XPeng

Phase A — Understand the business

Company Overview

XPeng sells Smart EVs and NEVs (now incl. EREV range-extender variants) to China's middle/upper-middle class, plus a fast-growing services line (technical R&D/licensing, supercharging, after-sales, parts, carbon credits). Two reported revenue buckets: vehicle sales and services & others.

• FY2025 total revenue RMB76,719.7M (vehicle RMB68,378.9M / 89.1%; services RMB8,340.8M / 10.9%).
• FY2025 deliveries 429,445 units, +125.9% YoY (190,068 in 2024; 141,601 in 2023).
• Model line (as of the filing): MONA M03 (low-ASP volume leader), Next P7, P7+ / P7+ EREV, G6 / G6 EREV, G7 / G7 EREV, G9, X9 / X9 EREV. New 2026 models incl. GX SUV.
• Distribution: 721 stores across 255 cities (mix of direct + franchised; franchise = asset-light).
• Customers = retail consumers (no buyer concentration), plus one structurally important B2B customer: the Volkswagen Group, which pays XPeng for E/E-architecture + platform technical services (see Lens 3/4). Employees: 19,884; R&D staff = 44.5% of headcount.
• Contract structure: vehicle revenue recognized at delivery; bundled lifetime warranties/charging/OTA are separate performance obligations deferred over time; the VW work is milestone/royalty (sales-based) technical-services revenue.

Read: this is a sub-scale-but-scaling OEM (≈430k units/yr vs. BYD's ~4M, Tesla's ~1.8M) that has pivoted from "ADAS-differentiated car company" to "physical-AI platform that happens to ship cars." The car P&L finally works at scale; the question is whether the AI optionality is worth the burn it demands.

Supply Chain

Upstream → XPeng → end customer, named where the filing names them:

• Battery cells (single-source-ish): XPeng does not make cells; it has "fully qualified only a very limited number of suppliers" and "very limited flexibility in changing battery cell suppliers". China's incumbent cell suppliers are CATL and BYD/FinDreams (industry knowledge; not named in-filing) — XPeng also holds minority equity stakes in two NEV-battery makers (RMB190M Dec-2021 + RMB50M Apr-2022 + HK$157M Oct-2022) to secure supply. Chokepoint.
• Semiconductors / compute (single-source + geopolitical): does not make chips; imports advanced silicon; explicitly exposed to BIS export controls (Oct-2022 → Dec-2024 tightening) and the 2025 memory-chip shortage. Its in-house Turing AI chip ("processing hardware") tape-out Aug 2024 is the vertical-integration answer — but it relies on a foundry to fabricate (TSMC-class; not named in-filing) and "may not ramp in a cost-efficient manner, or at all". Chokepoint.
• mmWave radar / sensors (single-source): has experienced mmWave radar shortages that hit deliveries.
• Raw materials: steel, aluminium, lithium, nickel, cobalt, manganese — commodity price exposure flagged.
• Manufacturing (owned): Zhaoqing + Guangzhou plants; Wuhan base construction completed (property cert obtained by Mar-31-2026). Plus third-party contract manufacturing in Europe + Southeast Asia (localized production initiated 2025).
• Downstream: direct + franchised stores; self-operated + partner supercharging network; OTA/cloud for software.

Read: the two named single-source chokepoints (cells, advanced compute) are exactly where US–China decoupling bites. The battery-maker equity stakes and the Turing chip are deliberate de-risking, but neither is proven at the volume XPeng needs.

Competitive Advantages (moats)

1. In-house full-stack AI/ADAS — XNGP rolled nationwide (no-HD-map) since Nov-2023; end-to-end model May-2024; Hawkeye vision (P7+, Nov-2024); VLA 2.0 rollout began Mar-2026; XNGP urban monthly-active penetration 85% (Jun-2025). Real software lead vs. most legacy OEMs — the durable edge if autonomy compounds.
2. The Volkswagen licensing annuity (the under-appreciated moat). VW pays XPeng to put XPeng's E/E architecture into VW's China cars — and the Aug-2025 expansion extended it from EV platforms to VW's ICE + PHEV platforms (>70% of China auto sales). XPeng booked RMB1.72B of technology-licensing profit in H1-2025. This is high-margin, recurring, and a third-party validation of XPeng's tech that no pure-play China-EV peer has. The single best moat in the story.
3. Cost/vertical integration — own chip, own OS, own powertrain → the FY24→FY25 vehicle-margin climb (8.3%→12.8%).
4. Brand/scale — weaker. XPeng is a value-tech brand (MONA = entry), not a premium one (vs. Li Auto's family-SUV franchise or NIO's premium/BaaS). 721 stores is mid-pack.

Bargaining power: WEAK over cell/chip suppliers (single-source, they need XPeng less than XPeng needs them); WEAK over consumers in a price war (it is cutting prices, not raising them — see GX); STRONG and unusual over VW (VW chose to buy XPeng's IP — a rare instance of a Chinese upstart with leverage over a global incumbent).

Segments

By product (FY2025, RMB000):

• Services margin is the story within the story: 68.2% gross margin (FY25) vs. 57.2% (FY24) — driven by VW technical-services/licensing, parts, and carbon credits. Services revenue grew +65.6% YoY and is structurally higher-margin than cars; Q1'26 auto-services revenue +43.6% YoY to RMB1.44B, "mainly due to technical services for Volkswagen". As services mixes up, blended margin compounds faster than the car business alone implies.
• By geography: the 20-F does not break out a clean geographic segment table; revenue is overwhelmingly China, with a nascent Europe/SE-Asia export + localized-production push (began 2025). Geography detail: n/a — not segment-reported.

Trend: vehicle revenue +90.8% (FY25, volume-led + better mix); services +65.6% (margin-accretive). The acceleration is real but FY25-specific — Q1'26 reversed it (Lens 5).

---

Phase B — Measure performance

Earnings Result

FY2025 (audited, the 20-F):

• Revenue RMB76,719.7M (+87.7% YoY) ≈ US$11.1B.
• Gross profit RMB14,472.9M; GM 18.9% (FY24 14.3%; FY23 1.5%) — a violent margin recovery.
• R&D RMB9,490.0M (+47.0%); SG&A RMB9,398.5M (+36.8%).
• Loss from operations RMB(2,771.4)M (FY24 −6,658.1; FY23 −10,889.4) — operating loss cut to −3.6% of revenue.
• Net loss RMB(1,139.5)M (FY24 −5,790.3; FY23 −10,375.8) — near breakeven, helped by RMB1,761.4M other income (government subsidies, +198.9%) and RMB1,163.2M interest income.
• Operating cash flow +RMB8,258.5M (FY24 −2,012.3) — a huge swing, largely working-capital (accounts-payable +RMB14,082.9M as volume ramped). CapexCapital expenditureMoney spent on long-lived things — buildings, machines, servers — rather than on running costs. RMB3,347.1M → FCF positive on a reported basis.
• Balance sheet: cash+ST-investments+time-deposits RMB47,656.5M (incl. RMB6,071.5M restricted); total borrowings RMB12,708.8M (ST RMB4,282.0M + LT RMB8,426.8M) → net cash. Total equity RMB30,368.6M; total assets RMB103,162.6M.

Q1 2026 (the latest print — and it's a warning):

• Total revenue −17.6% YoY and −41.4% QoQ; vehicle revenue RMB11.00B (US$1.59B), −23.5% YoY / −42.3% QoQ.
• GM 20.6% (Q1'25 15.6%; Q4'25 21.3%); vehicle margin 12.1% (Q1'25 10.5%; Q4'25 13.0%). Cash RMB42.09B (US$6.10B).
• Guidance: Q2'26 deliveries 100,000–106,000.

The tension in one line: margins held up impressively (GM ~20%, well above NIO/BYD/Tesla on this metric) even as volume/revenue fell off a cliff YoY. That is the whole debate — is the FY25 ramp the trend, or was it a subsidy-juiced peak now mean-reverting?

Earnings Calls (sentiment trend)

No transcripts on disk (transcripts=0). From `` proxies:

• Management's drumbeat through 2025→2026: margin discipline, services/VW monetization, AI leadership (VLA, Turing), and the physical-AI pivot (IRON/Robotaxi). He Xiaopeng personally took the CEO seat of the robotics unit (Jun-2026) — signaling robots are now a board-level priority, partly to fill a robotics-chief departure.
• Tone shift: from FY25 "we've cracked the margin/volume flywheel" to 1H26 "navigating a difficult NEV year (subsidy pullback + price war), bridging to a 2H26 new-model rebound." More defensive on near-term volume, more aggressive on the long-dated AI moonshots.
• What they're saying more of: VLA / world-models / robotaxi / humanoid. What they've gone quieter on: near-term unit guidance (the ~600k FY26 internal target now looks a stretch given −22.6% YTD).

Comps

Pure-play China-EV + global EV peers. Multiples are `` (Jun-2026) or n/a. EV-makers here mostly have negative/near-zero earnings → P/E and dividend are largely n/a; the live comps are scale, vehicle margin, and EV/Sales.

Stock-Price Catalysts (what moves XPEV >5%)

From `` + filing history:

• Monthly delivery prints — the dominant catalyst. XPEV trades tick-for-tick on CnEVPost monthly numbers. (Jan–May 2026: 125,851, −22.6% YoY; May 32,158, −4.1% YoY but +3.7% MoM).
• New-model launches / order books — GX SUV got 24,863 firm orders in its first 12 hours; ~50k total May orders (+40% MoM) → a positive catalyst. MONA M03 launch (Aug-2024) was the prior volume inflection.
• VW deal milestones — each expansion (Jul-2024 master agreement; Aug-2025 ICE/PHEV expansion) is a re-rate catalyst.
• AI/robot reveals — IRON unveils, robotaxi announcements, He taking the robot-CEO role.
• Macro/policy — China NEV subsidy/purchase-tax changes (purchase-tax exemption steps down to half-rate, cap RMB15k, for 2026–27) and the EV price war / "anti-involution" rhetoric.
• The big down-move: −47% from the Nov-2025 high of $28.24, on persistent YoY delivery declines + subsidy pullback + "overextension" worry; Macquarie cut PT to $26.

Pattern: this is a delivery-momentum + narrative-optionality stock. It rallies on volume re-acceleration and AI catalysts; it breaks on YoY volume declines and price-war fear. Right now the tape is pricing the bear (volume) and discounting the bull (AI/VW).

---

Phase C — Judge people & books

Management

• He Xiaopeng (Xiaopeng He), 48 — Co-founder / Chairman / CEO (and now CEO of the robotics unit, Jun-2026). Track record: co-founded UCWeb, sold it to Alibaba (2014), ran Alibaba Mobile / Tudou / Alibaba Games as president — a genuine founder-operator with a prior big exit. Founder archetype, technology-first, willing to make long-dated bets (chip, robot, eVTOL).
• Skin in the game: dual-class (Class A/B) → founder voting control. The 2025 CEO Award = 28,506,786 RSUs, vesting entirely on market conditions — three equal tranches unlocking only when the 30-day avg HK share price reaches HK$250 / HK$500 / HK$750. With shares far below those triggers, this is aggressively performance-aligned (he earns it only on a multi-bagger) — but also a tell of management's own price ambition and a future dilution source. Total board+exec cash comp RMB182M (FY25).
• Bench (unusually strong): Fengying Wang (President) — 30+ yrs auto, ex-Great Wall Motor vice-chairman/GM (a heavyweight industrial-operations hire, the adult-in-the-room for manufacturing/cost). Brian Gu (Hongdi Gu), Honorary Vice-Chairman & Co-President — ex-J.P. Morgan APAC IB chairman (capital-markets/deal brain; architected the VW and DiDi deals). Board includes ex-GGV/Granite Asia (Ji-Xun Foo), ex-Microsoft/Kingsoft (HongJiang Zhang), ex-Bosch China (Yudong Chen).
• Capital-allocation history: mixed-to-improving. Good: the DiDi smart-auto acquisition (2023) delivered the MONA M03 (the volume model) + robotaxi IP; the VW partnership (US$705.6M VW equity in + a recurring licensing annuity out) is a genuinely value-creative two-way deal. Riskier: minority stakes in battery makers + a US$150M Rockets Capital VC fund + eVTOL (Huitian) bets = capital sprayed across many frontier options. ROE/ROIC still negative (net-loss company) but the trajectory (−10.4B → −1.1B op... net loss) is sharply improving.
• Red flags (governance): standard China-ADR overhangs (VIE structure — though VIEs are immaterial to revenue; HFCAA/PCAOB audit-access risk; dual-class). The robotics-chief exit just as the robot unit ramps is a minor flag (mitigated by He stepping in). No related-party self-dealing of note beyond the disclosed DiDi/VW deals.

Read: this is one of the stronger management teams in China EV — a credible serial-founder CEO, a real industrial COO-type in Wang, and a deal-maker in Gu. The market-condition CEO grant is a positive incentive signal. The risk isn't competence; it's focus — they are running a car turnaround, a chip program, a humanoid robot, an eVTOL, a robotaxi, and an international expansion at once.

Forensic Red Flags

Forensic lens, FY2025 20-F:

• Earnings quality: net loss −RMB1,139.5M but operating cash flow +RMB8,258.5M — the cash beat earnings, which is favorable, BUT it was dominated by a RMB14,082.9M jump in accounts/notes payable (stretching suppliers as volume ramped). That payable build reverses if volume falls — and Q1'26 volume did fall, so expect FY26 operating cash flow to give a chunk back. Watch the payables/working-capital unwind.
• Subsidy dependence: other income RMB1,761.4M (+198.9%), "primarily government subsidies" — i.e., a meaningful slice of the move toward breakeven is government money, not core operations. With purchase-tax exemption halving in 2026–27, this tailwind fades.
• Inventory & receivables: inventory +RMB5,772.1M and installment-payment receivables +RMB3,105.1M in FY25 (volume-driven); inventory write-downs RMB555.4M (FY24 RMB943.7M) — recurring write-downs tied to model cessations (G3i, P5, P7/P7i). Normal for a fast-cycling OEM but a margin drag to monitor.
• SBC: RMB564.3M (modest, ~0.7% of revenue) — but the 28.5M-RSU CEO award + the new 2025 Scheme (up to 10% of shares) is future dilution not yet in SBC.
• Intangibles/goodwill: DiDi-acquisition intangibles (VPT 10-yr life; VMT 5-yr); no goodwill impairment taken FY23–25; a RMB117.3M fair-value loss on the DiDi contingent-consideration derivative in FY25.
• Structure: Cayman holdco + PRC opcos + immaterial VIEs; HNTE 15% preferential tax rates across subsidiaries (renewal risk). Standard but real China-ADR complexity.

Regulatory findings (required):

• SEC: regulatory-findings.md — 0 Litigation Releases, 0 AAERs naming XPeng (EDGAR EFTS, 2021–2026).
• Company's own disclosure (Item 8 / Item 4, Legal Proceedings): "We are currently not a party to any material legal or administrative proceedings.".
• Non-SEC web check: No material FTC/DOJ/CFPB/consent-decree hits found. The only historical items: an immaterial fine for unconsented facial-recognition use at Shanghai stores (terminated, data deleted); two product recalls — G3 (13,399 units, 2021, inverter) and P7+ (47,490 units, Sep-2025, steering-sensor wiring), both supplier-funded with minimal cost to XPeng; two historical trade-secret cases involving former employees (Apple criminal case — ex-employee pleaded guilty 2022, XPeng only subpoenaed for docs; Tesla civil case — dismissed with prejudice 2021, confidential settlement, XPeng not a defendant). No 2025 cybersecurity incident or data breach.
• Verdict: Clean — no material regulatory or accounting-fraud findings via SEC EDGAR EFTS (LR/AAER), web search, and 20-F Item 3/Item 8 as of 2026-06-21. The real forensic risks are quality-of-earnings (subsidy + payables-driven cash), not fraud.

---

Phase D — Project & stress-test

Forward Projection

XPeng is near breakeven, not yet profitable — so the right projection is path-to-GAAP-net-profit + delivery trajectory, not a clean three-year EPS ladder. Built bottom-up from FY25 actuals + Q1'26 + monthly run-rate. Fiscal year = calendar year; "next three" = FY2026 / FY2027 / FY2028. All outputs ``, inputs labeled.

Volume:

• FY25 actual 429,445. YTD Jan–May'26 125,851, −22.6% YoY. Q2'26 guide 100–106k. Internal FY26 ambition ~600k (now a stretch).
• Bear FY26 ~430–460k (flat-to-slightly-up; new models only offset MONA/subsidy weakness). Base FY26 ~500–540k (GX + 4 new models drive a 2H re-acceleration; ~+15–25% YoY). Bull FY26 ~600k (the internal target; strong GX + int'l).

Revenue & margin:

• Q1'26 GM 20.6%, vehicle margin 12.1% — margins are holding ABOVE FY25's 18.9% blended even on lower volume, helped by VW services mix. Services (≈68% GM) keep mixing up.
• Base FY26 revenue ~RMB80–88B (flat-to-up on price-war ASP pressure offset by volume + services); blended GM ~19–21%.

Path to profit (the number that matters):

• FY25 net loss −RMB1.14B on +88% revenue. The incremental drop-through (services + scale) suggests GAAP net breakeven is plausible in FY26–FY27 if volume re-accelerates and subsidies don't fall faster than mix improves.
• Base: FY2026 ≈ GAAP-breakeven to small loss; first full-year GAAP net profit in FY2027..
• Bull: FY26 small net profit (if 600k + VW licensing steps up). Bear: FY26 net loss widens back toward −RMB3–4B if volume stays −20% and the payables/working-capital tailwind reverses.

No our model create (watchlist/breadth mode — Lens 11 logs a forecast only on a genuine committed base case; deferred to a our position log pass). The scoreable line I would log: "XPEV reaches first full-year GAAP net profit by FY2027-end, p≈0.55."

Bull vs Bear

Bull case. XPeng is the cheapest way to own three converging S-curves at a 52-week-low multiple: (1) a margin-inflecting car business (GM 1.5%→18.9%→~20.6%, vehicle margin now leading NIO/BYD/Tesla on the headline), net-cash, FCF-positive on a reported basis; (2) a high-margin, recurring VW software-licensing annuity that just tripled its TAM (extended to VW's ICE+PHEV China platforms, >70% of sales) — a third-party-validated moat no China-EV peer has; (3) free embodied-AI optionality — IRON humanoid (mass-pro end-2026, 1M-by-2030 ambition), AeroHT eVTOL (7,000+ orders, <$300k), robotaxi (3 models, DiDi IP) — that the market is currently valuing at roughly zero. The contrarian view: the market is extrapolating a subsidy-driven 1H26 volume air-pocket into a permanent demand failure, while ignoring that XPeng's gross margin is rising and its software annuity is structurally separate from the car price war. If GX + the 2H26 model wave re-accelerate deliveries, the stock re-rates on both volume and AI narrative.

Bear case. Three things that could permanently impair: (1) China NEV price war + subsidy withdrawal structurally caps ASPs and demand — XPeng is cutting GX prices to chase orders, the opposite of pricing power; MONA M03 (the volume model, ~44% of units) is the most subsidy- and price-sensitive part of the mix. (2) Strategic overextension — running a car turnaround, an in-house chip, a humanoid, an eVTOL, a robotaxi, AND international expansion simultaneously, while still loss-making, is a classic capital-incineration setup; the robotics-chief exit hints at execution strain. (3) The AI optionality may never monetize — humanoids and eVTOLs are pre-revenue moonshots with no proven unit economics; if they consume cash for years and the car business stays sub-scale, XPeng is just a perennially-breakeven OEM in the world's most brutal auto market. Pre-mortem (18 months out, thesis broke): FY26 deliveries finished flat-to-down (~430k, the ~600k target missed badly), the subsidy step-down + price war kept net income negative, IRON/eVTOL slipped and burned cash, and a capital raise diluted holders into a still-falling stock. On multiples: at ~US$12.5B for a 20%-GM, net-cash, near-breakeven OEM, the car business isn't expensive — but the bull case requires paying up for AI options that may be worth zero.

Devil's Advocate (short-seller)

• Revenue concentration that bulls underweight: not customer concentration in cars, but (a) MONA M03 carrying ~44% of volume at the lowest ASP/margin and the highest subsidy sensitivity, and (b) the services-margin halo leaning on ONE customer — Volkswagen. "We primarily rely on the Volkswagen Group for the revenue arising from such [technical] services". If VW's China strategy shifts, in-sources, or the relationship cools, the single best part of the margin story evaporates. The 68% services margin is a VW-dependency dressed up as a moat.
• The margin recovery is partly borrowed: FY25's near-breakeven leaned on +199% government subsidies and a RMB14B payables build. Both fade/reverse. Strip them and the "we cracked profitability" narrative is thinner than the headline GM suggests.
• Most dangerous competitor bulls underestimate: Xiaomi. Xiaomi's EV unit is posting standout profitability and brand pull that XPeng's value-tech positioning can't match — plus BYD's cost wall below and Li Auto's family-SUV franchise beside. XPeng is squeezed from premium, volume, and cost simultaneously.
• What must hold for today's price: that the 1H26 −22.6% volume drop is a transient air-pocket (not structural demand loss), that GX/new models re-accelerate by 2H26, that subsidies don't fall faster than mix improves, and that the AI bets don't become a multi-year cash sink. If volume disappoints 20–30% vs. the ~600k target (i.e., finishes ~420–480k), there is no GAAP profit in FY26, subsidies are gone, and the stock's only support becomes the embedded options — which short-sellers value at ~zero. The single permanent-impairment scenario: a forced, deeply-dilutive equity raise into a sustained price war while the AI moonshots are still pre-revenue. Plausible but not base-case given the RMB42B cash / net-cash balance sheet — the cash cushion is XPeng's best defense against the bear.

Management Questions (ordered by information value)

1. FY26 deliveries are −22.6% YTD vs. a ~600k internal target — what is the honest exit-rate volume you're underwriting for 2H26, and what specifically (GX, new models, int'l) bridges the gap?
2. What share of the 68%-margin services line is Volkswagen, and what is the contracted floor/duration of that licensing revenue if VW slows its China program?
3. Of FY25's path to breakeven, how much was government subsidy + working-capital (payables) that won't recur — i.e., what is "clean" structural operating profit at this volume?
4. As purchase-tax exemption halves in 2026–27, what's the demand-elasticity assumption for MONA M03, your most subsidy-sensitive volume model?
5. What is the committed multi-year cash budget for IRON + AeroHT + robotaxi combined, and at what milestone would you cut or spin out any of them?
6. IRON: what is the realistic FY27–28 revenue and gross margin, not just unit ambition — who pays, and at what price, for the first 100k units?
7. Why is He Xiaopeng personally running the robotics unit, and what does the prior robotics-chief's departure tell us about execution risk there?
8. Vehicle margin 12.1% in a price war — what's the floor, and how much is structural (Turing chip, integration) vs. cyclical (subsidy, mix)?
9. Turing chip: is it in volume production yet, what % of the fleet runs it, and what's the cost-per-vehicle saving vs. merchant silicon?
10. What is your financing plan through FY27 — do you expect to need an equity raise, and under what conditions?
11. International (Europe/SE-Asia localized production): what unit volume and margin in FY27, and how do you clear tariff/geopolitical risk?
12. eVTOL (AeroHT/Huitian): 7,000+ orders at <$300k — what's the certification path and the first hard delivery-and-revenue year?
13. How do you defend the value-tech brand against Xiaomi's pull above and BYD's cost wall below?
14. Capital allocation: with net cash and recurring losses, why a US$150M VC fund (Rockets Capital) and minority battery stakes rather than concentrating on the core?
15. What is the single metric you want shareholders to judge you on over the next 24 months?

---