$XPeng(XPEV.US) released its Q2 2026 results after HK close and before US market open on Aug 24 Beijing time. Overall, XPeng delivered a middling print, while the all-important Q3 guide badly missed Street expectations. Details below:

1) Total revenue slightly missed: Q2 revenue came in at RMB 19.7 bn, just below the Street’s RMB 19.9 bn, mainly dragged by auto sales. But 'Services and other' was strong at RMB 2.7 bn, up RMB 660 mn QoQ from RMB 2.03 bn.This was driven by milestone-based recognition in tech R&D services and higher parts and accessories sales.

2) Auto revenue and ASP under pressure as mix shifted down: Auto sales were RMB 17.05 bn (up only ~1% YoY, below the Street’s RMB 17.5 bn). ASP fell to RMB 165k, down RMB 10k QoQ from RMB 175k, and below the Street’s RMB 169k.The decline reflected mix: lower-priced Mona M03 rose 8ppt QoQ to 41% of sales, while the high-end flagship X9 fell to 7%. Despite export volume rising 81% YoY (share up to 19.4%), the uplift could not fully offset ASP pressure.

3) Auto GPM missed as price cuts offset cost-down: Q2 auto GPM was 12.1%, flat QoQ and below the Street’s 12.4%. On costs, a sharp delivery ramp (up 65% QoQ to 103k units) brought scale benefits, while switching the lineup to in-house Turing chips lowered BOM.Unit cost fell ~RMB 9k QoQ to RMB 145k, offsetting higher raw material prices in batteries and memory chips. However, unit pricing fell too much, pushing unit GP down another ~RMB 1.3k QoQ to ~RMB 20k.

4) Overall GPM beat on high-margin services:

Q2 overall GPM reached 20.7%, well above the Street’s 19.3%. With auto GPM flat, the beat was driven by 'Services and other'.That segment delivered a 75.1% GPM, up 860 bps QoQ (vs. Street’s ~69%), primarily on a higher mix of high-margin tech R&D services.

5) R&D stayed elevated for AI; S&M rose on volume recovery:

To prep for a new product cycle and advance AI, R&D was RMB 2.9 bn (flat QoQ, slightly below the Street’s RMB 3.05 bn). Spend focused on new models (MONA L03, G9L, GX), in-house Turing chips, the VLA 2.0 LLM upgrade, Robotaxi testing, and the 'Iron' humanoid robot’s mass-production push.Meanwhile, S&M reached RMB 2.5 bn (above the Street’s RMB 2.28 bn), up ~RMB 610 mn QoQ. This was driven by higher commissions to franchised stores on stronger sales, new model launches lifting marketing and ad spend, and continued channel expansion.

6) Core operating loss narrowed:

Q2 net loss was RMB -1.24 bn, worse than the Street’s RMB -770 mn, due to larger FV losses on investments and lower other income. Core operating profit (GP minus core opex), a better proxy for underlying cash generation, was RMB -1.33 bn.That narrowed by nearly RMB 800 mn vs. last quarter’s RMB -2.1 bn, showing positive operating leverage from a volume recovery.

Dolphin Research view:

In short, XPeng’s quarterly results were mediocre, and the crucial next-quarter guide clearly missed Street expectations. The market had expected auto GPM to improve QoQ as the higher-priced GX lifted mix, but results showed only flat auto GPM.The overall GPM beat was mainly due to a sharp increase in 'Services and other' revenue, especially high-margin tech R&D services recognized upon milestones.

However, with S&M rising sharply (new model marketing and ads, plus higher dealer commissions tied to volume), auto net profit still missed. On guidance, XPeng severely missed Street expectations for next quarter.

1) Volume guide implies severe capacity constraints: Q3 delivery guide is only 115k–121k vs. the Street’s 147k. With 38k delivered in Jul, implied Aug/Sep monthly run-rate is only 38.5k–41.5k.Even as MONA L03 enters a full delivery quarter with over 50k orders on hand, the guide suggests the L03 faces serious capacity bottlenecks that cannot be fixed near term. The higher-priced GX has stronger-than-expected orders (~7k monthly sales, ~30k in backlog), but capacity ramp also limits its Q3 contribution.

2) Revenue guide implies lower-than-expected ASP: Q3 revenue guide of RMB 21.7–23.4 bn is far below the Street’s RMB 28.2 bn. The implied ASP is only ~RMB 168k, below the Street’s RMB 173k.This indicates the GX mix remains lower than expected due to capacity constraints, while the lower-priced MONA series still accounts for a high share.

As a result, with L03 in a full delivery quarter and GX still ramping, the much lower-than-expected Q3 guide leads Dolphin to cut its full-year volume forecast to 440k–470k (from 460k–500k, +2% to +9% YoY). Q4 launches of MONA L05 and G9L remain catalysts:

Exports as the core growth engine: The company plans to double 2026 overseas volume to ~90k, focusing on Israel, Norway, Denmark, and France, and expand its network to 680 stores across 60+ countries and regions. Q2 overseas monthly sales already exceeded 6k, with a Q4 target of sustaining 10k+ per month.Under a base case, exports reach ~80k (+~78% YoY).

Domestic base supported by a 'major new model cycle', but capacity bottlenecks cap growth: Despite ~30k GX backlog, nearly 50k L03 firm orders (stable-state monthly sales estimated at 13k–15k), and Q4 launches of MONA L05 and G9L, severe Q3 capacity constraints and longer wait times may cause order attrition.Dolphin expects domestic volume of 360k–390k (YoY -6% to +1%).

XPeng is accelerating AI from R&D to commercialization:

a. Smart driving stack upgrade: On hardware, the in-house 'Turing' chip (750 TOPS per chip) is now mass-produced across the lineup. On algorithms, the VLA 2.0 model, released in Mar with a 'vision-to-action' architecture, was further upgraded in Aug for higher intelligence, stability, and generalization, with sharply lower intervention/takeover rates.It will serve as a unified base for autos, Robotaxi, and robots.

b. Robotaxi ramp-up: XPeng secured road testing permits for intelligent connected vehicles in Guangzhou and is conducting routine L4 public road tests. The mass-produced Robotaxi car is off-line, with pilot passenger ops starting in Guangzhou in Q3 2026 and public experience expected on Tech Day.A dedicated budget Robotaxi is planned for 2027 to further validate the business model, with scaled commercial opportunities expected post-2028.

c. Humanoid robot push: The next-gen Iron robot is in a critical R&D phase, featuring all-solid-state batteries and three Turing chips, integrated with a multimodal large model system and targeting commercial, industrial, and household use.Mass production is targeted by end-2026 with monthly capacity over 1,000 units, with initial pilot deployments in XPeng-owned stores and deliveries to enterprise customers starting in 2027.

Based on 2026 volume of 440k–470k (+2%–9% YoY) and revenue of RMB 82.6–87.6 bn, XPeng’s current market cap of ~RMB 78.3 bn implies a 2026 P/S of 0.9–0.95x. Valuation looks relatively low.While Q3 guidance is soft, downside appears limited.

If P/S falls to 0.8x, implied market cap would be RMB 66.1–70.1 bn (10%–15% below current). That would be a relatively safe entry to capture dual value in autos plus AI, as this valuation largely reflects the core auto biz and excludes AI optionality from Robotaxi and robots.

XPeng disclosed a pre-money valuation of RMB 5.0 bn for Iron humanoids and post-money of RMB 6.3 bn (mass production by end-2026, deliveries in 2027). After dilution (XPeng’s effective stake ~73.8%) and a typical primary-market liquidity discount (20%), XPeng’s attributable robot equity value is ~RMB 3.7 bn (~US$500 mn).This could be re-rated as mass production progresses.

Thus, steady progress in AI (Robotaxi pilots, humanoid mass production) could drive a valuation re-rating. If Q4 capacity constraints ease, the new model cycle boosts volumes QoQ, and robots catalyze sentiment, a P/S re-rate to 1.2–1.3x (AI premium) implies market cap of RMB 99.1–114.0 bn.That is ~27%–46% upside vs. current levels.

Near term, shares may be pressured by the soft Q3 guide. But with L03 already a hit, L05 and G9L launching in Q4, Robotaxi demo ops starting in Guangzhou in Q3, and humanoid mass production by year-end, if AI delivers as planned, the valuation midpoint could move up and the stock retains upside.

Full text below:

I. Auto GPM 'below expectations'

With Q2 deliveries already disclosed, investors focused on auto revenue and margins. The Street expected Q2 auto GPM of 12.4%, roughly flat QoQ, as mix uplift from the GX flagship (Ultra >80% mix) and scale benefits offset price pressure in memory, batteries, and other materials.XPeng’s actual Q2 auto GPM was 12.1%, merely flat QoQ, mainly because ASP undershot expectations.

Unit economics details:

a) ASP: decline larger than expected

Q2 ASP was RMB 165k, down RMB 10k QoQ from RMB 175k and below the Street’s RMB 169k. The QoQ ASP decline was due to:① Mix: X9 share fell 7ppt QoQ to 7%, while lower-priced Mona M03 rose 8ppt QoQ to 41%, dragging ASP, partially offset by the higher-priced GX share rising 7ppt QoQ to 7%.② Exports: Exports reached 20k units (+81% YoY) while domestic volume fell 10% YoY to 83k; the export share rose only 1ppt QoQ to 19.4%, providing limited ASP offset.

b) Unit cost: scale offset raw-material inflation

Unit cost was ~RMB 145k, down ~RMB 9k QoQ from RMB 154k, thanks to factors that offset upstream cost inflation:① Scale: Q2 deliveries reached 103k (+65% QoQ), improving fixed-cost absorption in plant depreciation and amortization.② In-house chips: Full lineup switched to in-house Turing chips, lowering key component costs and offsetting price hikes in memory and other materials.③ Raw materials still rising: Battery (LFP prices doubled), memory, and aluminum remained elevated, but the above factors netted out the pressure, driving QoQ unit cost declines.

c) Unit GP: still declining QoQ

Unit GP was ~RMB 20k, down ~RMB 1.3k QoQ from ~RMB 21.2k and below the Street’s ~RMB 21k. The drop was mainly due to a larger fall in pricing than in costs.As a result, Q2 auto GPM was only 12.1%, flat QoQ and below expectations.

II. Q3 guide below expectations, likely due to supply bottlenecks

a) Volume guide shortfall, likely from MONA L03/GC supply constraints

After a tough Q1 under the dual energy strategy (deliveries -33% YoY), XPeng launched the refreshed Mona M03 and the all-new full-size SUV GX, lifting Q2 deliveries 65% QoQ to 103k and back to the same level as Q2 2025.For Q3, XPeng guided 115k–121k, well below the Street’s 147k. With 38k delivered in Jul, implied Aug/Sep monthly sales are only 38.5k–41.5k, a limited step-up vs. Jul.

Q3 growth hinges on two models: the new MONA L03 SUV launched in Jul and the full-size GX entering a full delivery quarter. The weak guide suggests severe supply-chain issues for both, not demand.Post the Jul 16 L03 launch, XPeng received 20k firm orders in 7 minutes and 47k in 1 hour, versus Mona M03’s 20k in 52 minutes and 30k in 48 hours in 2024, underscoring robust demand.

But L03 is constrained by supply capacity, especially for in-house Turing AI chips and other core components, slowing the ramp. As a result, L03 deliveries were limited in late Jul, pulling total Jul deliveries down to 38k.GX orders also beat expectations (~7k monthly sales, ~30k backlog), but the capacity ramp limits its Q3 contribution.

Thus, the Q3 delivery miss is not about demand but slower-than-expected capacity ramp.

b) Revenue guide implies flat ASP QoQ; GX mix offsets L03 share

The Q3 revenue guide of RMB 21.7–23.4 bn, below the Street’s RMB 28.2 bn, is mainly due to lower volume. The implied ASP is ~RMB 168k, up ~RMB 3k QoQ but still below the Street’s RMB 173k.This reflects a lower-than-expected GX share due to supply constraints and a still-high MONA share.

Given ongoing bottlenecks, limited scale effects, and only modest ASP uplift, XPeng still faces raw-material headwinds in Q3. Auto GPM pressure remains high.

III. Overall GPM slightly beat on services margin contribution

Q2 revenue was RMB 19.7 bn (below the Street’s RMB 19.9 bn), with overall GPM at 20.7% vs. the Street’s 19.3%. The beat came from high-margin services; auto GPM missed.① Revenue slightly missed, mainly on auto weakness: Auto sales were RMB 17.05 bn (up ~1% YoY, below the Street’s RMB 17.5 bn), as mix shift drove ASP lower.

'Services and other' was RMB 2.7 bn, up RMB 660 mn QoQ from RMB 2.03 bn. Drivers were:a. Higher tech R&D services revenue as several milestones triggered substantial recognition.b. Higher parts and accessories sales.XPeng had guided 2026 tech services and IP licensing revenue to be roughly flat vs. 2025 (RMB ~8.34 bn).

② GPM slightly beat, mainly from tech R&D services:

Auto GPM was 12.1%, flat QoQ, as scale and in-house AD chips could not fully offset ASP declines and higher raw-material costs. 'Services and other' GPM reached 75.1%, up 860 bps QoQ and above the Street’s ~69%.This was driven by a higher share of high-margin tech R&D services.

IV. Opex: building for dual-energy and AI ecosystem, but weighing profits

XPeng treats 'intelligence' as its core moat, requiring sustained heavy investment in AI driving and new platforms. Such front-loaded investment for a 2026 product super-cycle kept R&D elevated in Q1 and remained a key headwind to core operating profit in Q2.1) R&D: flat QoQ, still betting on new models and physical AI

R&D was RMB 2.9 bn, below the Street’s RMB 3.05 bn and flat QoQ. Spend was directed to new models and AI, including:

① Smart driving upgrades:

Hardware: From Q2, all XPeng models (incl. Max trims) switched to in-house Turing chips. The MONA M03 Max brought 750 TOPS to sub-RMB 140k price points, while the Ultra SE uses dual chips for 1,500 TOPS.Software: Since Mar, the 2nd-gen VLA model removed explicit language translation, boosting inference efficiency by 32x and cutting prediction error by 33%. A major upgrade arrived in Aug, with continued progress toward a 20B on-device parameter goal by year-end.

② Robotaxi (L4):

XPeng obtained Guangzhou permits for intelligent connected road tests and is running routine L4 public tests. The mass-produced Robotaxi is off-line, with pilot passenger ops in Guangzhou in Q3 2026 and a budget Robotaxi planned for 2027 (reference design with four Turing chips, 3,000 TOPS).Scaled commercial opportunities are expected post-2028. XPeng will provide the tech platform and charge commissions, partnering with multiple operators domestically and abroad.

③ Humanoid robot Iron:

Iron has entered small-batch trial production, with the mass-production line in final integration. The target is mass production by end-2026 for industrial and commercial use.It offers 82 DoF, runs three Turing chips, and debuts all-solid-state batteries. Initial deployments will be trial-commercial in XPeng stores with monthly capacity over 1,000 units, and deliveries to enterprise customers start in 2027.

Use cases: Initially product introductions and sales assistance, later open to ecosystem partners for retail checkout, professional services, and more.

④ New model development:

Q2 saw dense launches: the flagship 6-seat SUV GX in Apr with pre-installed L4-capable hardware, steer-by-wire, and rear-wheel steering; MONA L03 compact SUV reservations opened in Jul with simultaneous overseas shipments in Aug; flagship 5-seat SUV G9L opened presales in Aug with VLA 2.0 and Turing chips, starting at RMB 259.8k.Per guidance, 2026 physical AI R&D will be ~RMB 7.0 bn, and total R&D ~RMB 12.0 bn (+~26% YoY vs. RMB 9.4 bn in 2025), with ~RMB 7.0 bn in AI and ~RMB 5.0 bn in vehicle R&D.

2) S&M and G&A: up sharply QoQ on commissions and marketing

S&M and admin expenses were RMB 2.5 bn, above the Street’s RMB 2.28 bn and up ~RMB 610 mn QoQ. The increase reflected higher commissions to franchised stores on stronger sales, marketing and ad spend for new models (Mona M03 refresh and GX SUV), and faster channel rollout for the new product cycle.XPeng is accelerating store openings: net new stores in Q2 rose by XX (total XX), and cities covered increased by XX (total XX), preparing channels for volume ramp.

V. Operating losses narrowed

Q2 net loss was RMB -1.24 bn, below the Street’s RMB -770 mn. Beyond high S&M as the main drag, 'other income' fell another RMB 50 mn QoQ and FV losses on long-term investments widened by RMB 310 mn, further weighing on net.Core operating profit (GP minus core opex) was RMB -1.33 bn, narrowing by nearly RMB 800 mn vs. Q1’s -RMB 2.1 bn, driven by stronger GPM and a 65% QoQ volume recovery (Q2 103k vs. Q1 63k).

<End here>

Dolphin Research’s deep dives and updates on XPeng include:

Earnings season

Aug 19, 2025 earnings take 'XPeng: Record GPM — Can It Spread Its Wings?'

Aug 20, 2025 call notes 'XPeng (Trans): Aims to Make New P7 a Top-3 EV in the RMB 200k–300k Segment'

May 21, 2025 'XPeng: After the Hard Part, Ready to Soar?'

May 21, 2025 '(1Q25 Trans): Poised to Take Off?'

Risk disclosure and statement:Dolphin Research Disclaimer and General Disclosures