XPeng Inc.: Betting the Company Twice

I. Introduction & Episode Roadmap

On the morning of March 20, 2026, the finance team at 小鹏汽车 XPeng's Guangzhou headquarters published a sentence the company had waited eleven years to write. In the fourth quarter of 2025, XPeng had earned a net profit of RMB 380 million — the first profitable quarter in its history.¹ The stock jumped. The headlines wrote themselves. A Chinese electric-vehicle startup that had come within sight of the abyss three years earlier had finally, definitively, crossed over.

Five months later, on August 24, 2026, the same company reported a quarterly net loss of RMB 1.34 billion — wider than the loss it had posted in the same quarter a year earlier, back when it was supposedly still losing money on the way up.² Add the first quarter and XPeng's losses in the first half of 2026 came to RMB 3.12 billion,³ nearly three times the RMB 1.14 billion it lost across the whole of 2025.⁴

That is the puzzle at the centre of this story. XPeng is simultaneously one of the most impressive operational turnarounds in the global auto industry over the past three years and one of its least settled investment cases. Both statements are true, and they are true for the same reasons.

The temptation with a company like this is to pick a side — declare the turnaround real and the losses a rounding error, or declare the losses fatal and the turnaround a mirage. Neither reading survives the accounts. What XPeng actually offers is something more useful to a long-term investor: a business where the bull and bear arguments rest on different line items of the same income statement, which means the disagreement is resolvable by watching the right numbers rather than by arguing about the story.

The company delivered 429,445 vehicles in 2025, up 126% year over year, on revenue of RMB 76.72 billion.⁴ That is roughly three and a half times its 2022 volume. Its gross margin went from 1.5% in 2023 — effectively zero — to 18.9% in 2025.⁴⁵ It is dual-primary listed in New York as XPEV and in Hong Kong as 9868.HK, and it sells cars in sixty countries.⁶ And it still has not demonstrated that it can earn an operating profit for two quarters in a row.

The four things that make this story worth two hours. First, XPeng nearly died in 2022 — not from a technology failure but from something far more mundane and more instructive: it could not explain its own product to customers. Second, it was pulled back by a 30-year veteran of a rival Chinese automaker, a hire that is arguably the single best-evidenced management decision in the modern Chinese EV industry. Third, it signed a deal that ran the historical direction of automotive knowhow backwards — a loss-making Chinese startup licensing autonomous-driving software to Volkswagen, and getting paid enough for it that the money now props up the company's reported margins. Fourth, it is now spending an accelerating share of its research budget on flying cars and humanoid robots while its core vehicle margin goes down rather than up.

Here is the road map. We start with a browser company and a founder who sold his life's work to Alibaba. We go through the launch that nearly ended everything, the president who fixed it, and the German alliance that changed the company's economics. Then we war-game the industry XPeng actually competes in, take apart the profitability claim number by number, put current management's promises next to their outcomes, size the "physical AI" side bets against their own commercialization record, and lay out what would have to happen for either the bull or the bear case to win.

A note on posture before we begin. XPeng's management tells a coherent, well-argued story about why the company wins. Coherent is not the same as proven. Wherever this article records a claim — about moats, about margins, about robots — it also records the strongest evidence in XPeng's own history that cuts against it, and says which way the evidence leans.

↑ Back to Top

II. Origins: From Browsers to Bodywork (2004–2020)

In June 2014, 阿里巴巴 Alibaba announced it was buying out UCWeb, the maker of China's most popular mobile browser. Alibaba never disclosed the price. What it did say was that the deal was worth more than the $1.9 billion it had paid for 91 Wireless the year before, which made it, by the company's own description, the largest internet merger in Chinese history.⁷ One of UCWeb's co-founders was a 36-year-old computer science graduate of South China University of Technology named 何小鹏 He Xiaopeng.

He is worth understanding, because XPeng is a company built in his image — for better and for worse. He is not a car person. He is a software person who spent a decade in the brutal, iteration-obsessed world of Chinese mobile internet, where you ship weekly, measure everything, and assume your product will be copied within a quarter. That instinct — treat the car as a computing platform that happens to have wheels — is the founding premise of XPeng and the source of both its genuine differentiation and its recurring habit of over-engineering things customers did not ask for.

He backed the automotive venture as an investor from 2014 while still at Alibaba, where he ran the mobile business group after the acquisition. He left in August 2017 to take the chairman's role full-time. It was not an obvious trade: he was a wealthy executive at China's most powerful internet company, walking into the most capital-punishing industry on earth at the moment three or four hundred Chinese EV startups were being funded simultaneously.

The early capital came from the venture ecosystem he already knew — GGV, IDG, Morningside, Shunwei — and then, in January 2018, from a RMB 2.2 billion Series B led by Alibaba, Foxconn and IDG Capital. The Alibaba cheque mattered less for the money than for the signal: it told suppliers and local governments that this particular startup would not evaporate.

There is a small detail in the company's name that tells you something about the culture. XPeng is 小鹏汽车 — literally "Xiaopeng Motors," the founder's own given name on the badge. He reportedly resisted it. In an industry where Chinese startups were choosing aspirational abstractions, XPeng ended up with a name that made the founder personally identifiable with every product decision. Given what happened in 2022, that turned out to be more consequential than a branding footnote.

The first production car, the G3 compact SUV, arrived in December 2018 — competent, cheap, unremarkable. The car that established the brand was the P7 sedan, unveiled in April 2019: long range, sleek, and carrying an assisted-driving system XPeng developed in-house rather than buying from a supplier.

The P7 is where the "smart EV" positioning stopped being a slogan and started being a product feature customers would pay for. That distinction matters more than it sounds. Almost every automaker in 2019 claimed software differentiation; almost none of them could point to a specific feature that changed what a buyer was willing to pay. XPeng could, and the reason was structural: because it wrote the software itself, it could ship improvements to cars already sold. A conventional automaker's car is worth slightly less every year it sits in a driveway. XPeng's argument was that its cars would get better. Whether customers valued that enough to pay a premium was, at the time, an open question — and in the price war that arrived three years later, it became the central one.

The capital-markets sprint. In August 2020, XPeng sold 99.7 million American Depositary Shares at $15 each on the New York Stock Exchange, raising about $1.5 billion — an offering it had upsized twice on the way in, having originally planned to price between $11 and $13.⁸ The shares opened at $23.10 and closed above $21. Eleven months later, in July 2021, XPeng priced a Hong Kong offering of 85 million Class A shares at HK$165 apiece, raising roughly HK$14 billion.⁹ The Hong Kong listing was structured as a genuine dual-primary listing rather than a secondary listing, which mattered more than it sounded: it insulated the company from the delisting risk hanging over US-listed Chinese issuers at the time, and it made XPeng eligible for mainland investor flows through Stock Connect.

Both raises came at the peak of global enthusiasm for anything electric. That timing is a large part of why XPeng survived what came next: the balance sheet it built in 2020–2021 bought it the right to make an expensive mistake.

And 2021 looked like vindication. XPeng delivered 98,155 vehicles, up 263% from 2020, on revenue of RMB 20.99 billion.¹⁰ The IPO narrative — software-first Chinese EV maker compounds at triple digits — was, for one year, exactly what the numbers showed.

What investors should take from the origin story. XPeng's founding advantage was never manufacturing; it was a founder with capital, credibility and a software worldview entering an industry that had not yet been rebuilt around software. That advantage is real but inherently temporary — worldviews get copied. The durable asset from this period is duller and more important: two large equity raises completed at generous valuations before the window closed. That war chest is the reason the next chapter is a near-death experience rather than a death.

↑ Back to Top

III. The Fall: The G9 Launch Disaster and the 2022 Reckoning

On the evening of September 21, 2022, XPeng launched the G9. It was meant to be the moment the company graduated. The G9 was a genuinely advanced flagship SUV — an 800-volt high-voltage architecture with ultra-fast charging at a time when most rivals were still on 400-volt systems — priced from around RMB 309,900 and aimed squarely at the premium segment where Tesla's Model Y and NIO's SUVs made money.¹¹

Then the configuration slide went up, and the Chinese internet went to work on it.

The launch lineup ran to six versions, distinguished by letters spelling out X-P-E-N-G, layered on top of separate range options and separate feature packages. Buyers could not work out which version had which capability. Worse, XPeng's own assisted-driving system — the single feature the entire brand was built around — was not available at all on the entry trims. Reviewers who had spent the week praising the hardware spent the night trying to reverse-engineer the price list. Reservation holders began cancelling.

To understand why this was fatal rather than merely embarrassing, you have to understand how a Chinese EV launch actually works. Demand is concentrated into a launch window measured in days, not months. Buyers place deposits within hours of a price reveal, social-media sentiment forms overnight, and the order book built in that first week sets the production plan, the dealer allocation and the media narrative for the following year. A launch is not the start of a sales process; it is the sales process, compressed. Get the configuration wrong and there is no gradual recovery — the window closes and the segment moves on.

Forty-eight hours later, XPeng scrapped it. On September 23, the company republished the G9 range: the letter names were gone, replaced by Plus, Pro and Max in a straight borrowing of Apple's convention; the lineup went from six versions to eight; the XPilot assisted-driving system became standard across every trim; and two new intermediate price points were added at RMB 369,900 and RMB 419,900. XPeng's own explanation was that the changes were meant to reduce "the difficulty of choosing."¹²

It was, in isolation, an admirable correction — fast, public, and unspun. But the damage was structural rather than cosmetic. A flagship gets one launch. The G9's had been spent, and the segment it was aimed at was the one where XPeng most needed a foothold, because premium SUVs are where Chinese EV makers earn the gross margin that funds everything else.

The year the arithmetic broke. XPeng finished 2022 with 120,757 deliveries, up 23%.¹⁰ In almost any other industry, 23% growth would be a fine year. Following 263%, in a market growing far faster, against a build-out of factories, stores and R&D sized for a much steeper curve, it was a rout. Revenue rose to RMB 26.86 billion but the net loss widened to RMB 9.14 billion, from RMB 4.86 billion the year before, and the company ended the year with RMB 38.25 billion in cash and equivalents — still a large cushion, but one that had started to drain rather than fill.¹³

The market's verdict was brutal and fast. By the end of September 2022, XPEV was down 74% for the year.¹⁴ It did not recover before December; the ADSs finished 2022 below $10, against roughly $50 twelve months earlier.

The cleanest disconfirming evidence in the whole story. Companies rarely tell you plainly that their thesis has failed. XPeng did. On January 30, 2023, Bloomberg reported that XPeng had pushed back its profitability target — from the late-2023-to-early-2024 window management had previously pointed investors toward, out to 2025.¹⁵ That is not analyst scepticism; it is management conceding on the record that the "scale into profit" plan was not working on the schedule they had sold.

Read carefully, the 2022 failure was not really about a price list. He Xiaopeng's own diagnosis, delivered internally and later reflected on the Q4 2022 earnings call, was organisational: XPeng had grown a structure in which product planning, marketing and sales did not have a single accountable owner, and in which nobody with genuine automotive commercial experience had the authority to stop a launch. On that call, He described a year in which "a challenging macro environment and increasingly intense competition in the NEV market placed pressure on our performance," then spent most of his time describing a flattening of the hierarchy — design, R&D, production and supply chain moved to report directly to him — rather than defending the product.¹⁶

So what. For an investor, the G9 episode is the most useful single event in XPeng's history, because it establishes the company's actual failure mode. XPeng does not typically fail at engineering. It fails at commercial translation — at converting a technical advantage into something a buyer can understand and pay for. Every subsequent claim in this story, including the current one about humanoid robots, deserves to be tested against that specific weakness rather than against a generic "can they build it" question. Which brings us to the person hired to fix exactly that.

↑ Back to Top

IV. Rock Bottom to Reorg: Wang Fengying (王凤英) and the Turnaround

In January 1991, a young woman took a frontline sales job at an obscure vehicle workshop in Baoding, Hebei — a company that at the time built pickup trucks and had no meaningful brand. Over the next three decades she rose through sales to general manager and eventually to president, and the workshop became 长城汽车 Great Wall Motor, one of China's largest privately owned automakers and the country's dominant SUV brand. Her name is 王凤英 Wang Fengying, and she is one of the very few executives in the Chinese auto industry with three uninterrupted decades of commercial operating experience at a company she helped build from nothing.

On January 30, 2023 — a date that fell, not coincidentally, on the same day as the Bloomberg report about the deferred profit target — XPeng announced that Wang had joined as President, responsible for product planning, portfolio management and sales operations, reporting directly to He Xiaopeng.¹⁷¹⁸ CnEVPost described it as the most significant hire XPeng had made in years, arriving at what was plainly the lowest point in the company's life.¹⁷

The choice is more interesting than it first appears. He Xiaopeng did not hire another software executive, another autonomous-driving researcher, or a turnaround consultant. He hired the single function XPeng had proven, in public and at scale, that it could not do: figuring out what to build, how to name it, how to price it, and how to sell it. Founders very rarely hire against their own instincts. This one did.

What she actually changed. The first target was the sales organisation. XPeng had been running a fragmented structure with overlapping product lines and diffuse regional accountability. Wang consolidated it, redrawing the sales map into a smaller number of districts each with a single accountable head. In September 2023 she launched what XPeng internally called the "Jupiter Plan," shifting the company away from a purely owned-store direct-sales model toward a hybrid of company stores and franchised dealers — a pragmatic, unglamorous decision that let XPeng expand into lower-tier Chinese cities on somebody else's capital, at a moment when its own capital was precious.

The second target was procurement. Wang ran an anti-corruption cleanup through the supply-chain organisation, with at least one vice president suspended. This gets less coverage than the product decisions and probably deserves more: in an industry where component costs are 70–80% of the bill, procurement integrity is a margin line item, not a compliance footnote.

The third was scope creep in her own favour. By early 2024 she had also taken over the international division, which is the part of the business now growing fastest.

The MONA deal. The most consequential single transaction of the Wang era was announced on August 28, 2023. XPeng agreed to acquire the assets of 滴滴 DiDi's smart-car development business — the research, design and engineering work behind DiDi's own EV programme — for up to HK$5.835 billion, roughly $744 million, paid entirely in newly issued Class A shares representing about 3.25% of the enlarged share capital. DiDi became a strategic shareholder with a 24-month lock-up, agreed to plug XPeng's new mass-market brand into its nationwide ride-hailing ecosystem, and the acquired programme — codenamed MONA — was aimed at the RMB 150,000 price band, well below anything XPeng then sold.¹⁹

As capital allocation, this is worth pausing on. XPeng bought a finished low-cost EV platform, an engineering team that had already solved the cost problem, and a distribution channel, for paper rather than cash, at a moment when its own cash was the only thing keeping it alive. Compare that to the alternative — developing a clean-sheet entry platform internally over three years while burning RMB 10 billion a year — and the logic is hard to argue with. It also carried an obvious risk, which was that DiDi's assets might be worth less than the shares given up. The subsequent volume evidence, which we will get to, suggests they were not.

Does the turnaround claim survive contact with the record? This is the section where the temptation to write a hagiography is strongest, so let us test it properly. The claim is: a specific management change in January 2023 produced a durable operating improvement.

The affirmative evidence is unusually clean. Deliveries went 120,757 (2022) → 141,601 (2023) → 190,068 (2024) → 429,445 (2025).¹⁰⁵²⁰⁴ More importantly, the profitability of each of those units transformed. Vehicle margin — the gross margin on cars alone, stripped of everything else — was negative 1.6% in 2023, meaning XPeng lost money on every car before paying a single engineer or salesperson. It went to 8.3% in 2024 and 12.8% in 2025.⁵²⁰⁴ Net loss went from RMB 10.38 billion in 2023, the worst in company history, to RMB 1.14 billion in 2025.⁵⁴

The strongest counterevidence is timing. The 2023 numbers, which were Wang's first full year, were the worst in company history on margin — gross margin of 1.5% against 11.5% in 2022 — because that was the year XPeng cut prices to clear G9-era inventory and rebuilt its lineup.⁵ Turnarounds have J-curves, and this one's trough came after the new president arrived. An investor buying the "Wang fixed it" story in mid-2023 would have watched the evidence get worse for twelve months before it got better.

The second piece of counterevidence is attribution. The 2024–2025 volume explosion is inseparable from the MONA product, which came from an acquisition, and from a broader Chinese EV market that grew rapidly through 2025. Untangling how much of the recovery is organisational competence and how much is a well-timed purchase plus a rising tide is genuinely hard.

The fair conclusion: the historical record supports the claim in a narrower form than the headline version. XPeng under Wang demonstrably fixed the specific things it had broken — product definition, pricing clarity, sales-channel reach, and unit economics on the vehicle itself. It did not demonstrate that XPeng can grow profitably through a downturn, because it has not yet been through one under her. Her own three-year share grant, structured to vest on service rather than share price, fully vested in March 2026 — the first equity she had ever held in an employer after thirty stock-less years at Great Wall. That she stayed to collect it is itself a data point; whether the design rewards the right things is a separate question we return to later.

The turnaround also had an unexpected consequence. It made XPeng credible enough that the world's second-largest automaker came shopping.

↑ Back to Top

V. The Volkswagen Alliance: A Reversed Technology Flow

For roughly forty years, the technology trade between the global auto industry and China ran in one direction. Volkswagen entered China in 1984 through a joint venture with SAIC, and the bargain — repeated by every foreign automaker that followed — was straightforward: the Western partner brought platforms, engines and process knowledge; the Chinese partner brought market access. Foreign firms taught; Chinese firms learned.

On July 26, 2023, Volkswagen Group announced it was investing approximately $700 million for a 4.99% stake in XPeng, alongside a technical collaboration to jointly develop two Volkswagen-brand mid-size electric vehicles for the Chinese market on an XPeng-derived platform.²¹²² Volkswagen framed the rationale in its own announcement with unusual candour: partnering with XPeng would cut time-to-market by more than 30%.²¹ A ninety-year-old German industrial giant had concluded it could not develop a competitive Chinese EV fast enough on its own, and was buying the capability from a company that had lost RMB 9 billion the previous year.

The equity closed in December 2023, with XPeng issuing new Class A shares and receiving net proceeds of RMB 5.02 billion.³ Note what that means mechanically: this was a primary issuance, so it diluted existing shareholders. It was not free money. But it arrived at a moment when XPeng's alternative sources of capital were expensive and its cash was being consumed.

The part that actually matters is what came after. The equity investment was the headline; the licensing relationship became the economics. By late 2025, reporting indicated Volkswagen would license XPeng's assisted-driving software for its China-market vehicles rather than merely co-develop cars.²³ In early 2026, Volkswagen was named as the first commercial customer for XPeng's VLA 2.0 driving system.²⁴ The relationship had also widened to include XPeng's China Electronic Architecture — the underlying electrical and computing backbone of a modern vehicle — deployed across Volkswagen's China lineup.²⁴

Two things about that are genuinely novel. First, a global incumbent is now shipping a Chinese startup's autonomy stack in its own branded cars. Second, the money is real and visible in XPeng's accounts, which is a rarer thing than it sounds in the world of announced partnerships.

What "VLA" means, in plain language. Older driver-assistance systems were built like flowcharts: detect a lane line, apply a rule, execute a manoeuvre. Engineers wrote thousands of rules, and the systems broke whenever the road did something the rules had not anticipated. VLA — vision-language-action — replaces the flowchart with a single large neural network that takes camera images in and produces driving actions out, with a language-model-like reasoning layer in between. The analogy is the difference between a phrasebook and actually speaking a language: the phrasebook handles the sentences someone wrote down, and nothing else. XPeng claims VLA 2.0 works without pre-built high-definition maps and materially extends the average distance between driver takeovers on complex roads.²⁴ Those are the company's own performance claims, not independently audited ones, and should be read as such.

Where the money shows up. XPeng reports Volkswagen income inside "services and other revenue." That line was RMB 2.67 billion in 2023, RMB 5.04 billion in 2024, and RMB 8.34 billion in 2025 — up 65.6% in the most recent year, driven explicitly, in XPeng's own language, by technical research and development services provided to a car manufacturer.⁵²⁰⁴ In the second quarter of 2026 alone it reached RMB 2.70 billion, up 93.9% year over year, which XPeng attributed to hitting certain key milestones with Volkswagen Group.²

Now do the arithmetic that XPeng does not do for you. In that quarter, total revenue was RMB 19.74 billion at a 20.7% gross margin, implying about RMB 4.09 billion of gross profit. Vehicle sales were RMB 17.05 billion at a 12.1% vehicle margin, implying about RMB 2.06 billion. The remainder — roughly RMB 2.0 billion of gross profit on RMB 2.70 billion of services revenue — implies a gross margin on that line in the region of 75%.²

That single derived number is the most important fact in this article. Services and other revenue was under 14% of XPeng's second-quarter turnover and produced roughly half of its gross profit. XPeng's reported gross margin of 20.7% — the number that looks like a manufacturing success story — is substantially a software-licensing number wearing a car company's clothes.

The caveat belongs here, not in an appendix. A revenue stream that is high-margin, growing fast, and concentrated in a single counterparty is not the same as a moat. It is a contract. The joint-development vehicles have slipped from their originally indicated 2026 timing in some reporting, and milestone-based licensing revenue is by construction lumpy: it recognises when milestones are hit, which means it can be flat or absent in a quarter where none are. Any assessment of XPeng's margin trajectory that does not separate the Volkswagen line from the car business is measuring the wrong thing.

There is also a version of this relationship deepening into something structural rather than contractual. In May 2026, XPeng confirmed at the Financial Times' Future of the Car summit that it was in talks with Volkswagen about acquiring or using a European production site, having outgrown the contract-assembly capacity it uses at Magna's Austrian plant.²⁵²⁶ If that happens, the reversal is complete: the Chinese company would be building cars in a German company's European factory using its own architecture. Note the tell in the same reporting, though — XPeng's own executive described Volkswagen's available European plants as "a little bit old."²⁵ Deals get done or they do not; talks are not tooling.

So what. Volkswagen validated XPeng's technology in the only way that counts, by paying for it, and gave XPeng a high-margin revenue stream that no other Chinese EV challenger has. That is a genuine, differentiated asset. But it is one customer, in one market, on milestone-based terms of undisclosed duration, and it currently flatters a consolidated margin that the vehicle business alone does not support. Which means the real competitive question is not what Volkswagen thinks of XPeng. It is whether XPeng can survive the market it actually sells cars into.

↑ Back to Top

VI. Industry Structure: China's EV Coliseum

In February 2026, China's State Administration for Market Regulation did something that would be almost unthinkable in most developed markets: it banned automakers from selling cars below cost. Not below marginal cost — below fully loaded cost, including administrative, financial and selling overheads. Online car-buying platforms were conscripted as real-time monitors, required to flag abnormally low listings to both consumers and regulators, and violators were warned of significant legal risk.²⁷

Regulators do not outlaw price competition in healthy industries. They do it when an industry has entered what the Chinese press calls 内卷 involution — a self-defeating spiral in which every participant works harder and cuts deeper and nobody earns anything. Chinese passenger-car sales had fallen nearly 20% year on year in January 2026, sector profits dropped 18% in the first quarter, and average industry margins compressed to around 3.2%.²⁷ This is the arena XPeng competes in.

The scale problem, stated bluntly. In 2025, 比亚迪 BYD sold 4,602,436 vehicles.²⁸ XPeng sold 429,445.⁴ BYD holds 27.2% of China's new-energy retail market; the top ten also includes Geely at 12.2%, Changan, SAIC-GM-Wuling, Tesla China, 华为 Huawei's HIMA alliance, Chery, Leapmotor, 赛力斯 Seres and 小米 Xiaomi Auto — and does not include XPeng.²⁹ After tripling its volume in three years, XPeng was still outside China's ten largest NEV sellers by domestic retail volume.

BYD is not merely bigger; it is differently constructed. It makes its own batteries, its own semiconductors, and — critically — its own DM-i plug-in hybrid powertrains. That last point is the substitution threat that most directly damages XPeng. A plug-in hybrid gives a Chinese buyer electric running costs for the daily commute and a petrol engine for the road trip, at a lower sticker price than a comparable battery-electric car, with no charging anxiety. For the price-sensitive middle of the market — exactly where XPeng's MONA products live — DM-i is not a competitor product so much as an alternative answer to the same question, and BYD can price it aggressively because it owns the whole cost stack.

The peer group is not doing better. Among the original "NIO–XPeng–Li Auto" trio of listed challengers, 2025 sorted them decisively. 蔚来 NIO delivered 326,028 vehicles against a 440,000 target. 理想汽车 Li Auto delivered 406,343 against a 640,000 target — a 63.5% completion rate and an actual year-on-year decline, the only shrinking major challenger. XPeng delivered 429,445 against a 350,000 target, a 122.7% completion.²⁸ On the specific test of "did you do what you said you would do," XPeng was the best of the three by a wide margin in 2025.

The newer entrants are the more worrying comparison. Xiaomi Auto sold 411,837 units in 2025, up 200.9%, in its first full year of volume production — a smartphone company that entered the car market and matched a decade-old EV specialist's volume almost immediately.²⁹ Leapmotor sold 596,555.²⁸ Seres, building the AITO range around Huawei's software and retail network, sold 422,920.²⁹ If your thesis is that XPeng's software-first identity is hard to copy, Xiaomi's first year is uncomfortable evidence, and Huawei's model — supplying the whole intelligence stack to a hardware partner — is uncomfortable in a different way, because it competes with the exact thing XPeng sells Volkswagen.

A word on the supply chain, because it decides who survives. The most expensive component in an electric car is the battery, and in China the battery market is dominated by 宁德时代 CATL and by BYD's own in-house cells. For a company like XPeng, that creates an asymmetry it cannot engineer away: BYD buys its batteries from itself at cost, while XPeng buys them from a supplier that also serves its competitors and needs to earn a margin. Every point of battery cost is a point of vehicle margin, which is a large part of why the structural gap between BYD's economics and everybody else's has proven so persistent. XPeng's counter-move has been to vertically integrate the other expensive component — the computing silicon — rather than to fight the battery battle it cannot win.

Porter, applied honestly. Rivalry is the defining force and it is extreme: an industry that has consolidated from several hundred makers to a few dozen serious survivors, and is still shedding. Buyer power is high — the Chinese EV customer is unusually spec-literate, cross-shops relentlessly, and faces almost no switching cost between brands. Supplier power is moderate for batteries, where multiple qualified suppliers compete, but real for advanced computing silicon, which is why XPeng's in-house Turing chip programme matters strategically as much as economically. Substitutes — the plug-in hybrid — are the most underrated threat to a pure battery-electric maker. Entry barriers have been demonstrably lower than incumbents assumed; Xiaomi proved a consumer-electronics brand with capital and distribution can arrive and scale in a year.

7 Powers, applied honestly. XPeng's best claim is counter-positioning: it built a software-defined vehicle from a blank sheet, which incumbents cannot fully copy without cannibalising their supplier relationships and margin structures. That is a real power, and Volkswagen's decision to license rather than build is the strongest external evidence for it. XPeng has some process power in its owned factories at Zhaoqing, Guangzhou and Wuhan, and in the vertical integration of its Turing autonomous-driving chip. Turing packs roughly the AI compute of three Nvidia Orin X processors into a single part, taped out in August 2024 and shipped first in the G7 SUV launched in June 2025 — making XPeng, after NIO, only the second Chinese EV maker to mass-produce its own driving silicon.³⁰ That matters less as a cost saving than as insurance: it removes a single foreign supplier's veto over the company's most important product feature.

What XPeng does not have should be stated just as plainly. It has no scale economies — BYD outsells it more than tenfold and buys components accordingly. It has no network effect; a car does not become more valuable because your neighbour bought one. Its switching costs are weaker than NIO's, whose battery-swap network genuinely locks owners in, and weaker than Huawei's, whose device ecosystem does the same. It has no cornered resource and no branding power comparable to a luxury marque.

So what. XPeng's moat, to the extent one exists, is technical and temporal: a lead in software and in-house silicon that must be re-earned every product cycle, in a market where the regulator has just had to intervene to stop competitors from selling at a loss. That is a very different asset class from a structural moat, and it should be underwritten differently — as a capability that generates returns only while the lead lasts, not as a permanent barrier. The question then becomes whether the current business converts that lead into cash. It is time to open the accounts.

↑ Back to Top

VII. The Business Today: MONA, Margins, and the Profitability Question

On August 27, 2024, at a press conference in Beijing, XPeng launched a compact sedan called the MONA M03 at a starting price of RMB 119,800 — roughly $16,800, and about half what a Tesla Model 3 cost in China. He Xiaopeng told the audience that the "M03" name was a tribute to the Model 3.³¹ It was an unusual thing for a CEO to say out loud, and it told you exactly what the car was for: not to beat Tesla on technology, but to take Tesla's design language down two price bands and sell it to people who could not previously afford any of it.

Within 52 minutes, XPeng had more than 10,000 firm orders with non-refundable deposits.³¹ Within 48 hours, more than 30,000. Buyers skewed young — 55% under thirty — and He personally handed over the first cars at the Chengdu auto show three days later.³²

MONA is the pivot on which the modern XPeng turns, and it is a genuinely double-edged product. It solved the volume problem — the 2024–2025 delivery explosion is largely a MONA story — and in doing so it dragged XPeng's average selling price down by roughly a third, because a company that was selling RMB 300,000 SUVs started selling RMB 120,000 sedans in enormous numbers.

Here is the part that is actually impressive. Average selling price collapsed and vehicle margin still went up — from 8.3% in 2024 to 12.8% in 2025.²⁰⁴ That combination is rare and it is not achievable through pricing tricks. It requires the cost of the car to fall faster than its price, which means real engineering: shared platforms, integrated die-casting, fewer parts, cheaper battery chemistry, in-house silicon replacing bought-in chips, and the plain volume leverage of running factories closer to full. On the evidence, XPeng did the unglamorous cost work that its 2022-era self did not do.

Now the part that is not. In the first half of 2026, that improvement stalled and then reversed. Vehicle margin in the second quarter was 12.1%, down from 14.3% a year earlier.² On the earnings call, management attributed the decline to "the production generation transition" — the cost overhang of switching model generations — and He also cited supply-chain disruption and extreme weather slowing the ramp of the new MONA L03, adding that two-shift production had begun to catch up.³³

Two-shift production is a specific, checkable commitment, which is more than most explanations offer. But the underlying pattern deserves attention rather than reassurance: XPeng has now had one full generation cycle in which volume growth was purchased with vehicle margin, and the company is entering another one. The 2025 margin improvement was real; the claim that it is a permanent, structural feature of the business does not survive the first half of 2026.

The deliveries picture, told honestly. XPeng delivered 62,682 vehicles in the first quarter of 2026, down 33.3% from the same quarter of 2025.³⁴ Management pointed to the industry's traditional slow season, but the more specific driver was policy: China halved its new-energy vehicle purchase-tax exemption from January 1, 2026, which pulled an enormous amount of demand forward into late 2025 and left a hole in the following quarter. The second quarter recovered to 103,295 units — but that was only 0.1% above the same quarter a year earlier.² For the first half as a whole, deliveries fell 15.8% and revenue fell 3.8%.³

Put plainly: after growing 126% in 2025, XPeng's vehicle business stopped growing in 2026. That is the single fact most at odds with the way the company is usually described.

The profitability claim, taken apart. This deserves precision, because the headline and the accounts say different things.

In the fourth quarter of 2025, XPeng reported a net profit of RMB 380 million — genuinely the first in its history. In the same quarter, its GAAP loss from operations was RMB 40 million.⁴ The operating business, in other words, still lost money; the reported profit came from below the operating line — interest and investment income on a large cash pile, and other non-operating items. On a non-GAAP basis, which excludes share-based compensation, XPeng did report a small operating profit of RMB 80 million.¹ That is a real milestone and a very thin one.

The two quarters that followed removed the ambiguity. The first quarter of 2026 produced a net loss of RMB 1.78 billion, against RMB 660 million a year earlier.³⁴ The second produced an operating loss of RMB 1.14 billion and a net loss of RMB 1.34 billion.² Half-year losses of RMB 3.12 billion now exceed the RMB 1.14 billion lost in all of 2025.³⁴

But losses caused by what? This is where a fair analysis has to slow down. Research and development spending in the second quarter reached RMB 2.91 billion, up 32.1% year over year, which XPeng attributed to new vehicle models and AI-related technologies.² Half-year R&D was RMB 5.82 billion — an annualised pace above the RMB 9.49 billion spent in the whole of 2025.³⁴ A meaningful slice of the widening loss is therefore a deliberate reinvestment decision, not an operating collapse. Investors are entitled to disagree about whether that reinvestment is wise; they should not confuse it with the business failing to cover its costs.

The honest formulation is this: XPeng's core vehicle operation is close to breakeven at the gross level but does not yet generate enough gross profit to cover a research budget that management is deliberately expanding. Whether that is investment or dilution of shareholder value depends entirely on what the R&D buys, which is Section IX's problem.

The cash question. XPeng's headline cash position — including short-term investments and time deposits — was RMB 40.48 billion at the end of June 2026, down from RMB 47.66 billion six months earlier.²⁴ Meanwhile total borrowings rose from RMB 12.71 billion to RMB 20.76 billion over the same period, and net cash used in operating activities was RMB 11.72 billion for the half.³ The company remains in a net cash position on the broad measure, and by the standards of loss-making EV makers it is well capitalised. But the direction is unambiguous: the gap between cash and debt narrowed by roughly RMB 15 billion in six months, funded partly by borrowing. This is not distress. It is the beginning of a clock.

Guidance: better than the reputation, worse than the ambition. XPeng's forecasting record has three distinct phases. In 2022–2023 it was poor — culminating in the profitability-target retreat already discussed. In 2025 it was good: the company set a 350,000-unit annual target and delivered 429,445, a 123% completion rate at a time when NIO hit 74% of its target and Li Auto 63%.²⁸ Then in the fourth quarter of 2025 it guided to 125,000–132,000 deliveries and delivered 116,249, missing the bottom of its own range by nearly 9,000 units.⁶

Through 2026 the pattern changed again in an instructive way. XPeng has hit its quarterly guidance — the first quarter came in at 62,682 against a 61,000–66,000 range, the second at 103,295 against 100,000–106,000 — but the guidance itself has been modest, and the third-quarter range of 115,000–121,000 units landed below Wall Street expectations, sending the shares down roughly 7% on the print.⁴²³⁵

And then there is the annual number. XPeng entered 2026 targeting 550,000 to 600,000 deliveries.⁶ With 165,977 in the first half and at most 121,000 guided for the third quarter, the company would need to deliver well over 260,000 vehicles in the final quarter to reach the bottom of that range. On the second-quarter call, He said XPeng expected monthly deliveries "targeting more than 60,000 units" in the fourth quarter³³ — which implies roughly 180,000, and a full-year outcome closer to 470,000 than to 550,000.

That is the credibility ledger in one paragraph: quarterly guidance now met, annual ambition still set well above what the company delivers. The market has adjusted accordingly. Deliveries stopped being the story sometime in 2025; what is priced now is whether the profits are durable.

↑ Back to Top

VIII. Current Management: Ownership, Incentives, and the Credibility Ledger

In March 2025, XPeng's board proposed granting He Xiaopeng 28,506,786 shares at zero purchase price — a package worth about HK$2.55 billion, or $328 million, at the HK$89.55 closing price on the day it was set.³⁶ Headline compensation numbers of that size usually mark the point where an article stops being analytical. This one is worth reading in full, because the structure is the substance.

The shares vest in three equal tranches, and none of them vest on time served. The first third unlocks only if XPeng's Hong Kong shares sustain a 30-day average closing price of HK$250. The second requires HK$500. The final third requires HK$750 — roughly 8.4 times the price at grant. The whole award expires worthless if none of the hurdles are met by March 18, 2035.³⁶

That is an unusually demanding design. The first tranche alone requires the stock to roughly triple. Set against a nominal base salary, it means He Xiaopeng's economics are almost entirely a long-dated call option on the company he controls.

Control, and what it costs shareholders. He holds all 348.7 million Class B shares, which carry ten votes each, plus a small Class A position — about 18.5% of shares outstanding and 69.5% of the voting power.³⁶ The dual-class structure means that on any matter put to a vote, the outcome is known before the meeting. At the annual general meeting held in Guangzhou on June 26, 2026, every resolution passed, including the re-election of independent director Donghao Yang.³⁷

An activist would make two arguments here, and both are fair. First, XPeng's board is majority independent and its committees are independently chaired, but with a 69.5% voting bloc that independence is advisory rather than binding — there is no realistic mechanism by which shareholders could force a change of strategy, a change of capital allocation, or a change of CEO. Second, routine share-issuance mandates that pass without meaningful recorded dissent are a governance formality, not a governance endorsement. Investors in XPeng are, structurally, minority partners in a founder's project.

The counterargument is behavioural rather than structural. He bought Class A shares and ADSs personally after a share-price slump in August 2024 — a founder adding to an already dominant position with his own money rather than selling into strength. On the evidence available in XPeng's public filings and announcements over 2023 to 2026, there is no disclosed programme of insider share pledging or systematic founder selling. That is a bounded observation about disclosed records, not a general assurance.

The credibility ledger, weighed rather than listed. The 2022–2023 period was genuinely bad: an internal target badly missed, a flagship product mispriced in public, and a profitability timeline publicly walked back. The 2025 period was genuinely good: annual target beaten by a wide margin while both listed peers missed theirs. The 2026 period is mixed: quarterly commitments met, an annual target that will almost certainly not be, and a business whose growth stalled.

The fair conclusion is that management's credibility is partially rebuilt and still being tested. The specific thing to watch is not whether XPeng hits a number, but whether it stops setting annual targets it has no path to. Overpromising is a habit, and habits recur under pressure.

One point in management's favour deserves recording alongside that: when XPeng has missed, it has generally said so plainly and given a specific mechanical reason — a production generation transition, a supply-chain disruption, a policy-driven demand shift — rather than blaming the macro environment in the abstract. Those explanations are checkable, which is the property that makes an explanation worth anything. The 2022-era pattern of vague attribution has not recurred in the recent calls.

The president. Wang Fengying's record is discussed in Section IV and remains the strongest concretely evidenced management case at XPeng. One structural point belongs here, though: her equity award vested on service rather than on share price. That is a materially softer design than the CEO's, and it rewards a different thing — staying, rather than creating shareholder value. For an executive brought in precisely to change commercial outcomes, that asymmetry is worth noticing.

The dealmaker. 顾宏地 Brian Gu, Vice Chairman and Co-President, came from J.P. Morgan's Asia-Pacific investment banking business and runs corporate strategy, finance, international expansion and regulatory affairs. He is XPeng's public voice on capital markets and is generally credited with architecting the Volkswagen relationship. That credit is worth flagging as thinly sourced: most accounts of his personal negotiating role trace back to XPeng's own materials rather than to independent reporting. What is independently observable is his conduct on calls. On the second-quarter 2026 call, asked about robotics, Gu said plainly that it was "too early for us to provide a volume prediction guidance," while separately projecting that humanoid robots would eventually carry "much higher gross profit potential compared to the automotive business."³³ Declining to forecast volumes while forecasting margins is a recognisable executive move, and analysts noticed: Goldman Sachs pressed on profitability timing and did not get one.³⁵

The robot that had to be unzipped. On November 6, 2025, XPeng unveiled the next generation of its IRON humanoid robot. Its walk was smooth enough that Chinese social media immediately concluded a person was inside the suit. He Xiaopeng's response was to have a colleague cut open the robot's synthetic skin on stage, exposing the mechanism, the cooling system and the fan noise underneath. Two hashtags about the incident became the top trending topics on Douyin the following day.³⁸

The episode is a genuine data point in both directions, and it is easy to over-read either way. In XPeng's favour: when a claim was doubted, management substantiated it immediately, physically, and in public — which is more than most companies do about anything. Against: the reason the doubt existed at all is that XPeng's demonstrations had, by that point, outrun what audiences believed the company could actually deliver. Proving a robot is a robot is not the same as proving a robot is a business.

So what. The incentive design at the top of XPeng is better than the market average and the governance is worse; those two facts do not cancel out, they simply describe what kind of investment this is. A shareholder is backing a founder with overwhelming control, a personal payoff tied to an eightfold share-price move, and a demonstrated willingness to spend heavily on ideas the market has not yet validated. Which brings us directly to what he is spending it on.

↑ Back to Top

IX. Beyond the Car: Physical AI, Flying Cars, and Robots

Start with the number that anchors this entire section: the combined revenue XPeng has publicly confirmed from flying cars and humanoid robots is not disclosed, because there is no evidence of any. Everything that follows is capital raised, orders claimed and factories built — none of it is shipped product generating recorded sales. That is not a criticism; it is the correct frame for evaluating an option.

The flying car. XPeng's affiliate AeroHT — since rebranded Aridge — has been developing what it calls the Land Aircraft Carrier: a six-wheeled road vehicle that carries a detachable two-seat electric aircraft in its rear compartment, priced under RMB 2 million, or about $280,000.³⁹ The company completed a Guangzhou manufacturing facility designed for more than 10,000 aircraft a year, and reports a global order book of 7,000 units, including a 600-unit block from four Middle Eastern buyers — Ali & Sons Group, Qatar's Almana Group, Kuwait's AlSayer Group and the UAE Chinese Business Council.⁴⁰

Now apply the test that matters. In September 2024, mass production was described as moving to 2026, having previously been targeted for the fourth quarter of 2025.³⁹ By late 2025, mass production and delivery were planned for 2026, with Middle East market entry pushed to 2027.⁴⁰ A factory sized for 10,000 units a year has been standing complete since September 2025 against zero confirmed commercial deliveries.

This is the single most useful precedent in the XPeng story for how to read the company's other frontier claims. AeroHT has raised very large sums, built real hardware, flown public demonstrations, secured named institutional orders, and completed a factory — and has still not converted a single one of those milestones into disclosed revenue, while the delivery date has moved repeatedly. Certification, prototypes and order books are not sales. XPeng's own record on this specific product is the strongest available evidence for how long its frontier programmes take to become businesses.

The humanoid. In August 2026, XPeng's robotics business raised more than $900 million at a post-money valuation above $6.3 billion — led by IDG Capital, with Gaorong Ventures participating and 腾讯 Tencent and Alibaba as strategic investors — which the company described as the largest single private financing round in China's embodied-AI sector.³³ IRON, the robot, carries 76 degrees of freedom in its body and 21 in each hand, and runs on three Turing chips delivering up to 2,250 TOPS of compute.³³ XPeng targets scaled production by the end of 2026 and output in the "several thousand units per month" range during 2027.³³

The strategic logic is not absurd. A humanoid robot and a self-driving car are, at the level of software, close cousins: both need to perceive a messy physical world, predict what happens next, and act. XPeng's VLA models, its in-house silicon and its manufacturing capacity are genuinely reusable. He told analysts that the robotics problem is "at least 20x" more complex than the vehicle problem, which is at least an honest framing of the difficulty.³³

Three pieces of counterevidence belong right here, next to the claim.

First, the use case has already been walked back. XPeng's initial framing pointed at industrial and factory work; the current plan starts with retail and service settings — store greeters and showroom assistants. Management's stated reason, given to Citi on the second-quarter call, is that retail environments best demonstrate "hardware, environmental understanding and emotional value."³⁵ That is a coherent explanation. It is also, functionally, a retreat from the harder and far more valuable application to the easier and more photogenic one.

Second, the people are leaving. Mi Liangchuan, who headed the robotics business, departed in June 2026, after which He Xiaopeng took personal charge of the robotics centre. Shi Xiaoxin, the core product lead who had spent more than four years building the IRON programme from scratch, resigned in the same month.⁴¹ In July, XPeng's head of AI infrastructure, Lu Siyuan, left for OpenAI's embodied-AI effort, with his team reportedly being split across multiple managers rather than replaced by a single successor.⁴² Losing the product lead, the business head and the infrastructure head of a programme within eight weeks of its largest-ever funding round, six months before its stated mass-production deadline, is a material execution signal.

Third, there is no disclosed revenue model. Gu's projection of higher-than-automotive gross margins and subscription-based software upgrades describes an aspiration, not a price list or a customer.

There is also a governance dimension that rarely gets discussed. Financing the frontier units through outside rounds means XPeng shareholders own only part of whatever these programmes become, while the parent company carries the research expense that makes them possible. The robotics unit's headline valuation is also a number set by a private financing round, not by a market — and it is large relative to what public investors have been willing to pay for the entire listed company. Private marks are opinions with money attached; they are not prices, and they do not become prices until somebody sells.

Robotaxi. XPeng began pre-installed mass production of a robotaxi in Guangzhou in 2026, powered by VLA 2.0. He told analysts it had completed more than 2,000 internal test orders and validated the process for trial passenger operations, with the goal of running passenger service without a safety operator in the car "next year."³³ Weigh that against the pattern: He's public timeline for full autonomy has moved more than once since 2024, and "next year" has been the answer at multiple points along the way.

So what. These programmes are correctly understood as options, not businesses. They are financed substantially with outside capital rather than XPeng's own balance sheet, which limits the direct cash cost, though it also creates minority interests in XPeng's most-hyped assets. The claim that they represent real optionality tied to the core software stack is defensible on technical grounds and unproven on commercial ones — and XPeng's own flying-car record is the reason to hold that judgment at "unproven" rather than "promising." The falsifying test is simple and will arrive soon: whether IRON records disclosed external revenue in 2027, and how that compares with the capital committed to it.

↑ Back to Top

X. Bull Case vs. Bear Case

Every investment case eventually reduces to a single question: why does this company earn a return that competitors cannot compete away? For XPeng, both the affirmative and the negative answers are unusually well documented, which makes the argument sharper than most.

The bull case, stated at its strongest. XPeng is the only Chinese EV challenger that has proven a Western incumbent will pay it for technology. That is not a soft validation — it is a recurring, high-margin cash flow from a customer with alternatives, and it is the reason XPeng's consolidated gross margin now sits above 20% while its cars alone earn about 12%. The company designs and ships its own autonomous-driving silicon, which almost no competitor at its scale does, and which removes a supplier chokepoint on its most differentiated feature. Its founder is compensated almost entirely on a share price that must multiply severalfold before he receives anything. It fixed a broken commercial organisation by hiring the specific competence it lacked, and the fix is visible in three consecutive years of vehicle-margin improvement. And it is diversifying away from the worst price war in the industry's history: overseas revenue reached RMB 8.23 billion in the first half of 2026, about 24% of the total, against RMB 5.07 billion a year earlier, with export prices averaging above EUR 40,000 — meaningfully richer than what XPeng gets at home.³³³

That last point deserves emphasis because it is underappreciated. XPeng's international business is not a volume story; it is a mix story. Selling cars in Europe at forty thousand euros while selling them in China at a fraction of that is the single cleanest path to a structurally better margin that does not depend on Volkswagen.

The bear case, stated at its strongest. The profitability milestone was thin, non-operating in character, and did not repeat. The core vehicle margin is falling, not rising. Roughly half the company's gross profit comes from a services line dominated by one customer on milestone-based terms. Growth in the vehicle business went to zero in the first half of 2026 while the industry's regulator was busy outlawing loss-making price competition — meaning the competitive pressure that produced that stall is unlikely to abate quickly. Borrowings rose sharply in six months while operating cash burn ran to nearly RMB 12 billion. And research spending is accelerating into two frontier programmes, one of which has a documented multi-year record of building factories and order books without generating disclosed revenue, and the other of which lost its business head, its product lead and its AI infrastructure head within eight weeks of raising nearly a billion dollars.

Myth versus reality. Three consensus descriptions of XPeng deserve correcting, because each is repeated often enough to have become background assumption.

Myth: XPeng turned profitable at the end of 2025. Reality: it reported a net profit for one quarter while its operating line was still slightly negative, and both of the quarters that followed were losses. The milestone was real and it was not a trend.

Myth: XPeng's margin improvement proves it has cracked the cost problem in mass-market EVs. Reality: it partly proves that, and partly reflects a high-margin licensing line that has nothing to do with building cars. The vehicle-only margin — the number that tests the cost claim — fell year over year in the most recent quarter.

Myth: XPeng is the fastest-growing Chinese EV challenger. Reality: it was, spectacularly, in 2025. In the first half of 2026 its deliveries fell 15.8%. Growth rates in this industry are policy-sensitive and model-cycle-sensitive to a degree that makes any single year a poor guide to the next.

The activist's version. A skeptical fund would put it more bluntly. It would argue that XPeng's headline gross margin is not a manufacturing achievement but a licensing contract in disguise, and would ask the company to disclose Volkswagen-related revenue separately so investors can see the car business unclothed. It would ask why the robotics and flying-car units are financed through outside rounds that create minority interests in the assets the market is most excited about, while the parent absorbs an R&D line growing 32% year over year. It would note that a 69.5% voting bloc means none of these questions can be forced. And it would point out that setting a 550,000-to-600,000-unit annual target and then guiding quarterly to numbers that cannot reach it is a pattern, not an accident.

None of that constitutes a fraud allegation, and none has been made. XPeng has not been the subject of an activist campaign or a governance-focused short report on the public record through mid-2026; the bearish positioning that has existed has been a valuation and momentum view rather than an accusation about the accounts.

Why win. XPeng's honest answer is that it monetises the same technology stack twice — once as a differentiated feature in its own cars, and once as licensed software sold to companies that cannot build it. If both channels work, XPeng earns software-like margins on a manufacturing base, which is exactly the economics every automaker in the world has been trying to reach for a decade.

Why not. The vehicle channel is being fought in a market where the average industry margin is around 3% and the regulator has had to ban below-cost selling. The licensing channel is one customer deep. And the company is choosing, right now, to spend the proceeds of both on a third channel that has not yet sold anything.

The calibrated conclusion. On the evidence available, three of XPeng's four material thesis claims survive in narrowed form. The turnaround claim survives, restricted to the specific commercial and cost failures it addressed, and untested through a downturn. The technology-differentiation claim survives, evidenced by Volkswagen's payments and by in-house silicon actually shipping — but as a time-limited technical lead rather than a structural moat, since it has no scale, network or switching-cost support. The management-incentive claim survives essentially intact, with the caveat that alignment and governance are different things. The claim that XPeng has crossed into durable profitability does not survive; the record rejects it in its current form, and the burden of proof sits squarely with the next few quarters.

↑ Back to Top

XI. Risk Radar

The price war has not ended; it has changed shape. China's below-cost sales ban addressed the mechanism but not the underlying overcapacity, and discounting has migrated into forms that are harder to police — zero-interest financing stretched over five to seven years, inflated trade-in valuations, and driver-assistance software packages worth thousands of yuan bundled in as standard.²⁷ That last method is particularly awkward for XPeng, because its core differentiation is precisely the software that rivals are now giving away to move metal. The mechanism by which this hurts is visible in the accounts already: vehicle margin down year over year while volumes were flat.

Concentration in the profit pool. The risk here is not that Volkswagen walks away; it is subtler. Milestone-based technical-services revenue is recognised when milestones are achieved, which means the line can be flat or absent in a quarter where the joint programmes are between deliverables. Because that line carries an implied gross margin far above the vehicle business, a pause would hit reported profitability disproportionately, with no change whatsoever in the underlying car operation. Investors reading XPeng's consolidated gross margin as a proxy for manufacturing progress will be misled in both directions.

Tariffs and the European build-out. The European Union imposed definitive countervailing duties on China-built battery-electric vehicles on October 31, 2024, on top of the standard 10% import tariff — ranging from an additional 17% for BYD to 35.3% for SAIC, with company-specific rates depending on cooperation with the anti-subsidy investigation.⁴³ XPeng's response has been to assemble G6 and G9 models from semi-knocked-down kits at Magna's Austrian plant, which shifts final assembly inside the tariff wall.²⁶ That is a genuine mitigation, but it is contract capacity that XPeng has already outgrown, and the replacement — a European plant of its own — remains a negotiation rather than an asset. Meanwhile the Commission has been preparing to extend duties to plug-in hybrids, closing a gap that currently favours Chinese PHEV exporters.⁴³ European localisation is the right strategy and it is not yet done.

Capital allocation, with a precedent. The concern is not that XPeng invests in frontier technology; it is that it is doing so while its core business burns cash and its borrowings climb. The relevant historical check is AeroHT: a programme that consumed large amounts of capital, hit real technical milestones, and has still not produced disclosed revenue on a timeline that has slipped repeatedly. If the robotics programme follows the same conversion rate, the cash cost will be larger and the distraction cost larger still — and the recent departure of its senior technical and product leadership makes that scenario more plausible, not less.

Regulatory scrutiny of autonomy claims. China has been tightening the rules on how driver-assistance capabilities may be marketed following industry-wide safety incidents, and export markets are moving in the same direction. XPeng's entire brand equity is built on assisted-driving performance, and its robotaxi programme depends on regulators granting permission to remove the safety operator. A restriction on marketing language would damage the domestic premium; a delay in driverless approval would push the robotaxi business case out by years. Neither is speculative — both are live regulatory processes.

Data, and the export problem nobody prices. A modern XPeng is a rolling sensor array: cameras, radar, and a driving model that improves by ingesting road data. That is the engine of the company's technical lead, and it is also a regulatory exposure that grows with every European sale. Chinese automakers exporting to the EU operate under data-protection rules written for a world in which cars did not continuously record public streets, and under a political climate in which the provenance of that data is a live question. XPeng has not disclosed a material incident on this front. The risk is not an event that has happened; it is that the same capability which differentiates the product is the one most likely to attract restriction in the markets where the product earns its best prices.

Financing costs, and the shape of the next raise. XPeng funds itself today from a large cash balance, growing borrowings, and outside capital raised into its subsidiaries. That third channel is the interesting one. Financing the robotics and aviation units through private rounds keeps their burn off the parent's cash flow, but it also means that if those units succeed, XPeng shareholders own a diluted share of the upside, and if they need more capital, the parent faces a choice between further dilution at the subsidiary level or funding them itself. This is not a problem today. It becomes one the first time a frontier unit needs money in a market that has stopped paying for embodied-AI stories.

A quieter one: key-person concentration. He Xiaopeng now personally runs the robotics centre in addition to chairing and running the company. Founder-led focus is an asset until it becomes a single point of failure across three simultaneous programmes.

↑ Back to Top

XII. Durable Lessons and What to Watch

Lesson one: a product failure that triggers organisational change is recoverable; one that triggers only an apology is not. XPeng's G9 relaunch inside 48 hours was admirable but insufficient on its own. What actually changed the company's trajectory was hiring a president with the commercial competence the founder lacked and giving her real authority over product, portfolio and sales. The relevant investing signal was never the apology. It was the org chart.

Lesson two: R&D built during loss-making years can be sold, and selling it changes the economics of the whole enterprise. XPeng spent years capitalising nothing and expensing everything into losses, and then found a buyer for the output at a gross margin roughly six times what it earns on cars. That is a genuinely underused playbook for capital-intensive technology companies. The caveat travels with the lesson: a licensing contract is not a moat, and a licensing customer is not a distribution channel.

Lesson three, and the most portable one: check the operating line before you believe a first profitable quarter. XPeng's inaugural profit was reported at the net line while operations were still slightly negative, and it did not repeat. This is not unusual and it is not deceptive — the company disclosed both numbers clearly. It is simply that the headline and the economics measured different things, and only one of them was durable.

Lesson four: watch what management stops saying. XPeng's narrative has shifted three times in four years — from "scale into profit" in 2021–2022, to "cost discipline and volume recovery" in 2023–2025, to "physical AI company" in 2026. Each shift was explicable and each was accompanied by real operational change. But the shifts also had the effect of moving the goalposts before the previous target was reached, and that is a pattern worth tracking independently of whether any single strategy is sound. The most informative sentence on any earnings call is often the one management has quietly stopped repeating.

The KPIs that matter. Three, and only three.

Vehicle gross margin. This is the cleanest available read on whether the MONA-led volume strategy works economically at scale, and it must be tracked separately from consolidated gross margin, which the Volkswagen line flatters. A second consecutive year of decline would meaningfully damage the cost-engineering claim.

Consecutive quarters of operating profit. Not net profit — operating profit, which is the bar the fourth quarter of 2025 did not clear. One quarter is an event; two is the beginning of evidence.

Disclosed revenue from robotics and flying cars, measured against capital committed. This is the test that separates real optionality from a repeat of the AeroHT pattern. The number to watch is not units produced or orders claimed; it is revenue recognised.

What would change the case. Two consecutive operating-profitable quarters with vehicle margin rising rather than falling would materially strengthen the constructive view. A second consecutive year of falling vehicle margin, or a visible pause in the Volkswagen milestone revenue, would materially strengthen the skeptical one. Both outcomes are observable within four quarters, which is unusually fast for a thesis this contested.

↑ Back to Top

XIII. Epilogue

As of the most recent reporting, XPeng occupies an odd position in the Chinese EV hierarchy: the challenger that grew fastest and then stopped, running the best margin trajectory of the loss-making trio while posting its widest half-year loss since 2023. It sells cars in sixty countries from 740 stores across 257 Chinese cities and 380 international outlets, operates 3,780 charging stations, and generates almost a quarter of its revenue outside China — up from a negligible share three years ago.³⁶ By the standards of 2022, when the company could not price its own flagship, that is a transformation.

The through-line of this story is that XPeng has now bet the company twice. The first bet was defensive and it worked: after the G9 failure, He Xiaopeng handed the commercial engine of his company to an outsider from a rival firm, bought a low-cost platform from a ride-hailing company rather than building one, and sold his technology to a German incumbent. Those three decisions, taken within about eighteen months of each other, are why XPeng exists in its current form.

The second bet is offensive, and it is being placed right now. It says that the software stack XPeng built to make cars drive themselves is the foundation of a broader business in machines that move through the physical world — robotaxis, humanoids, aircraft. If that is right, the current losses will look like the cost of entry to a category. If it is wrong, they will look like a company that got very good at making cars and then spent the proceeds on something else.

There is a version of the future in which neither reading is quite right, and it is probably the most likely one. In it, XPeng remains a mid-sized manufacturer in a consolidating market — profitable in good quarters, not in bad ones — while the licensing business quietly becomes the most valuable thing it owns, because selling autonomy software to companies that cannot write it requires no factories, no dealers and no price war. That outcome would be less dramatic than either the robot future or the collapse, and it would depend on something the company has done exactly once: persuading a second global automaker to buy what Volkswagen bought. Nothing in the public record suggests that has happened yet.

What makes XPeng genuinely interesting as a case study is that both readings are supported by its own history. The company has a demonstrated ability to fix things it has broken, and a demonstrated inability to convert technical milestones into revenue on the schedule it announces. Investors do not get to pick which pattern repeats. They get to watch the vehicle margin, count the operating-profitable quarters, and wait for the first robot invoice.

↑ Back to Top

XIV. Outro

That is the story of XPeng — a browser company's founder, a flagship launch that nearly ended it, a president who rebuilt the commercial machine, a German giant that bought the software, and a company now spending its recovery on the next bet. Whether the second bet pays is not yet knowable from the evidence available.

What is knowable is exactly which numbers will tell you, and roughly when they will arrive.

XPeng Inc.

Table of Contents

XPeng Inc.: Betting the Company Twice

I. Introduction & Episode Roadmap

II. Origins: From Browsers to Bodywork (2004–2020)

III. The Fall: The G9 Launch Disaster and the 2022 Reckoning

IV. Rock Bottom to Reorg: Wang Fengying (王凤英) and the Turnaround

V. The Volkswagen Alliance: A Reversed Technology Flow

VI. Industry Structure: China's EV Coliseum

VII. The Business Today: MONA, Margins, and the Profitability Question

VIII. Current Management: Ownership, Incentives, and the Credibility Ledger

IX. Beyond the Car: Physical AI, Flying Cars, and Robots

X. Bull Case vs. Bear Case

XI. Risk Radar

XII. Durable Lessons and What to Watch

XIII. Epilogue

XIV. Outro

References