There is an easy question in this morning's Chinese car data and a hard one, and from a distance they look the same.
The easy question is how big the export boom is. It answers itself. The association publishes the number, every wire prints it within the hour, and the only thing left to argue about is which denominator makes the share look most dramatic. Answer it and you are holding a national statistic and no position in anything.
The hard question is whose boom it is. Nobody owns a country's exports. What an investor owns is a company, and the distance between a market moving 78% and a company moving with it is where the entire result lives. So the question this piece asks is narrower and more awkward: does the surge show up inside the two Chinese carmakers this site covers — and specifically, does it move the one driver our XPeng upside is written on?
Answering that needs a company-level export number, monthly. Neither company publishes one on a monthly schedule. Getting it is the work, and it turns out to point the other way.
The month, as the association reported it
The China Passenger Car Association published August this morning. Passenger-car retail was 1.541 million units, down 23.6% year over year and up 5.5% on July, with the year to date at 11.716 million, down 20.8%. Passenger-car exports were 888,000, up 77.8%. Manufacturers wholesaled 2.369 million units, down 5%.
New-energy vehicles were 65.2% of domestic retail, a record, and 518,000 of the exports — up 154.7%, taking the new-energy share of everything China ships abroad from 40.9% a year ago to 58.4%.
The export share is 37.5%, and the denominator tells you what you are measuring
Two shares are circulating. Both are computable, and the difference between them is not a mistake.
Wholesale is the only total the association publishes that contains exports at all, so it is the consistent base. 888,000 of 2.369 million is 37.5%. A year earlier, exports of 499,000 sat on a wholesale base of about 2.49 million — derived, because the association reports the year-on-year change as a rounded −5% rather than the level, which puts the base between 2.48 and 2.51 million. That gives 20.0%, and the rounding does not threaten it: the range runs 19.9% to 20.1%.
Reuters counts the month at 894,000 exports rather than the association's 888,000. On the same wholesale base that is 37.7%, which rounds to the 38% now in circulation. So the figure being repeated is a rounding of a slightly different export count on the right denominator — not, as it first looks, a basis error.
Divide exports by retail plus exports instead and August gives 36.6% against 20.0% a year earlier. That is also a real ratio, and it is lower for a reason worth knowing. Retail plus exports comes to 2.429 million against wholesale of 2.369 million, because the two are counted at different moments: domestic wholesale was 1.481 million against retail of 1.541 million, so Chinese dealers sold about 60,000 more cars last month than manufacturers shipped them. Inventory came down, which flatters retail and shrinks the retail-based share below the flow that actually left the factories.
Either way the shift is roughly seventeen points in a year, and either way the same caution applies. Retail counts a car when a customer takes it. An export counts when it leaves the factory. A car on a ship to Rotterdam is an export today and may be a sale in November or a discount in January. Any argument treating 888,000 as evidence of demand is treating a shipment number as a sales number.
An exported car is not a domestic car sold somewhere else
Three things change when a Chinese maker sends a car abroad, and they do not all cut the same way.
- Price roughly doubles. XPeng's average export selling price is above EUR 40,000, He Xiaopeng said on the June-quarter call. Group revenue per vehicle that quarter was RMB 165.0 thousand. Across any euro rate of the last two years that puts an exported XPeng at roughly 1.8 to 2 times the revenue of the average car it sells at home.
- Duty takes a bite out of that. A China-built battery EV entering the EU carries the standard 10% car tariff plus a countervailing duty, 20.7% for XPeng's cooperating-exporter band — 30.7% landed. A minimum-import-price mechanism intended to replace the duty reached framework agreement in January 2026, and the Commission published its guidance for price-undertaking offers on 12 January; it is not yet the regime XPeng ships under.
- Cash arrives later. An export is recognised when it ships and collected when it clears the far end, with a hull and a port in between. XPeng consumed RMB 11.72 billion of operating cash in the first half against RMB 7.64 billion generated a year earlier, and its gearing ratio went from 41.8% at year end to 73.2%. Growth that lands as inventory on water is the expensive kind on that cash path.
Higher price, thinner after duty, slower to cash. It is a better car to sell and a harder one to fund.
Tesla's August proves the subtraction works
The association publishes each brand's Chinese retail. Companies publish their global deliveries. The difference is, approximately, what left the country — and Tesla settles whether "approximately" is good enough.
Tesla shipped 86,166 China-built cars in August on the association's wholesale count and retailed 50,047 of them in China. The difference is 36,119 — exactly the Shanghai export figure reported independently for the month. The subtraction is not a guess.
It also carries its own warning. Tesla exported a record 66,330 cars from Shanghai in July, 70.9% of that month's wholesale, and 36,119 in August, 41.9%. The line nearly halved in four weeks while nothing about Chinese demand for Teslas halved. Monthly export counts are shipping schedules as much as they are commerce.
That is the one Tesla point here. Our Tesla model bear case opens on "vehicle demand stalls against a cheaper Chinese field," and a month in which China's domestic market contracted 23.6% while the new-energy field more than doubled its shipments abroad is one month of evidence aimed at that sentence. One month is a datapoint. This piece is not where that file gets opened.
The market moved. XPeng's own export line went the other way.
XPeng delivered 39,107 vehicles in August and retailed 30,982 of them in China. The gap is about 8,100 cars — roughly 21% of its deliveries, and 0.9% of China's 888,000 passenger-car exports.
Run the same subtraction against a month where the company published the answer, and the estimate holds up. XPeng disclosed 9,700 overseas deliveries in July, a monthly record, up 223% year over year, and the association's own July export table put XPeng at the identical 9,700. Through July the company had delivered 41,299 vehicles overseas, up 90.3%, against 45,008 in all of 2025.
So the reading is this. In the month China's exports rose 78%, XPeng's own overseas line fell about 16% from its record — from 9,700 in July to roughly 8,100 in August. Treat the August figure as indicative rather than exact: a company delivery count and an association retail count are struck on different bases at different moments. But the direction does not depend on the precision, and the tide did not lift it.
NIO is the cleaner case. It delivered 35,836 and retailed 35,655 in China. The gap is 181 cars. No reasonable adjustment turns 181 into a number that matters to a quarter. This export surge is happening beside NIO, not to it.
What XPeng's bull case actually needs, and a mix shift is not it
This matters because our XPeng model writes its upside on this exact line. The bull case rests on two things compounding: overseas deliveries passing 20,000 in a quarter for the first time, up 81% year over year, and services and others growing 93.9% at a 75.1% gross margin while vehicle margin fell. Only the first is in play today.
Management put a number on it on the June-quarter call, and the wording is precise: "Overseas delivery of the MONA L03 are expected to begin in the fourth quarter, driving firm-wide quarterly overseas deliveries to exceed 40,000 units."
Above 40,000 a quarter is about 13,300 a month. July's record was 9,700 — 27% short. August's derived 8,100 is 39% short. Getting there means beating the best month XPeng has ever had by 37% and then holding it three times in a row.
Now put that against the second target from the same call: monthly deliveries above 60,000 in the fourth quarter. 13,300 of 60,000 is 22% overseas, against about 21% in August. The mix barely moves. What has to move is the absolute number — total deliveries up 53% from August, with two new SUVs behind it, the G9L launching in September and the MONA L05 in the fourth quarter.
That is the finding, and it is the opposite of the intuitive read. A national shift in mix toward exports is not what gets XPeng to its target, because XPeng is not being asked to export a larger share of what it builds. It is being asked to build and sell far more. A rising export tide gives it a wider channel; it does not supply the cars.
So no model input moves on this month, and the size of that "no" is the point. The bull case adjusts a company-level growth rate, a margin and an exit multiple, and an association month contains none of the three. What the month does supply is an instrument: XPeng discloses overseas revenue only as a six-month geographic total, so deliveries less Chinese retail, monthly, is the only interim read anyone has on the driver its upside depends on. It now reads about 8,100.
One basis note, because we have published both sides of it. Our XPeng June-quarter piece carries He Xiaopeng's "more than 25% of first-half revenue" from international sales. That is a share of revenue, on cars worth roughly twice the domestic average. The 21% above is a share of units. They are different quantities and setting them side by side would overstate how exported XPeng is.
Two more reconciliations. Our Xiaomi June-quarter analysis put the Chinese market down 22% by retail volume in that quarter; August extends the same retail series to −23.6%, and Xiaomi — which retailed 30,153 cars in China in August and exports none — takes that contraction with no offsetting line. And our NIO June-quarter analysis left one question the print and seven analyst questions never settled: why the June quarter missed its own volume guide. A shrinking domestic market is the obvious candidate, and August argues against it — NIO's deliveries grew 14.5% year over year in a market whose retail fell 23.6%. Whatever cost NIO those 2,342 cars, it was not the market.
What to watch
- The association's per-automaker export table, around the middle of September. It publishes the August split by manufacturer and will give XPeng's actual overseas count, replacing the subtraction used here. July's put XPeng at 1.8% of China's new-energy exports.
- XPeng's September deliveries, on 1 October, against its Chinese retail two weeks later. The September-quarter guide of 115,000 to 121,000 needs 37,866 to 43,866 in September, against 39,107 in August. Two more months of the subtraction says whether 8,100 is a level or a slope.
- NIO's September figure, in the first days of October. The September-quarter guide of 108,000 to 111,000 needs 36,230 at the floor and 39,230 at the ceiling, against 71,770 delivered in July and August and a June-quarter monthly average of 35,886. Every month of that guide is better than either month already printed.
- Whether the EU minimum-price mechanism clears, and which makers get undertakings accepted. That is the difference between a 30.7% landed duty and a price floor, on the cars carrying twice the domestic revenue.
- September's wholesale-to-retail gap. August destocked about 60,000 units. If that reverses while exports hold near 888,000, the cars are going onto ships and into overseas inventory rather than to buyers, and the export figure stops being a demand signal at all.
Every August 2026 and August 2025 Chinese market figure here — retail, wholesale, exports, new-energy penetration and the brand-level Chinese retail counts — is the China Passenger Car Association's, published on 8 September of each year. The August 2025 wholesale base, both export shares, the 60,000-unit destock and XPeng's and NIO's implied overseas volumes are R40 arithmetic on association figures; implied overseas volume is company deliveries less association Chinese retail, an approximation across two counting bases that reproduces Tesla's separately reported Shanghai export figure exactly and matches XPeng's own disclosed July count. Reuters reports the month as 894,000 exports on 1.55 million retail; we use the association's 888,000 and 1.541 million throughout. The association's stated 23.6% retail decline implies an August 2025 base near 2.02 million, above the 1.995 million it published a year ago. XPeng's and NIO's delivery counts, guides, targets, cash and export pricing are as each disclosed on 24 August and 1 September 2026. Scenario, growth, margin and exit-multiple assumptions are ours, struck against an $11.15 XPeng price as of 27 August 2026.
