Why XPeng's Stock Is Lagging Despite the News

XPeng has had, by any product measure, an extraordinary few months. It launched three SUVs since May. It began building humanoid robots on an automated line. It named its robotaxi business and opened it to the public. Its flying car is weeks from the delivery date it has been promising for two years.

Its stock is at a two-year low.

XPEV closed at $9.58 on 7 October, down 52.8% year to date. The Hong Kong line (9868) is down 52.7% and closed at HK$37.50 — the lowest since September 2024. The Hang Seng is down 6.2% over the same stretch, and XPeng is the worst performer in its peer group: NIO and Li Auto are both down about 33%, BYD (Hong Kong) about 22%.

Look further back and the move is starker still. XPeng peaked at $28.24 in the week after its AI Day in November 2025, briefly worth more than Geely Auto. It has given up roughly two-thirds of that value in eleven months — and the AI Day is precisely where the robot and robotaxi story was launched.

This is not a market that is falling; it is a stock that is being repriced.

The question of why is worth taking seriously, because the usual answers — "the market is irrational", "short-termism", "the robots aren't understood" — do not survive contact with the numbers. Here is what the numbers say.

The catalogue the bull case is built on is real

It is worth stating plainly, because none of the following is in dispute:

• MONA L03: launched 16 July at ¥123,800–¥156,800 across nine variants, having quietly landed ¥20,000 below its pre-sale price. More than 14,000 deliveries in September (the IR release gives a floor of "over 10,000"), on sale in Hong Kong, the Philippines, Australia and Indonesia, with over 90% of orders taken on the Max or Ultra SE trims that carry the assisted-driving stack.
• GX: launched 20 May with 24,863 firm, non-cancellable orders in the first 12 hours and more than 80% of early orders on the top Ultra trim. 7,224 deliveries in September, a third consecutive month above 7,000, and a second consecutive month in China's top three for large SUVs above ¥300,000. More than 3,000 orders in the Middle East.
• G9L: China launch 17 September from a limited-time ¥231,800 — ¥28,000 below its own pre-sale price, on a car positioned between the GX and the G9. First right-hand-drive units were built on 11 September for Australia; the global launch is in Paris on 12 October across a stated 64 markets.
• MONA L05: filed with China's MIIT on 9 May — a 4,870 mm mid-size crossover, BEV and range-extended. He Xiaopeng confirmed on the August earnings call that it launches in China in Q4 2026. No price has been announced.
• IRON: production line commissioned in Guangzhou in September; the first unit completed automated assembly and walked itself off the line. Over 80% of core processes automated, three Turing chips at 2,250 TOPS, and a stated capacity target above 1,000 units a month by year-end. Mass production by end-2026 for XPeng's own stores and campuses; external sales in 2027.
• Robotaxi: branded XPENG YOYO and opened to public registration on 8 October 2026 — four Turing chips, ~3,000 TOPS, second-generation VLA, no HD maps, no lidar, over 2,000 internal test orders completed in Guangzhou. Notably, XPeng will not operate fleets itself: it is positioning as a technology and hardware supplier, with Amap as its first ecosystem partner and revenue from hardware, technology service fees and operating commissions. That is asset-light, and it also means XPeng does not capture the fleet economics.
• Flying car: AeroHT's Land Aircraft Carrier — a ground vehicle carrying a detachable two-seat aircraft — is still guided for mass production and delivery "within 2026", with more than 7,000 orders at under ¥2 million and a factory designed for 10,000 units a year. No customer delivery had been confirmed as of 8 October.

Any one of those would be a good year for most carmakers. The pile of them is why the disconnect feels wrong.

The financials say something different

Here is the same company's P&L.

Period	Deliveries	Revenue	Net loss
Q4 2025	—	—	+¥0.38bn profit
Q1 2026	62,682	¥13.03bn	−¥1.78bn
Q2 2026	103,295	¥19.74bn	−¥1.34bn
H1 2026	165,977 (−15.8% YoY)	¥32.78bn (−3.8%)	−¥3.12bn (vs −¥1.14bn)

XPeng reported its first profitable quarter in Q4 2025. It has not repeated it. The H1 net loss is roughly 2.7× the prior year's, while revenue went backwards.

The reason is visible in the segment split. Group gross margin was a healthy-looking 20.7% in Q2 — but vehicle margin was 12.1%, down from 14.3% a year earlier, and it was services and others that carried the number: ¥2.70bn of revenue (13.7% of the total) at a 75.1% gross margin, contributing close to half of group gross profit. Most of that is technical R&D services to another carmaker — the Volkswagen agreement — where revenue lands on development milestones and, later, on vehicle volumes.

That is the single most important line in the accounts. XPeng is currently a company whose reported profitability comes from selling technology, attached to a car business earning 12.1%. When the milestone timing moves, the group margin moves with it — and milestones are, by nature, lumpy.

Meanwhile the cost base is being rebuilt in public. H1 R&D rose 39% to ¥5.82bn, across VLA, Turing silicon, Robotaxi operations and IRON. Operating cash flow swung from an inflow of RMB 7.64bn to an outflow of RMB 11.72bn in the same half, and cash fell to RMB 40.48bn from RMB 47.66bn at year-end.

You can fund that from a licensing stream and a strong balance sheet for a while. You cannot do it forever while the car business runs at a loss.

The guidance miss is what broke the spell

Product announcements are narrative. Guidance is a promise with a date on it.

On 24 August, XPeng guided Q3 to 115,000–121,000 deliveries and ¥21.7–23.4bn of revenue. Consensus was ~147,000 units and ¥26.6–27.3bn — the volume guidance midpoint was roughly 20% below what analysts had modelled, and the revenue midpoint about 15% below. The shares fell 8.5% in New York that day and 9.5% in Hong Kong the next — the biggest one-day fall in Hong Kong since November 2025 — on the same day the company announced its robotics unit had raised over $900m.

The delivery guidance, notably, was fine: XPeng delivered 118,390 in Q3, inside the range and a record. Analysts were not worried about volume. They were worried about revenue per vehicle — which is to say, about price and mix.

That distinction matters. On 1 October XPeng reported 41,256 deliveries for September — its best month of 2026 — and the shares fell 1.5%. Nine-month deliveries of 284,367 remain 9.2% below last year, and the comparison is unforgiving: 2025 was a record year, with deliveries up 126% to 429,445. The market spent 2025 pricing a company growing at triple digits and is now pricing one that is shrinking.

And the demand was there. The L03 took 46,859 orders in its first hour — against roughly 10,000 for the M03 — and Deutsche Bank's channel checks put firm, non-cancellable L03 orders above 50,000. XPeng's own commentary referred to record locked orders in July and August, up more than 50% quarter on quarter. That is the strongest order book in the company's history, sitting alongside a delivery number that went backwards.

Which is the real diagnosis, and it is more specific than "demand is weak": this was a supply problem and a mix problem, not a demand problem. The L03 was constrained by Turing chip supply through its ramp; the GX was constrained by its own popularity. XPeng could not build fast enough, and what it could build skewed cheap. That distinction matters for what happens next — supply can be fixed, and mix can be improved — but it does not change the fact that the reported numbers got worse.

Three structural problems

1. The new models are in the right segments and the wrong price bands

Look at where the volume is coming from — and at what price.

• The L03 launched at ¥123,800, having landed ¥20,000 below its own pre-sale price.
• The G9L launched from ¥231,800, ¥28,000 below its pre-sale price.
• The GX, the one genuine margin carrier, launched from ¥269,800 on a limited-time basis — against a pre-sale price of ¥399,800 a month earlier.

That pattern is the margin story in three lines. XPeng is launching new generations of products and then pricing them sharply below where it had signalled, to win volume. The GX is the exception that proves the rule: it is the model holding its price best, it is capacity-constrained by Turing chip supply, and it is the reason group vehicle mix has not collapsed further.

The L05, when it arrives, is expected by Chinese media to start near ¥150,000 — though no official price exists. That would put a third of XPeng's current product offensive below ¥160,000.

That is a rational strategy for a company that needs volume, data and manufacturing scale. It is also a strategy that puts XPeng's growth directly into China's most brutally contested price band, in a year when the average selling price of a Chinese NEV has been under pressure from both directions. Q2's average selling price was RMB 165,000, down 5.9% quarter on quarter.

2. China's EV market is contracting, not just consolidating

This is the context that gets lost outside China. China's NEV retail is down roughly 13% year to date, with eight consecutive months of year-on-year declines. Total passenger-car retail fell 29% year on year in the first 27 days of September — the steepest month of 2026. The NEV purchase-tax exemption halved on 1 January, pulling demand into December 2025 and gutting the 2026 comparison.

Industry profitability is at a decade low — 2.4% in July. The cause is not only price war: lithium carbonate roughly doubled to over ¥200,000/tonne, and automotive-grade memory rose 150–300% as AI datacentres outbid carmakers for fab capacity. Fifteen-plus NEV brands have raised prices or cut incentives in 2026, because they have no choice.

XPeng is competing in that market as a premium-leaning, technology-led brand, which is the hardest position to hold when the market's cheapest models are winning. Six of China's top ten models are now priced below ¥150,000.

Meanwhile the relative position has slipped. In September, Leapmotor delivered 105,656 vehicles to XPeng's 41,256 — 2.56× the volume. Zeekr's nine-month total (288,404) has already overtaken XPeng's (284,367), growing at triple digits while XPeng shrinks. Xiaomi hit a 2026 high above 40,000 — its YU7 SUV and new range-extended series land squarely in the ¥200,000–300,000 band where the G9L and GX compete.

3. The market stopped paying for pre-revenue physical AI — this year

This is the part that is hardest to accept, because XPeng's robots and robotaxis are genuinely further along than most. But 2026 has been a "show me" year for exactly this theme, and the evidence is unambiguous:

• Unitree, the sector's flagship, IPO'd in Shanghai in August at a 219× P/E, opened 629% up, and has since fallen ~55% from its peak — still at roughly 332× earnings. Its own filings show only 9% of humanoid revenue is industrial, and 50–70% of that is corporate guided tours.
• Pony.ai's Hong Kong shares trade below half their IPO offer price. Its CFO has said the company needs 40,000–50,000 robotaxis deployed for positive free cash flow. Its end-2026 target is 3,500.
• WeRide is ~40% below its issue price; its CFO targets full-year breakeven in 2029.
• EHang — the only company globally with a full Chinese eVTOL certification stack — is down 62% this year and in August withdrew its FY2026 revenue guidance with no replacement.
• MSCI deleted Pony AI from the MSCI China Index in August, six months after adding it, while adding AI and semiconductor names.

The market's test has shifted from technical leadership to demonstrated earnings. That is a bad regime for a company asking to be valued on robots, and it is not unique to XPeng. Tesla is the only listed carmaker still receiving a robotics premium — and it is a US mega-cap with an index-driven bid.

The thing that actually explains the multiple

XPeng's market capitalisation on 7 October was $9.17bn.

Its robotics unit was marked, in August, at a $6.3bn post-money valuation — roughly two-thirds of the entire company.

Even setting aside how much of that round came from external investors rather than XPeng and affiliated entities, the arithmetic is the point: there is very little room left in the share price for the robot story to be the upside. If you credit the whole $6.3bn mark, you are valuing XPeng's cars, its charging network, its 64-market rollout, its VLA stack and its Turing silicon at a few billion dollars — and the market has decided that is about right, because the car business is loss-making and the robot business has no disclosed revenue.

Citi did that sum explicitly in August and concluded that, with the robotics valuation fully reflected, the car business was worth about $6.5bn — roughly 0.5× its 2026 estimated sales. UBS, the same month, put XPeng at 0.9× forward sales against Li Auto at 0.7× and NIO at 0.5× — the two measures are not directly comparable, but both point the same way: even after a 53% decline, XPeng has not become cheap relative to its peers. The multiple has compressed because the earnings went away, not because the stock reached a discount.

Meanwhile shareholders fund the moonshots through the P&L. The robotics round went into a subsidiary — XPeng retains a majority and consolidates it, which means XPEV holders capture most of the upside and all of the R&D line. But the claim is thinner than the headline suggests: XPeng's stake in the robotics business fell from 100% to about 82% at the close, and to roughly 68% if the attached warrants and employee incentive plan are fully exercised. The incoming investors hold preferred stock with redemption rights if there is no qualifying IPO within seven years, plus liquidation preference and anti-dilution protection. And the robotics business is not yet a profit centre: it lost RMB 369m in FY2025 and had net liabilities of roughly RMB 447m at 31 March 2026, so its cash burn still lands in XPeng's consolidated accounts.

AeroHT sits inside the group too. So the "free option" argument cuts both ways: holders are paying for the optionality in cash burn, but the market is not yet willing to capitalise three separate pre-revenue businesses inside one loss-making car company — and roughly a third of the robotics upside now belongs to someone else.

What analysts did, and what it tells you

Every major house cut its target in the six weeks after the August print:

Date	Firm	Action	Target
24 Aug	Barclays	Underweight	$15 → $14
Aug	Macquarie	Outperform	→ $18
Aug	Bernstein	Market-perform	→ $18
Aug	Jefferies	Buy	HK$98.30 → HK$75.70
15 Sep	Morgan Stanley	Overweight kept	HK$96 → HK$70
29 Sep	JPMorgan	Overweight → Neutral	$24 → $11.50

The consensus target is still around $18 against a $9.58 price — over 80% implied upside. Treat that number with suspicion. When the average target is double the spot price and the price is right and the targets are falling every few weeks, the targets are the lagging indicator, not a floor. JPMorgan's cut cited weak domestic demand, higher input costs, policy uncertainty and tariffs — none of which are XPeng-specific, and all of which are structural.

Barclays put the bull case's real constraint in one sentence in August: before the first batches of IRONs leave the line, "it is hard to include them in our current financial build for FY27. And the stock will continue to be dictated by vehicle deliveries until IRON sales reach a material amount."

So why is the stock lagging?

Because it is being valued on the part of the company that is losing money, not the part that is in the news. Four things have to be true at once for the story and the price to converge, and only one of them is about robots:

1. Volume and mix have to carry the car business. He Xiaopeng has targeted over 60,000 deliveries a month in Q4 — roughly 45% above September. The L05 in Q4 will help volume but not mix.
2. Vehicle margin has to recover. 12.1% is not enough to fund ¥5.8bn of half-year R&D. The GX mix is the lever, and it is chip-constrained.
3. The services line has to normalise without a cliff. Licensing revenue flattered a 20.7% group margin. The market wants to see the car business stand on its own.
4. Robots have to become countable. Not demonstrated — sold, with revenue and a customer. XPeng's own plan has external IRON sales starting only in 2027.

None of these is a reason to think the products are bad. All of them are reasons why the market is not paying for the products yet.

What would change it

The next hard data point is Q3 earnings in mid-to-late November — market calendars estimate 16 November, though XPeng has not confirmed a date. Watch three lines specifically: vehicle margin, the services contribution to gross profit, and whether Q4 guidance keeps the 60,000-a-month target intact. A quarter where the car business produces the gross profit would do more for the stock than any robot demo.

Beyond that: disclosed external IRON orders with pricing, a vehicle-margin recovery above 13–14%, and evidence that the L05 can lift volume without further ASP dilution.

There is one more thing worth noting, because it is the quietest signal in the data. XPEV's RSI is around 23 — deeply oversold. Call volume has been running at roughly 1.8× normal with implied volatility up. Short interest, meanwhile, is only moderately elevated at around 6% of the float, having risen about 28% during 2026 — so this is not a crowded short that a single headline can squeeze. Somebody is positioning for a rebound. That is not evidence the thesis has changed. It is evidence that at $9.58, the market has priced a great deal of bad news already.

It is also worth saying what is not driving the decline, because it comes up constantly. There is no evidence of insider selling — He Xiaopeng's only 2026 filing of ownership showed no shares traded — and no XPeng-level equity raise, convertible note or ATM programme this year. The dilution that has occurred is inside the robotics subsidiary, not at the listed company.

The honest summary is this. XPeng has built the products, the silicon and the robots. What it has not yet built is a car business that makes money on its own — and until it does, the market will keep valuing it like the car company it still is, however good the other story gets.

Caveats

• The $6.3bn robotics valuation is a private mark, not a market price. The round was the largest single private financing in China's embodied-AI sector, led by IDG Capital with Tencent and Alibaba participating, but press reporting of its composition indicates only about $600m was genuinely external, with the balance from XPeng's own subsidiary and entities controlled by He Xiaopeng and Brian Gu. We have not been able to verify that split against a primary filing, so treat the headline and the split with corresponding caution. The correct number to watch is the external pre-money, not the post-money headline.
• XPeng's vehicle margin for H1 2025 is reported inconsistently. The interim report gives 12.6%; trade coverage cites 14.3% for the Q2-on-Q2 comparison. This article uses the Q2 figure when discussing Q2, and both are correct for their respective periods.
• The robotics and AeroHT ownership structures are not fully transparent in public sources. XPeng confirms it retains a majority of the robotics business and consolidates it. AeroHT is a group subsidiary. The exact economic interest XPeng holds in each has not been established here.
• Source quality varies. Several 2026 figures on delivery rankings and industry margin come from trade compilations (CnEVPost, CPCA, Gasgoo) rather than company releases. Where a figure is company-sourced — the buyer profile, the order counts, the "record firm orders" claim — this article says so.
• Nothing here is a forecast. The next reporting date will settle several of these questions in either direction.

---

Sources: XPeng Q2 2026 results and September/Q3 delivery releases; CnEVPost; CPCA; Form 6-K interim report; Reuters; CNBC; Barclays, JPMorgan, Morgan Stanley, UBS and Jefferies notes as reported; Caixin; company filings and press releases. Prices as of 7–8 October 2026 via Xueqiu. Nothing here is investment advice.