disagreed with this, because they think IRON is their child." He overruled them. His reasoning was arithmetic: the company did not need only "1% of industry users" to believe, but "99% of non-industry users." When someone asked whether the robot itself might one day understand what had been done to it, he gave the only honest answer available. "To put it bluntly, I don't know."
— He Xiaopeng, interview with 36Kr's Smart Car Reference, November 2025If the robot knew what we were doing to it, it wouldn't have let us cut it open today.
It is tempting to treat the zipper as a stunt and move on. But the question the internet asked about IRON (is there a man inside?) is the question that has followed He Xiaopeng through two decades, three companies, and one industry that was never supposed to be his. It was asked when he lent his given name to a car startup he did not yet run. It was asked when he was called, with varying degrees of contempt, the Chinese Elon Musk. It was asked in 2022, when the company bearing his name nearly collapsed under a flagship SUV that buyers did not want. It was asked in 2025, when he tore out a self-driving system that had cost billions of yuan, on the theory that the thing had been assembled, not grown.
A note on the name, since it causes a problem in English. His surname is He (何), pronounced roughly huh. In English sentences it collides with the pronoun, and every profile written in English becomes a small comedy of ambiguity. The given name, Xiaopeng (小鹏), means something like "little roc," the giant mythical bird of Chinese literature that flies ninety thousand li in a single ascent. That is the name on the cars. It is on the robots, and on the flying vehicles. A man who wanted, at eighteen, simply to stop worrying about money ended up with a given name that is now a ticker symbol on two exchanges, XPEV in New York and 9868 in Hong Kong.
The robot was cut open, and it walked. Three months later, in a shopping mall in Shenzhen, it fell down. Both things are true, and the distance between them is roughly the distance this story covers.
By the Numbers
The XPeng Ledger
$4.3BAlibaba's 2014 buyout of UCWeb, then the largest deal in Chinese internet history
482Vehicles XPeng sold in 2018, its first year of production
429,445Vehicles delivered in 2025, up 126% year over year
RMB 0.38BNet profit in Q4 2025, XPeng's first profitable quarter
RMB 1.72BTechnology licensing fees from Volkswagen, first half of 2025
~RMB 15B2026 R&D budget across XPeng and AeroHT, 13–14% of revenue
384,934,965XPeng shares held directly by He Xiaopeng, per a March 2026 filing
Four Roads to Freedom
At eighteen, by his own account, He Xiaopeng set himself a single goal: financial freedom before forty. He worked out that there were four ways to get there. "After college, I thought there were several ways to achieve financial freedom," he told the entrepreneur-turned-broadcaster Luo Yonghao in 2025. "The first was to win the lottery, the second was to buy real estate, the third was to get stock options in a good company, and the fourth was to start a business."
The taxonomy is so blunt it is almost funny, the reasoning of a computer science student who treated his own life as a problem with a finite solution space. It is also revealing. Nothing in the list involves cars, robots, flying machines, national industrial strategy, or the Chinese Communist Party's vision of a "new form of the intelligent economy." All of those would arrive. At eighteen, the target was simply not having to think about money anymore.
He was born on November 3, 1977, in Huangshi, an industrial city on the south bank of the Yangtze in eastern Hubei. He entered Huangshi No. 2 Middle School in 1992 and in 1995 was admitted to the computer and applications program at South China University of Technology in Guangzhou. That move south mattered more than it looked. Every company he has started since has started in Guangdong, a fact he now recites in provincial government video series with the fluency of a man who has given the talk before.
He graduated in July 1999 and took a job at AsiaInfo, the company then laying much of the plumbing of China's early internet. He cycled through roles as technology manager, testing manager, and project manager. When AsiaInfo went public he received roughly $30,000 in shares. That was real money for a recent graduate from Huangshi, and nowhere near the third road to freedom. By one account he tried to jump to HP and was turned down.
At AsiaInfo he worked alongside Liang Jie, a fellow South China University of Technology graduate whom colleagues remembered as a gifted code writer. The two built a large-capacity email system together. Liang was the engineer's engineer, the man who would later serve as chief technology officer of UCWeb inside Alibaba. He Xiaopeng was something else: the product man, the one who sat between the code and the customer and argued about what should exist.
In 2004 they left to try the fourth road.
The money came from Ding Lei, the founder of NetEase, who had already made his own fortune from the first wave of Chinese internet portals and who liked the idea of mobile email. "Ding Lei took a fancy to the mobile email business and lent us 800,000 RMB," He recalled. "He originally wanted to convert it into an investment but didn't. We paid him back after a few years. 800,000 RMB was not a small amount at that time. With that money, the company could survive for two years."
Hold onto that detail. A future billionaire's first company ran on a loan rather than equity, from a peer rather than a fund, and the lender passed up the chance to own a piece of what became the largest acquisition in the history of the Chinese internet. In the Chinese tech world of the mid-2000s, the line between friend, creditor, and investor was porous. He spent those years building a network that included Ding Lei, Li Xueling of the live-streaming company YY, and Lei Jun, who would found Xiaomi in 2010. They backed one another's ventures. A decade later, nearly every name in that network would reappear on XPeng's capitalization table.
The company was UCWeb. Its product, UC Browser, was a bet that hundreds of millions of people would get onto the internet through cheap phones on bad networks. The engineering answer was compression. UCWeb processed pages heavily on its own servers and later built its own rendering kernel, the U3, so that the web would load on hardware that had no business displaying it. It was not glamorous work. It was the work of making something heavy feel light on hardware that was underpowered, and the company did it better than almost anyone. Yu Yongfu joined and took the chief executive role in 2006. He Xiaopeng stayed president of product from January 2005 until June 2014, nine and a half years. UC Browser became, by the company's later accounting, the third most used mobile browser in the world, a default gateway across China, India, and Indonesia.
In June 2014, Alibaba, which already held a stake, bought the rest for a reported $4.3 billion, then the largest merger in Chinese internet history. He Xiaopeng was thirty-six. The goal he had set at eighteen was met with three years to spare.
He became president of UCWeb under Alibaba, then president of the Alibaba Mobile Business Group, then chairman of Alibaba Games, then president of the video platform Tudou. The titles accumulated the way titles do inside a conglomerate after an acquisition: generously, and with a faint air of consolation. It was a soft landing. He was wealthy, senior, and comfortable.
He has described what came next, in more than one interview, as a kind of emptiness. He had solved the problem he set for himself, and the solution had not been the point.
The Orange That Was Already Taken
In 2014, the same year he sold UCWeb, He Xiaopeng took delivery of a Tesla. He was, by one widely repeated account, among the first private buyers in China to receive one. That June, Elon Musk had announced that Tesla would open its patents to anyone who wanted to build electric cars in good faith. The two facts fit together neatly in the founding lore, perhaps too neatly. Inside Alibaba, He proposed that the company build a smart electric car of its own. The proposal was turned down.
He was not, at first, inclined to do it himself. "Internet people were overly confident at that time," he told Luo Yonghao, and they "needed a process of re-growth." Hardware was hard. Large hardware was harder. He had watched peers from the internet business assume that a car was a phone with wheels, and he did not think it was.
So he did what a newly liquid internet executive does with a conviction he is not yet ready to act on: he wrote a check.
The engineers came from somewhere else entirely. Xia Heng was born on November 7, 1983, six years after He Xiaopeng, and studied automotive engineering at Tsinghua University for six years, finishing a master's degree in 2008. Tsinghua is the cradle of Chinese automotive talent, the institution that has produced the engineers of the state carmakers for decades. Xia went to Guangzhou Automobile Group, the state-owned GAC, where he supervised the New Energy Vehicle Control Section. He worked as a resident engineer in several European countries and published more than ten papers and patents on electric and intelligent vehicles. His GAC colleague He Tao (no relation) led the group's work on intelligent and driverless vehicles. The two men were, in effect, the next generation of the Chinese state auto industry, trained in its best school, raised in its best labs, and impatient with its pace.
On August 15, 2014, they registered a company in Guangzhou. Its first name was Guangzhou Orange Mobility Technology. The founders liked "Orange Car," which suggested freshness, vitality, something young. Chinese internet companies of that era tended to name themselves after fruits and animals, a deliberate rejection of what He called "the lofty names you used to have." But the orange trademark was taken. So were most of the fruits and most of the animals. "The company was very small at the time, and could not afford a brand name that someone else had already registered," He told CNA in 2024. "The co-founders couldn't register a good brand name and asked if (my name) Xiaopeng could be one of the registration options for the brand. Then they registered it."
The early backers read like a roster of He Xiaopeng's address book: Li Xueling of YY, Fu Sheng of Cheetah Mobile, Wu Xiaoguang, formerly of Tencent's e-commerce arm, Zhang Ying, and Lei Jun of Xiaomi. The investor whose personal reputation anchored the round was the man whose given name went on the door. He had not yet left Alibaba. He was not the chief executive. He was a backer and a name, and as later events would show, the name proved the heavier commitment.
— He Xiaopeng, on Luo Yonghao's Crossroads, August 2025In China, it seems strange to use the boss's name as a company name. For seven or eight years, many friends advised me to rename XPeng Motors, saying that the sales would double as long as we changed the name.
He never changed it. His explanation for holding on has shifted over the years, from cultural confidence to brand recognition to stubbornness. The most revealing version came in 2024: "If in 10, 20 or 30 years, a Chinese (car) brand uses its own Chinese name to go global, and people like it, I think it would be very cool." The ambition in that sentence is not modest. A man's given name, chosen because an orange was unavailable, was being asked to carry a national argument.
Xia Heng, meanwhile, built the company. He recruited from GAC, BMW, Lamborghini, BYD, Alibaba, Tencent, Huawei, and Samsung, a mix of factory and internet talent that would define XPeng's character for a decade. From 2015 the team built and tested four prototype cars. A Chinese tech outlet reported in September of that year that the "first exposed internet car team" had already logged 5,000 kilometers of road testing. In September 2016, XPeng showed a beta prototype of a small electric SUV in Beijing. In 2017 came the Identy X, a limited run of fifteen pre-production cars reserved for founders and investors. It used Samsung 18650 battery cells with an energy density of 152 watt-hours per kilogram, good for 300 kilometers of range. These were not cars so much as letters of intent on wheels.
Years later, in April 2019, Xia Heng would stand at the founding ceremony of Tsinghua's new School of Vehicle and Mobility, beside an XPeng G3, as the alumnus who had made good. "The automobile department of Tsinghua University, as the cradle of Chinese automobile talents, has witnessed the development of China's automobile industry," he said. "Today, the establishment of vehicle college indicates that the automobile industry will start again in the future under the environment of intelligent travel." Start again. It was a quiet heresy to say at Tsinghua, and an accurate description of what he had left GAC to do.
By then, the company's public story had rearranged itself around the man whose name it carried. That happens to co-founders, and it happened to Xia Heng.
Seven-Yuan Chicken Rice
On August 22, 2017, He Xiaopeng resigned from every position he held at Alibaba. Seven days later, on August 29, he joined XPeng full time as chairman. He was thirty-nine. By one account he told Alibaba's leadership he was "all in for ten years," and his wife opposed the move. The ten-year clock he started that August runs out in August 2027.
The company he walked into had about five hundred employees and an office in one of Guangzhou's chengzhongcun, the "urban villages" left behind when the city grew around old farming settlements: dense, cheap, a little lawless, full of migrant workers and cheap restaurants. Here the founding lore turns, as founding lore does, to food. A Chinese industry commentator, contrasting He with the professional managers who flew business class to their failing auto startups, noted that he could be found eating seven-yuan chicken-leg rice in the urban village while already a billionaire and a former senior Alibaba executive. The detail is offered as a moral, and it flatters him a little too much. It is still useful. A man who had optimized web pages to load on bad phones over bad networks was now trying to optimize a company to survive on capital he would increasingly supply himself.
The pace was frantic. Headcount went from roughly five hundred to more than 2,600 by October 2018. In January 2018, XPeng took its first production-intent model, the G3, to the Consumer Electronics Show in Las Vegas, the first Chinese electric vehicle shown there. The name had come from a crowdsourced naming contest. The "G" stood for Geek, an internet company's idea of a car name. At CES, He told a Chinese journalist something that would be quoted against him for years: "Elon Musk is now better than me, but I might be better in the future."
That August a funding round of 4 billion yuan valued the company at more than 25 billion yuan. In March 2018 he had recruited Brian Gu Hongdi, chairman of J.P. Morgan's Asia-Pacific investment bank, who would become vice chairman and president. Gu had spent his career on the other side of the table from companies like XPeng, structuring the deals that kept them alive. XPeng later described its leadership as a fusion of three cultures: the internet, car manufacturing, and finance. In practice those were He, Xia, and Gu.
Production of the G3 began in November 2018. XPeng did not yet hold a Chinese manufacturing license, so the cars were built under contract by Haima Automobile in Zhengzhou, a thousand kilometers north of Guangzhou. XPeng sold 482 of them in 2018.
The first years of actual business were humbling in the specific ways that hardware humbles software people. In July 2019, XPeng was accused of clearing out old G3 inventory days before announcing a longer-range version, and early buyers felt cheated. Each G3 that caught fire made the news. One ignited at a charging pile in Guangzhou on April 14, 2021, the same day XPeng unveiled the P5, billed as the world's first mass-produced car with lidar. A rival had sued one of XPeng's engineers: in March 2019, Tesla accused Cao Guangzhi, a former Autopilot engineer, of taking source code to XPeng. Cao admitted uploading Tesla code to his iCloud before leaving, though he denied theft. XPeng was never a party to the suit. It handed its own source code to a court-appointed neutral party, which found that XPeng's code did not use Tesla's intellectual property, and the case settled. When Musk took to Twitter, XPeng called his posts "bullying." An earlier case, in which a former Apple engineer named Xiaolang Zhang was charged in July 2018 with stealing autonomous-driving trade secrets while seeking a job at XPeng, ended with his guilty plea in August 2022. XPeng said the alleged theft predated his employment.
At the Fortune Global Tech Forum in Guangzhou in November 2019, the magazine's Adam Lashinsky described XPeng's "snazz factor" as "high while its unit volume is low." Asked what other ideas he was thinking about, He named two: freezing human brains, and building cities on the water. "He'd fit right in Northern California," Lashinsky wrote. The line was a joke. It was also accurate about the impulse, if not the geography.
That same month XPeng closed a $400 million round with Xiaomi as a strategic investor. Then the pandemic arrived. He has called this the worst stretch in the company's life.
— He Xiaopeng, post-launch interview with Chinese media, April 2023The toughest time for us was in 2019 and early 2020 during the pandemic. I personally contributed half of the $400 million we raised. If XPeng didn't make it, I'd pay everyone's salary and suppliers first.
Roughly $200 million of the $400 million round, in other words, came from the chairman himself. The name on the door was also the largest single source of cash in the building.
Then the window opened. The P7, a long, low four-door sedan designed by Rafik Ferrag and aimed squarely at the Tesla Model 3, began deliveries on June 29, 2020. In July, XPeng raised $500 million from Aspex, Coatue, Hillhouse, and Sequoia China. In August it raised $400 million more from investors including Alibaba, the Qatar Investment Authority, and Abu Dhabi's Mubadala. On August 27, 2020, it raised $1.5 billion in a New York IPO, and the stock rose 41 percent on its first day. He's stake was worth $4.2 billion on paper. That year he was named EY Entrepreneur of the Year. On July 7, 2021, XPeng listed in Hong Kong, the first U.S.-listed Chinese company to do so as a dual primary listing, a hedge against the possibility of being expelled from American markets. By May 2021 he owned 23 percent of the company; Alibaba, the employer that had declined to build a car, owned 12.
In November 2021 Forbes put his fortune at 48 billion yuan, 76th in mainland China. Neither of his closest rivals, William Li of Nio and Li Xiang of Li Auto, made the list. A Chinese EV news site noted, with the precision of the era, that Elon Musk was forty-five times richer.
XPeng delivered 98,155 cars in 2021. The P7 alone sold 60,569 in China that year and won the Xuanyuan Car of the Year award. In September 2022, it was added to Forza Horizon 5, which for a certain kind of young Chinese buyer was its own kind of award.
Then came the G9.
The Chairman Who Was Too Nice
The G9 was meant to be XPeng's flagship, a large SUV loaded with every piece of technology the company could fit into it, and it launched in September 2022 into a market that had already stopped caring about long specification sheets. Its configurations confused buyers. The pricing confused them further. Sales disappointed. XPeng's 2022 deliveries rose only to 120,757, while the industry around it was being transformed by BYD's scale, Tesla's price cuts, and a domestic price war that turned new Chinese carmakers into casualties.
In the telling of one founder-profile firm, the G9 was the moment a "consensus-seeking chairman" was forced to become a "hands-on CEO." In this version, Lei Jun told his old friend that the company's biggest problem was He himself. That account is secondhand. The facts around it are not. XPeng replaced more than a dozen senior executives in a single year. On Luo Yonghao's program in 2025, He called this the "darkest moment" of his entrepreneurial life. The voices urging him to rename the company peaked during this stretch, as if the trouble might be in the letters.
In March 2023, He Xiaopeng became legal representative of Guangzhou XPeng Motors Technology, the operating entity. The same month, filings showed Xia Heng stepping down as executive director and manager of the company. The co-founder who had registered the business, recruited the first engineers, and built the first four prototypes receded from the formal structure that had carried his name on the paperwork. The man whose given name was on the paperwork's letterhead moved in.
By April 2023, He was speaking with the conviction of a convert. "Now I'm overseeing all six R&D teams," he told reporters after unveiling the G6 at Auto Shanghai, "so there's no doubt I'm involved in the products." He talked about scale with a new bluntness: "I've realized the importance of thinking big and scaling up. Annual revenue of one to two hundred billion RMB might be okay in other industries, but it's not enough for the automotive industry." He had once said a carmaker needed three million vehicles a year to survive. Now he revised the figure upward: "for us it should be closer to 5 million." He also said, almost in passing: "We've always focused on R&D, even when we were losing money." It was less a boast than an alibi.
The G6, unveiled on April 18, 2023, was the first product of the new regime. It sat on a new platform, SEPA 2.0, built to cut costs and speed up development across every future model. It added up to 300 kilometers of range in ten minutes of charging. Its design had been developed in collaboration with the science fiction writer Liu Cixin, whose Three-Body novels had become the unofficial literature of Chinese technological ambition. This is the kind of detail that makes Western readers smile and Chinese buyers lean in. It sold 44,545 units in China in its first half-year.
Then, on July 26, 2023, Volkswagen announced it would pay $700 million for 4.99 percent of XPeng.
The Student Sells to the Teacher
To understand why the Volkswagen deal mattered, it helps to remember who Volkswagen had been in China. For four decades it was the default, the first foreign carmaker to commit to the market, the company whose Santana sedans were the taxis of Shanghai and whose joint ventures with state carmakers trained a generation of Chinese engineers. In a sense, GAC, Tsinghua, and the whole apparatus that produced Xia Heng had been built in Volkswagen's shadow. Volkswagen was the teacher.
In 2023 the teacher came to buy lessons.
The agreements accumulated with the patience of a slow-moving takeover in the other direction. In February 2024, the two companies signed a joint development agreement covering platforms and software, promising to cut Volkswagen's product development cycle by more than 30 percent. In April 2024, they agreed to develop a new electronic and electrical architecture, the digital nervous system of a car, based on XPeng's own, for Volkswagen's China platform, with mass production in 2026. In July 2024, they opened joint project houses in Guangzhou and Hefei. In January 2025, they agreed to open their charging networks to each other. In August 2025, XPeng's half-year report revealed 1.72 billion yuan in technology licensing fees from Volkswagen, which Chinese financial press reported at a gross margin near 60 percent. The same month, the architecture deal was expanded beyond electric vehicles to Volkswagen's combustion-engine and plug-in hybrid platforms in China, the old business as well as the new. In November 2025, XPeng opened its new narrow-road driver-assist system to other carmakers, with Volkswagen as the first client. In February 2026, He revealed that Volkswagen would be the launch customer for VLA 2.0, XPeng's second-generation driving model.
The licensing revenue line, more than any delivery figure, marks what changed. The company that had once outsourced its first car to Haima, whose very name had come about because a fruit was unavailable, was now selling its software stack to the company that had taught China to build cars.
In August 2023, XPeng also agreed to acquire the autonomous-driving unit of Didi, the ride-hailing giant, for about $744 million in shares. Didi's car project became the foundation of MONA, a cheaper brand aimed at buyers XPeng had never reached. The MONA M03 launched on August 27, 2024, the tenth anniversary of XPeng's founding.
The anniversary event was oddly mixed. Video congratulations came from Changan's Zhu Huarong, Alibaba's Cai Chongxin, GAC's Zeng Qinghong, and Huawei's Yu Chengdong. Lei Jun attended in person. Nearly a thousand journalists registered. On stage, He reminisced about the hard years. In the audience, owners who call themselves "Peng Friends" wept. One commentator compared the spectacle to the doomed LeEco Supercar launch years earlier, which was not entirely a compliment. The numbers underneath were sobering: XPeng had delivered 52,028 cars in the first half of 2024, just 18.6 percent of a 280,000 target. July brought 11,145. Gross margin had recovered to 13.5 percent from negative 1.4 a year earlier. That was improvement, not triumph.
That month, He bought roughly HK$107 million of additional XPeng stock with his own money.
The turn came in the fourth quarter. The MONA M03 delivered more than 15,000 units in each of two consecutive months. The P7+, a larger sedan, passed 20,000 deliveries in its first two months. In January 2025 XPeng delivered 30,000 cars and regained the top spot among China's new-energy startups. He's annual letter to employees that February caught the shift in tone: "We should cherish the hard-won upward momentum, and the essence of momentum is confidence."
XPeng delivered 429,445 vehicles in 2025, 126 percent more than the year before and 113 percent of its own target. In the fourth quarter it reported its first positive net profit, 380 million yuan. Full-year revenue reached 76.72 billion yuan, with an 18.9 percent gross margin. When the second-generation P7 opened for pre-orders on August 6, 2025, it took 10,000 refundable orders in six minutes and thirty-seven seconds.
Floor and Ceiling
In the spring of 2025, just as the numbers turned, He Xiaopeng decided to throw away the thing that had made them turn.
The thing was XPeng's first-generation driving system, the end-to-end model the company internally called VLA Gen 1, built over years at a cost he put at "a few billion" yuan. It worked. It was the reason XPeng's cars could navigate city streets without high-definition maps, the feature that justified the brand's premium. He decided it had a ceiling.
He explains the problem with a scoring metaphor, and the metaphor is worth reproducing because it reveals how he thinks. The old approach, whatever you call it, stitched together rules written by engineers with AI models that filled in the gaps. It could produce a dependable product with a floor of 900 points and a ceiling of 1,000. No amount of post-training would push it much past 1,000, and 1,000 was not enough for genuine autonomy, Level 4 or Level 5. The new approach rebuilt the entire system around a much larger foundation model. Its theoretical ceiling ran to 100,000 points, perhaps a million. Its floor, at the time, was about 100.
"You want to build a product with a ceiling of 1,000 and a floor of 900," he told the podcaster Zhang Xiaojun in May 2026, "but back then the floor might have been only 100." Most of his team, he said, was in wait-and-see mode. He has called the old method a fengheguai, a stitched-together monster, a Frankenstein patchwork. His test case is the underground parking garage. No self-driving system can actually navigate one. Every car that claims to park itself in a garage is relying on "memory": it drives the route once, remembers it, and repeats. Its understanding of the physical world, he said, is "extremely low."
He chose the high ceiling and the low floor. In one account he put the odds of that technical path succeeding at about one in five.
— He Xiaopeng, on Zhang Xiaojun's podcast, May 2026Most people don't dare to gamble in the physical world. I'm probably bolder than most.
Then he cut. "You can't take a small knife to a big tree," he said. "Once you've thought it through, chop it down." The autonomous-driving organization was rebuilt from the business layer to the roots. More than two hundred people walked out. The company stopped calling itself XPeng Motors and started calling itself XPeng Group. Its self-description shifted from "smart electric vehicle company" to "physical AI company," with three growth curves: cars, robots, and globalization. In early February 2026, the autonomous-driving center and the smart-cockpit center were merged into a single "General Intelligence Center," on the theory that a car's driving decisions and its conversations with passengers should run on one shared model. In He's imagined future, a passenger says take that small road ahead and avoid the traffic lights, and the car simply does it.
This was the second time he had blown up a part of the company that worked. The first was hardware.
On April 14, 2021, XPeng had launched the P5 as the world's first mass-produced car with lidar, the laser rangefinder that Musk famously dismissed as a crutch. XPeng's engineers defended it in vivid terms: lidar would catch "small moving obstacles even like kids and pets," and the motorbikes that were "a nightmare for anybody who's working on driving." The bet was a public rebuke to Tesla, and the press covered it that way. The South China Morning Post's headline said XPeng was "defying Elon Musk."
In late 2024, XPeng launched the P7+ without lidar. Its sensor suite relied on cameras and new 4D millimeter-wave radars. In 2025, models that had carried lidar were refreshed and had it removed. The company that had built its reputation on putting lasers in cars took them out, quietly, once its software no longer needed them.
There was a third reversal, too: silicon. XPeng began hiring a chip team in late 2020 and contracted Marvell for back-end design and integration. In 2022 it cancelled that contract, judging Marvell's experience and investment insufficient, and moved to Socionext. The chip taped out in August 2024. In June 2025 it shipped in the G7 under the name Turing: forty CPU cores, two neural processing units, 750 trillion operations per second, able by XPeng's account to run AI models of up to thirty billion parameters. By January 2026 the refreshed G6 could carry three of them, two for driving and one for a voice model in the cabin. That made 2,250 TOPS in a mid-size SUV starting at 176,800 yuan. XPeng offered Turing to other carmakers, and Volkswagen agreed to put it in some China models.
On a podcast in 2026, He said something that clarifies all three reversals, lidar, the driving model, and the chip vendor. Physical AI, he argued, cannot be judged only on its "long board," the headline benchmark. It needs a floor too: safety, regulation, stability. And it needs breadth: cost, materials, range of conditions. The language is from carpentry and old Chinese proverbs about the shortest stave in a barrel. The temperament is a product manager's, someone who spent ten years making web pages load on phones that could not handle them.
In June 2022, years before any of this, he had posted a video of himself riding in a development version of XPeng's city-driving system. He complained the whole way. When a big truck appeared, he said, the car's "IQ seems to have dropped by a generation." Its driving, he said, was less efficient than his own. It was a teaser for a product launch, and it was also, unmistakably, a man criticizing his own child on camera before anyone else could.
A Ten-Year Plan for Leaving Home
The first XPeng cars to leave China went to Norway in late December 2020, a small shipment of G3 SUVs with a press release attached. "Our launch in Europe comes just as consumers are shifting in increasingly large numbers to more sustainable personal transport," He said, in the language press releases require. By 2021 XPeng had sold 211 G3s in Norway. In a market that took tens of thousands of electric cars a year, that was a rounding error.
He did not seem bothered. "We have taken a very slow path in the past three years," he told CNA in March 2024, "because many people think that you should be able to achieve results quickly as soon as you enter the international market. I never thought so." He kept returning to a decade. "From my perspective, going global is at least a 10-year plan." He also explained, in hindsight, why XPeng had been slower than rivals to fill its domestic lineup: "Why did we not create too many localised products and services in China at the beginning? Because I think they would not be able to go global. We may take 12 more months to prepare than others, but we will earn it back later."
It is impossible to know how much of that was strategy and how much was a story told about delays. Either way, the global expansion that followed was methodical. Germany came in 2024, following an announcement at IAA Mobility in Munich. Singapore came in July 2024, where by XPeng's account the G6 soon matched Tesla's monthly sales. Malaysia launched on August 27, 2024, with a first dealership in Glenmarie. Thailand followed, then Spain and Portugal through the Salvador Caetano Group. Indonesia came through Erajaya, a local electronics distributor, a detail that would have pleased the man who once sold browsers through phone makers. The United Kingdom arrived in February 2025, with a right-hand-drive G6 from £39,990. Italy was announced at Milan Design Week in 2025. Kit assembly began in Purwakarta, West Java, in July 2025. And in September 2025 came full production in Graz, Austria, at Magna Steyr, which has built cars under contract for some of Europe's most storied brands.
Not every expansion went cleanly. XPeng entered Australia through a third-party distributor, TrueEV. In March 2026, TrueEV entered receivership and sued XPeng, which then launched its own factory-backed operation in the country. Selling cars abroad, it turned out, involved lawyers.
In 2025, XPeng delivered 45,008 cars outside China, up 96 percent, about a tenth of its total. By year-end it operated in sixty countries and regions. That April, in Hong Kong, He had offered two numbers. Half of new cars sold in China in 2024 were electric or plug-in hybrids, and 55.7 percent of those came with hands-off driver assistance. When XPeng was founded in 2014, both figures had been below one percent. "Over the next decade," he said, "we aim for over half of our sales to come from overseas, striving to really dominate as China's leading mid- to high-end export brand." His February 2026 letter set a target of one million overseas sales a year by 2030.
The Fortune reporter who wrote down those numbers also noted that XPeng was still "jockeying for position around seventh or eighth place among China's new energy vehicle brands." Both observations were accurate.
Make the Other Man Run
He Xiaopeng has played table tennis since childhood. Asked by CNA in 2024 whether the sport resembled entrepreneurship, he gave two answers. The first was about stamina, since entrepreneurs need a good physique. The second was about style. "When playing table tennis, my strength lies in my agility and I am able to make others run around the whole court, while I don't run as much. So it may be somewhat similar to my corporate style."
It is an unusually candid answer, and it describes much of how XPeng has been built. Haima built the first cars. CATL supplied the first batteries; CALB and Eve Energy supplied the later ones. Socionext integrated the chip. Magna Steyr builds the European cars, and Erajaya distributes them in Indonesia. AutoNavi, Alibaba's mapping unit, would anchor the robotaxi network, which He pointedly described as "more open" than Tesla's, with each XPeng branch able "to independently verify its business feasibility" so that "resources can be reorganized at any time to reduce the financial burden." Huawei co-developed the 87-inch augmented-reality head-up display in the new P7. Volkswagen, finally, paid to run on XPeng's architecture. The fixed costs sat with partners. XPeng kept the parts it considered the brain: the driving model, the chip, the electrical architecture, the operating system.
Even the founding fits the pattern. Xia Heng and He Tao registered the company and built the prototypes. He Xiaopeng wrote checks and lent his name, then arrived three years later to take the chair. The capital he contributed was enormous, the half of $400 million in 2019 above all. The labor of the early years was mostly other people's.
His rivals noticed. Chinese internet shorthand groups the leading smart-EV players as Hua-Mi-Wei-Xiao-Li: Huawei, Xiaomi, Nio (Weilai), XPeng (Xiaopeng), and Li Auto (Li Xiang). In that phrase the syllable Xiao stands for him. Lei Jun, the friend who invested in XPeng in 2014 and attended its tenth birthday, built his own car company in the meantime. Xiaomi now competes with XPeng for the same young, technology-obsessed buyer. Friend, investor, rival: in Chinese technology these categories have always overlapped.
He predicts the overlap will end in a shakeout. The industry's "elimination round," he has said, runs at least from 2025 to 2027. He expects China's EV market to move within five years from a knockout stage to an "all-star game" with about five serious players. In 2024 he guessed seven. The number keeps shrinking as his own position improves.
Three Miracles in One Year
On February 24, 2026, the first working day of the Year of the Horse, He Xiaopeng sent a letter to XPeng's employees worldwide. Its title, in translation, was "Steady Progress, Breaking the Game." He reported that he had spent the Lunar New Year holiday personally testing the second-generation VLA driving model, and that it had moved, in his words, from "showing off" to "emergence." He declared 2026 a historic inflection point. He set the company a goal that no technology company anywhere had met: mass production of three frontier businesses in the same calendar year. Those were robotaxis, humanoid robots, and flying cars.
Read as a list, it resembles the joke Lashinsky made in 2019, freezing brains and building cities on water. Read as a schedule, it is steadily being kept. The flying-car unit, XPeng AeroHT, was a Dongguan startup founded in 2013 that XPeng invested in and renamed in 2020. It had shown a six-wheeled, range-extended van called the Land Aircraft Carrier, which carries a detachable two-seat electric aircraft in its rear, and the first units had come off a trial production line. In 2025, XPeng and AeroHT together spent about 10 billion yuan on R&D. The 2026 budget is nearly 15 billion. He has said he believes the flying-car market will eventually be bigger than the EV market, and that flying vehicles could enter daily life within five to ten years.
On May 18, 2026, Reuters reported that XPeng had begun mass-producing its first robotaxi at its Guangzhou headquarters, built on the new GX platform, with pilot operations planned for the second half of the year and fully driverless operation in early 2027. On July 9, XPeng's robotaxis began internal testing with employees as passengers.
On June 10, He issued another internal letter, announcing that he would personally take direct control of the robotics business. Twelve days later he was appointed an independent director of Ant Group, Alibaba's financial affiliate, a return in a new role to the conglomerate that had once said no to his car. On August 24, XPeng's robotics business closed its first external round, more than $900 million at a valuation of $6.3 billion. On September 8, XPeng opened IRON's production line.
A man who had sat in the National People's Congress since January 2023 brought three proposals to its March 2026 session, one each on autonomous driving, humanoid robots, and flying cars. He said he was "both excited and under pressure." He pronounced 2026 "the first year" of physical AI in China. He promised to improve the company's driving capability five- to tenfold by year-end, from "multiple takeovers per 100 kilometers" to "one takeover per 1,000 kilometers or even per 10,000." He suggested that humanoid robots, being less dangerous than cars or airplanes, could be regulated later, once the industry was larger. In a separate interview in 2024 he had cautioned against the wilder AGI timelines: "We need to believe in the coming of the AI era, but at the same time be prepared for decades of gradual maturation and gradual scale." He did not seem to find these positions in tension, and perhaps they are not.
In the first quarter of 2026, XPeng delivered 62,682 cars and lost 1.78 billion yuan. Its gross margin was 20.6 percent, the highest it had reported. Profit had arrived for one quarter and left again. The company had 19,884 employees and planned to add 8,000 more.
Before all of that, on the last weekend of January 2026, XPeng brought IRON to a shopping mall in Shenzhen.
Hundreds of people gathered. The robot did its catwalk, the one that had made the internet sure it was a woman in a suit, smoothly, to the center of the stage. Then, standing still, it lost its balance and fell backward. A staff member caught it and broke the fall. It ended face down on the stage. The clip spread across Chinese social media; a single topic about it on RedNote drew some ten million views. Commentators noted the difficulty of spine and waist mechanics. He compared the fall to a toddler learning to walk. "Developing robots means getting used to failure," he would say later, in a short film released after the production line opened.
That afternoon, after adjustments, IRON came back. It finished its program with the children, asking them trivia, among other things what color a giant panda's poop would be.
On Sunday, the second day, it did not walk. It stood strapped to a frame, talking with the crowd, and somewhere near it, out of frame in every video, stood a person whose job was to catch it.
Part IIThe Playbook
What follows are eleven principles drawn from He Xiaopeng's three companies and, more instructively, from the reversals inside them. These are not the lessons he would put on a slide. Several run against the grain of conventional founder advice, and at least two of them, the decision to kill working systems and the decision to lend his name before he ran the company, carry risks that a less capitalized operator could not survive. Each is grounded in a specific decision from the record above, and each ends with a tactic.
The principles
- 1.Write the exit condition early, then treat reaching it as data, not destination.
- 2.Back the thing before you run it.
- 3.Be the investor of last resort in your own round.
- 4.Let your name become the liability.
- 5.Know when the chairman must become the operator.
- 6.Choose the high ceiling over the safe floor, and say so in numbers.
- 7.Reverse publicly, without ceremony.
- 8.Sell your brain to the incumbent.
- 9.Price globalization in decades, not quarters.
- 10.Prove it by cutting it open.
- 11.Make the other side run the court.
鹏
The Decision Record
The moments when He Xiaopeng bet, broke, or rebuilt, from a browser to a robot line.
1995
Sets a goal, by his account at eighteen, of financial freedom before forty; enters South China University of Technology.
2004
Co-founds UCWeb with Liang Jie on an 800,000-yuan loan from NetEase's Ding Lei.
2014
Alibaba buys UCWeb for $4.3B. He backs Xia Heng and He Tao's startup, which takes his given name when "Orange Car" is unavailable.
2017
Resigns from Alibaba on August 22; joins XPeng as chairman on August 29.
2019
Personally funds about half of a $400M round as the pandemic approaches.
2020
XPeng raises $1.5B in its NYSE IPO; shares rise 41% on day one.
2021
P5 launches as the first mass-produced car with lidar.
2022
G9 disappoints; more than a dozen senior executives are replaced within a year.
2023
Takes over as legal representative; Volkswagen invests $700M; Didi's autonomous unit acquired.
2024
Lidar dropped with the P7+; MONA M03 launches; Turing chip tapes out.
2025
Kills VLA Gen 1; 429,445 deliveries; first profitable quarter.
2026
Takes direct control of robotics; robotaxi and IRON production begin.
Principle 1
Write the exit condition early, then treat reaching it as data, not destination
The four-road taxonomy (lottery, property, options, a startup) is the least romantic founding philosophy imaginable, and that is its strength. By defining success at eighteen as a measurable condition, financial freedom before forty, He gave himself a test he could actually pass. He passed it at thirty-six. The more interesting fact is what he did with the result. Rather than treating freedom as an endpoint, he treated it as an experiment that had returned an unexpected answer: the problem he set had been solved, and he was not satisfied.
Most founders either never define their exit condition or never admit when they have hit it, and either way they confuse momentum with meaning. He's sequence was cleaner. He set a goal, met it, noticed the emptiness, and then used the freedom the goal had bought to attempt something with no defined endpoint at all. The car company carries no "freedom before X" clause. It carries his name.
Tactic: Write down, in one sentence, the measurable condition under which you would stop working on your current company, and decide in advance what you will do with the information if you hit it and keep going anyway.
Principle 2
Back the thing before you run it
From 2014 to 2017, He Xiaopeng was XPeng's investor and namesake while remaining an Alibaba executive. That arrangement is usually told as hesitation, but it worked as a three-year option. He paid to keep the right to commit, while Xia Heng and He Tao did the expensive, uncertain work of turning a concept into four prototypes and fifteen Identy X cars. When he finally arrived in August 2017, he walked into a company with a team, a vehicle, and a direction. He had also been honest with himself about his own weakness: that internet people were "overly confident" about hardware and "needed a process of re-growth."
The lesson is not to be passive. It is to buy time to learn a domain you do not yet understand, while people who do understand it build the base. The cost of the option was capital. The value was three years of watching someone else's mistakes before making his own.
Tactic: Before you leave a secure role to found in an unfamiliar industry, fund and advise a team already operating there for a set period, and keep a running list of what you were wrong about.
Principle 3
Be the investor of last resort in your own round
In late 2019, with the pandemic approaching and the auto startup market souring, He personally supplied roughly half of a $400 million round. He also made a public pledge that if XPeng failed he would pay salaries and suppliers first. In August 2024, with deliveries far below target, he bought about HK$107 million more of XPeng stock. These were not symbolic gestures. They put the founder's balance sheet between the company and its creditors, and they told every investor watching that the person with the most information was buying.
The approach requires money most founders do not have, and it concentrates risk in a way a portfolio manager would hate. But it changes the signaling problem completely. When the person whose name is on the door writes the biggest check during the worst stretch, outside investors stop asking whether he believes. They start asking how long he can afford to.
Tactic: When the market will not fund you at a fair price, decide in advance what share of the gap you will cover personally, and announce it in specific terms (what you will pay, and to whom first) rather than as a vague statement of commitment.
Principle 4
Let your name become the liability
XPeng carries He's given name by accident, because an orange trademark was taken. For seven or eight years, friends told him that renaming the company would double sales. He refused. In China a company named after its boss is faintly strange, and a brand tied to one person absorbs every one of his mistakes: the CES boast about Musk, the G9, the robot falling in a mall.
What he gets in exchange is accountability with no exits. There is no corporate abstraction to hide behind. When the city-driving system was poor, he said so on video himself. When the robot fell, he answered for it. The name became a commitment device, a way of making retreat more expensive than persistence.
Tactic: If your personal identity is fused with your company, stop treating that as a branding risk to manage and use it as a discipline: commit to being the first to announce your own failures.
Principle 5
Know when the chairman must become the operator
From 2017 to 2022, He ran XPeng largely as a chairman, delegating product and engineering to co-founders and executives. The G9's failure in 2022 ended that arrangement. Within a year more than a dozen senior executives were gone. By March 2023 he had become legal representative of the operating company, Xia Heng had stepped down as executive director, and He said he was "overseeing all six R&D teams."
The shift was painful, and by his own description it was the darkest moment of his career. But it reflects a precise judgment: a consensus-driven structure works in a growing market and fails in a price war, where decisions must be fast, unpopular, and sometimes contradictory. The founder who keeps the chairman's distance when the company needs an operator is choosing his own comfort over the company's survival.
⇄
Chairman Mode vs. Operator Mode
How XPeng's governance changed after the G9.
| Dimension | 2017–2022 | 2023 onward |
|---|---|---|
| Product decisions | Delegated to co-founders and executive teams Consensus | Founder oversees all six R&D teams Direct |
| Executive bench | Stable, broad, collegial | More than a dozen senior leaders replaced in a year Painful |
| Platform strategy | Feature-rich flagships (G9) | SEPA 2.0 standardization, cost focus |
| Outside partners | Buyer of components | Seller of architecture to Volkswagen Licensor |
| Hardest calls | Deferred | Made by the founder personally (robotics, VLA) |
Tactic: Set a concrete trigger, such as a flagship miss or two consecutive quarters below plan, at which you will personally take over the function that failed, and write it down before you need it.
Principle 6
Choose the high ceiling over the safe floor, and say so in numbers
The decision to kill VLA Gen 1 in spring 2025 is the clearest example of He's thinking. The old system had a floor of 900 and a ceiling of 1,000. The new one had a floor of 100 and a theoretical ceiling in the hundreds of thousands. He chose the low floor, knowing it meant billions in sunk cost, more than two hundred departures, and perhaps a one-in-five chance on that particular technical path.
What makes this a usable principle rather than a slogan about boldness is the framing. By putting both options on a single numeric scale, he made the trade-off something people could argue about, rather than a vague matter of conviction. His team could disagree about the numbers, and the numbers were not the decision. But the frame turned "Are we brave?" into "Which ceiling do we need?" For a company aiming at Level 4 autonomy, 1,000 points was simply not enough, and no amount of incremental work would change that.
Tactic: When weighing an incremental path against a disruptive one, estimate each one's floor and ceiling on a single scale, and ask whether the safe option's ceiling clears the bar your strategy actually requires.
Principle 7
Reverse publicly, without ceremony
XPeng put lidar into a mass-produced car first, publicly and against Musk's mockery, and then took it out of every model once its camera-and-radar system no longer needed it. It hired Marvell for its chip and fired Marvell for its chip. It built a hands-free driving system its chairman mocked on camera, and then tore it down. None of these reversals came with a grand mea culpa. They were simply announced through the product.
The founders who survive long hardware cycles are rarely the ones with the best initial bets. They are the ones who treat their own past positions as no more binding than a competitor's. He seems almost immune to the sunk-cost attachment that makes most leaders defend their prior decisions. The lidar he championed had been a means, not an identity.
Tactic: For every flagship technical bet, write down the condition under which you would abandon it, and review that list each quarter with the same coldness you apply to competitors' bets.
Principle 8
Sell your brain to the incumbent
The Volkswagen relationship began as an equity investment and became something stranger: China's oldest foreign carmaker paying XPeng 1.72 billion yuan in six months for technology, at a margin car sales could never approach. It also adopted XPeng's electrical architecture on its electric, combustion, and hybrid platforms, and agreed to be the launch customer for XPeng's next driving model. The student sold to the teacher.
The strategic logic is double. Licensing revenue subsidizes the R&D the core business cannot yet fund on its own. More important, an incumbent's adoption validates the technology in a way no marketing could. A startup competing with an incumbent on cars loses. A startup supplying that incumbent with its brain gains a source of revenue, a stamp of credibility, and a reason for the competitor to want it alive.
Tactic: Identify the one component of your product that an incumbent cannot build fast enough, and design it from the start to be licensable as a standalone module.
Principle 9
Price globalization in decades, not quarters
"Going global is at least a 10-year plan," He said in 2024, from a starting point of 211 cars sold in Norway. He claimed that XPeng had deliberately avoided China-only features early on because they would not travel, accepting being "12 more months" behind rivals. The rollout that followed was methodical. It used local partners (Salvador Caetano, Erajaya, Magna Steyr), local assembly in Indonesia, Malaysia, and Austria, and patience with early disappointments like the TrueEV collapse in Australia.
The underlying principle is that most companies treat overseas expansion as a quarter-by-quarter growth lever and abandon it when the early numbers disappoint. Treating it as a decade-long build changes which decisions make sense. It justifies building global-ready products from the start, and it absorbs the inevitable early failures without strategic panic. In 2025, overseas deliveries nearly doubled. The target is one million a year by 2030.
Tactic: Before entering a new market, set a ten-year objective and a deliberately modest year-one target, and judge the market on the trajectory of learning rather than on early volume.
Principle 10
Prove it by cutting it open
When the internet accused IRON of being a person in a suit, He did not issue a statement. He unzipped the robot on camera and, over his team's objections, had it dissected live, layer by layer, while it kept walking. When it later fell in a Shenzhen mall, he did not hide the clip. He compared it to a toddler learning to walk, and later included the fall in a public film about the robot's development.
Radical transparency of this kind is expensive. It means accepting humiliation as a cost of credibility. But skepticism toward ambitious hardware is total and quick, and the only argument that works is a physical one. His reasoning was explicitly about the "99% of non-industry users," the people who would never read a technical paper but would watch a video of a robot walking with its leg cut open.
Tactic: When your claim is doubted, find the most direct physical demonstration that settles it, and be prepared to show your failures with the same openness.
Principle 11
Make the other side run the court
He's table tennis self-description, making opponents run around the whole court while he moves less, is a surprisingly exact account of XPeng's architecture. Contract manufacturers built its first cars and still build its European ones. Battery suppliers carry the cell risk. Local distributors carry market-entry costs. Volkswagen pays for technology XPeng would have built anyway. AutoNavi anchors the robotaxi network. XPeng has spent heavily on the narrow set of things it considers the brain (the driving model, the chip, the electrical architecture) and let partners absorb the capital weight of nearly everything else.
The risk is dependence: a distributor can collapse, and a contract partner can lag. The advantage is that capital goes where differentiation lives, and the company stays able, in He's words, to "reorganize resources at any time."
Tactic: List every function in your company, mark the two or three that truly differentiate you, and find a partner willing to carry the fixed costs of the rest.
Free playbook
Get The Business Model Playbook
58 business models, one visual page each: how the money flows, the metrics that matter, and who runs it. Free when you join the Faster Than Normal email.
Free. No spam. Unsubscribe anytime.
Part IIIQuotes and Maxims
In their words
— He Xiaopeng, interview with CNA, March 2024We have taken a very slow path in the past three years because many people think that you should be able to achieve results quickly as soon as you enter the international market. I never thought so.
— He Xiaopeng, interview with CNA, May 2024When playing table tennis, my strength lies in my agility and I am able to make others run around the whole court, while I don't run as much. So it may be somewhat similar to my corporate style.
— He Xiaopeng, after XPeng's AI Day, November 2025XPeng has never fully believed that simply making a car bigger, more beautiful, cheaper, and of higher quality can lead to victory. These are only necessary conditions, not sufficient ones.
— He Xiaopeng, on Zhang Xiaojun's podcast, May 2026You want to build a product with a ceiling of 1,000 and a floor of 900, but back then the floor might have been only 100.
— He Xiaopeng, on AGI, 2024We need to believe in the coming of the AI era, but at the same time be prepared for decades of gradual maturation and gradual scale.
Maxims
- Define the finish line, then distrust it. A measurable goal tells you when you have arrived, and whether arriving was the point.
- Buy the option before you take the job. Funding a team in an unfamiliar industry is cheaper than learning that industry by running it.
- The founder writes the last check. When the market won't fund you, your own money is the clearest signal of what you believe.
- A name on the door closes the exits. Personal identification with a company is a liability you can use as discipline.
- Chairmen deliberate; operators decide. When the market turns brutal, consensus is a luxury the company can't afford.
- Compare ceilings, not comfort. The safe path is a bad bet if its ceiling sits below the bar your strategy requires.
- Yesterday's flagship bet is today's line item. Drop the lidar, the vendor, or the model the moment it stops serving the goal.
- Sell the brain, not the body. Licensing your core technology to an incumbent turns a competitor into a customer and a validator.
- Going global takes a decade. Judge foreign markets by the trajectory of learning, not first-year volume.
- When they doubt the machine, open it up. Physical proof beats any statement, and a public failure is the price of public credibility.
