Three Chinese EV makers posted July delivery numbers on Saturday, and each release is arithmetically true. Nio delivered 35,934 vehicles [1], XPeng 38,027 [3], Zeekr 35,837 [2]. What the headlines built on those figures actually claim is where the reading skill comes in, because record deliveries and slowing growth are frequently the same number wearing different clothes.
Nio's year-over-year line is the spectacular one. July's 35,934 vehicles are up 70.98 percent from a year earlier [1]. The same figure is down 11.49 percent from June's 40,597, which made July, in CnEVPost's words, "the first sequential decline in three months" [1]. Both statements are correct. Year over year measures how much bigger the company is than its 2025 self; month over month measures whether the growth is still happening now. A 4,663-vehicle sequential drop is the momentum story, and a year-over-year headline will never contain it.
The release detail worth checking every month is whether the parts sum. Nio's split does: 20,008 vehicles for the Nio brand, 10,155 for Onvo, 5,771 for Firefly, which adds to exactly 35,934 [1]. That is not a trivial check; a delivery figure whose brand-level parts do not reconcile is the first sign a release is mixing wholesale, retail, or registration counts, and this one passes. For scale, Nio has now delivered 1,224,649 vehicles all time, 227,057 of them in the first seven months of 2026, up 67.98 percent on the same period last year [1].
XPeng's release says the company "delivered 38,027 vehicles in July, marking an increase of 4% year over year" [3]. The division is worth running. XPeng's July 2025 base was 36,717 vehicles, so July's gain of 1,310 vehicles works out to 3.57 percent, which is 3.6 percent at one decimal. Rounded to a whole number, 3.57 does become 4, so the release is not wrong; it chose the coarsest precision available, and the coarse version reads meaningfully larger when growth is this slow. A true 4 percent would have been roughly 38,186 vehicles, 159 more than XPeng delivered. We print 3.6 percent.
Zeekr's headline is a genuine streak. July's 35,837 vehicles are an all-time high and the fourth consecutive monthly record, up 111.09 percent year over year [2]. The momentum sits in the footnote: per CnEVPost, "the month-on-month gain narrowed to 1.90 percent, the smallest since March" [2]. June was 35,169, so the record was set by 668 vehicles. A company can be at its all-time peak and flattening at the same time; that is what a plateau looks like from the front.
Li Auto, whose July figure of 30,468 crossed our wire earlier this week, fell on both comparisons, against last July and against June. No record framing was available, so no record headline appeared. The contrast is the lesson: the companies that grew chose the comparison that grew fastest, and the one that shrank had no favorable comparison to choose.
The general method fits in a paragraph. A delivery release will usually volunteer the flattering comparison, so the reading starts with finding both, the year-over-year and the month-over-month, and pulling last month's absolute figure for the subtraction the release skipped. Brand and model breakdowns should sum to the total, as Nio's does. Every percentage should be recomputed against its published base, and where the decimal disagrees with the release's whole number, the decimal wins. A "record" is a claim about one month measured against history, not a claim about direction. July's releases pass those arithmetic checks and still, read together, describe a market decelerating into its record books.
